TEM Journal. Volume 6, Issue 3, Pages 479-483, ISSN 2217-8309, DOI: 10.18421/TEM63-07, August 2017.

Design of an Accessible, Powered Myoelectrically Controlled

Osman Onur Akirmak 1, Çağdaş Tatar 1, Ziya Gökalp Altun 1, Yuriy Mishchenko 2

1 Toros University, Bahcelievler 45, Mersin, Turkey 2 Izmir University of Economics, Sakarya Cad 56, Izmir, Turkey

Abstract – In this paper we describe accessible developments have been confined to advanced myoelectric prosthetic hand design based on biomedical and laboratories in main modification of existing mechanical prosthesis and off- industrial countries, whereas the end designs are the-shelf parts and components. Despite significant complex and extremely costly thanks to reliance on advances in myoelectric prosthetics, such existing complex parts and advanced . i- devices remain out of reach of the majority of the patients needing them due to high costs and prosthetic hand – one of such offerings sold complexity. We describe a simple design that can be commercially – comes with a price tag of 20,000 assembled based on existing or readily acquirable parts USD [10]. Despite a significant need existing in at approximately 1/100 of the cheapest commercially myoelectric prosthetics in developing countries, the available alternative. Our design offers above factors make it all but impossible for such disarticulation patients in developing countries an patients to obtain a myoelectric prosthetic. affordable myoelectric prosthesis with significant In this report, we describe a design for myoelectric capacity for improving their quality of life. hand prosthetic that is simple, inexpensive, and does not rely on advanced components or expertise. The Keywords – , sEMG, robotic prosthesis, bionic hand, neural prosthesis. design utilizes all off-the-shelf components and is based on alteration of conventional mechanical hand prosthesis to grant it bionic function. The simplicity, 1. Introduction , and the absence of compulsory use of advanced parts make it possible for anyone to In recent years, myoelectric (EMG) prosthetics implement our design with limited resources, which have attracted significant . Significant work can become a viable alternative for hand and has been done on such systems in research literature amputees in developing countries for acquiring [1-9] as well as myoelectric prostheses are now myoelectric powered prostheses, which have the available commercially on the market including i- potential to significantly improve their quality of life. Limb (Touch , UK), Michelangelo (Otto Bock Healthcare Product GmbH, Austria), Bebionic 2. Materials and Methods (Steeper Group, UK), and DEKA arm (Deka Integrated Solutions Corporation, USA). 2.1. EMG signal sensing and acquisition system Unfortunately, the vast majority of such Our design relies on an off the shelf surface electromyographic sensor based on ECG 3M Red Dot 2560 Ag/AgCl electrodes [11] and a simple DOI: 10.18421/TEM63-07 signal electronic signal amplification and filtering https://dx.doi.org/10.18421/TEM63-07 circuit [12,13], which can be either acquired new –

such as Grove EMG sensor from Seeed [14] – or Corresponding author: Yuriy Mishchenko, assembled from basic electronics components such as Izmir University of Economics, Izmir, Turkey INA106 instrumentation amplifier and active RC Email: [email protected] filters, following the detailed presentations that can © 2017 Osman Onur Akirmak et al; published be found elsewhere in the literature [15-17]. by UIKTEN. This work is licensed under the Creative Electromyography (EMG) is a technique for Commons Attribution-NonCommercial-NoDerivs 3.0 sensing activity in the skeleton muscles of human License. body that develops during activation of those The article is published with Open Access muscles. The principle of EMG sensing from the at www.temjournal.com surface of the relies on placing electrodes next to target body muscles onto the skin surface [1]. In

TEM Journal – Volume 6 / Number 3 / 2017. 479 TEM Journal. Volume 6, Issue 3, Pages 479-483, ISSN 2217-8309, DOI: 10.18421/TEM63-07, August 2017. this work, we use the placement of electrodes first to the raw EMG signal in order to filter out slow identified by the end and the middle points of flexor signal variations otherwise affecting the prosthetic’s carpi radialis and palmaris longus muscles involved performance (Figure 3.). Secondly, the software in the flexion of middle , identified by the implements a simple 1 hidden-layer neural network numbers 3 and 4 in Figure 2. [2]. These electrodes to classify incoming signal into rest and target can be activated by a user upon contraction of his or muscle activation states (Figure 4.). her fist. Respectively, the two sensing electrodes are The digital filter was designed by examining the placed at the end and the mid-points of carpi radialis EMG signal acquired from the sensor using a PC and palmaris longus muscles with the distance computer, by using the Filter Design and Analysis between them of approximately 1.5 cm, and measure tool (FDAtool) in Matlab. During the examination, it the electric potential difference developing between had been observed that the raw EMG signal could these points upon fist contraction. A third, ground contain contaminations in the form of low frequency electrode is placed over the of the same hand, variations at and below approximately 2-3 Hz, which over a bone, to provide a reference for the EMG significantly distorted the signal. Respectively, a high signal to be used with the signal amplification and pass digital filter with cutoff frequency of 2 Hz was filtering circuit. designed using FIR equiripple option in FDAtool in order to clean the EMG signal before passing it to a neural network classifier detecting the target muscle contraction waveforms.

Figure 1. The overall block schema of the design of the myoelectric hand prosthetic in this work.

The signal from the sensing assembly is digitized by passing it onto a clone of Arduino Uno microcontroller via the ADC module of the microcontroller, and is then used by an embedded control software to direct a servo motor enacting the motions of the mechanical mechanism of the prosthetic hand, as will be described below.

Figure 3. Example of the raw (top) and filtered (bottom) EMG signal as well as the waveforms associated with Figure 2. Possible electrode locations for surface EMG target muscles’ activations. measurements as related to hand movements. Locations 3 and 4 have been used in this work. To detect the muscle contraction waveforms, a one-hidden layer artificial neuronal network 2.2. Muscle activation detection and embedded classifier was designed as described by control software

The embedded control software is realized as s = 5 h 40 ( ) software on the base of Arduino Uno microcontroller clone in C programming language. The software 푧 푔 �� 푗 ∙ 푔 �� 푤푖푠 푡푖 �� implements a two stage signal processing strategy 푗=1 푖=1 with a simple digital filter in direct form 1 applied

480 TEM Journal – Volume 6 / Number 3 / 2017. TEM Journal. Volume 6, Issue 3, Pages 479-483, ISSN 2217-8309, DOI: 10.18421/TEM63-07, August 2017. where ( ) = 1/(1 exp( )) is the sigmoid assembly was then covered with the original activation function, ℎ and are the neural network cosmetic glove. The motorized prosthetic was then weights 푔in 푥the output −and the− hidden푥 layer, and ( ) controlled by activations of the flexor carpi radialis 푗 푖 is the waveform input produced푤 from the raw EMG and/or palmaris longus muscles in lower arm such as data by sampling the EMG signal at 40 control points푠 푡푖 triggered by closing the user’s hand. In response, the uniformly distributed at 20 ms intervals over the prosthetic executed alternately hand-open and hand- most recent 800 ms part of the EMG signal. close motions, allowing the user to grab and hold objects just as the original prosthetic did but now controlled entirely by the myoelectric signal.

Figure 4. The diagram showing the layout of the artificial neural network used for detection of muscle activation waveforms in the EMG signal.

The neural network was trained by using the Neural Network design tool in Matlab and a 2 minute fragment of EMG data obtained while user was Figure 5. The drawing of the modified non-motorized continuously contracting the target muscles at a rate Otto Bock hand prosthesis designed in this work. Numbers of approximately one contraction per 2 seconds. An indicate: 1. The rod of the servo motor. 2-3. Fitting activation threshold of 0.999 was set for the neural screws. 4. The rod place bracket. 5. The - network after the training, in order to minimize the prosthetic connection area. 6. The fitting screw mounting rate of false positive errors while retaining most of holes. the true responses. The final accuracy of the neural network with respect to the detection of muscle 3. Results activations was calculated to be 94%. Upon the detection of muscle activations, the In this work we develop a design for powered embedded software activated the servo motor in the myoelectric hand prosthetic based on modification of mechanical subsystem of the prosthetic hand in a an existing conventional mechanical prosthesis. The manner related to the current state of the mechanical modification involved replacing the internal springs hand, which could be “open” and “closed” and was and mechanical pull-cord with servo motors and maintained by the software, so that the motor moved electronic EMG activation circuit. in alternate positions alternating between “open” and Overall, the altered design matched very closely “closed” prosthesis’s states as controlled by the user the original mechanical prosthesis in terms of form, contracting the target muscles. functionality, usability, effectiveness, and non- obtrusiveness. The entire motorized assembly could 2.3. Implementation of the mechanical prosthetic be covered using the original cosmetic glove and mechanism used with the original prosthetic socket, in place of the original prosthesis. In that respect, modified We implemented the mechanical subsystem of our motorized prosthetic was a close to one-to-one copy prosthetic by modifying an existing mechanical hand of the original unit and could be used entirely in its prosthesis Otto Bock (Otto Bock Healthcare Product place, now driven by the EMG control signal. GmbH, Austria), which is a commercially available, The modifications resulted in slight increase in the movable prosthesis that can be activated total weight of the modified prosthesis due to added mechanically by pulling a pull-cord by the user. sensing circuit, Arduino, and servo motor. These Together with internal springs that allowed the modifications added approximately 100 gram to the prosthesis to perform grab-and-hold motions and weight of the original prosthesis. Together with the hold objects as activated by the user manually weight of the original prosthesis’s mechanism of operating the cord with the other hand. approximately 450 gram, that resulted in the total We modified the said prosthesis by removing the new weight of about 550 gram, perfectly compatible pull-cord and internal springs and replacing them with the original unit. In terms of the usability of the with an EMG controlled servo motor mounted at the original unit, these alterations resulted in nearly no centre of the finger joint as shown in Figure 5. The changes.

TEM Journal – Volume 6 / Number 3 / 2017. 481 TEM Journal. Volume 6, Issue 3, Pages 479-483, ISSN 2217-8309, DOI: 10.18421/TEM63-07, August 2017.

provides the same EMG activation capacity as the significantly more expensive existing units and fulfils the most essential grab-and-hold function required by most wrist amputees. With the total costs in the range of 1-2% of the cheapest comparable alternative, our design presents a significant advance over the current state of affairs. Even though commercial prosthetics offer advanced sets of functions such as multiple grab positions which can be changed via smartphone application or manually, considering that taking and carrying objects is the main function required by Figure 6. The modified prosthesis (right) and the EMG- wrist amputees, the simplicity and dramatically controlled activation of the modified prosthesis (left). reduced total cost of our design – in 1-2% of the

cheapest alternative – is an enormous advantage for The Hitec HS-311 servo motor used to animate the wrist amputee users who desire to take advantage of mechanical hand mechanism generated a maximum myoelectric prosthetic over existing torque of 3.5 kg-cm, which was less than commercial units. produced by the original springs that we removed. We tested the effect of this change on the usability of 4. Conclusion the new prosthetic with respect to the original grab- and-hold action that the original unit performed. The In this work we describe a design for a powered above torque produced by the servo motor was found myoelectric hand prosthetic produced by modifying to be sufficient to hold objects of up to 0.5 kg, an existing mechanically activated prosthesis. The adequate for daily tasks. A more powerful servo proposed design is simple and robust, and can be motor can also be installed in the described design if implemented with minimal resources and expertise. a more heavy duty prosthetic is required. The design uses only readily available, off-the-shelf parts and can be implemented at a total cost of less than 2% of the cheapest alternative, which together with the most important holding-capability successfully fulfilled by the design offers feasible alternative for wrist amputees in developing countries who desire to take advantage of powered myoelectric prosthetic technology, which has substantial potential to improve the quality of life of such amputees.

Acknowledgements

The authors acknowledge financial support via Figure 7. The modified prosthetic was able to grab and TUBITAK ARDEB 1001 grant No 113E611. hold objects up to 0.5 kg in weight controlled by EMG activation. References

Currently in the market, four myoelectric hand/arm [1] Day, S. (2002). Important factors in surface EMG prosthetic offerings are available. These are iLimb, measurement. Bortec Biomedical Ltd publishers,1-17. Michelangelo, BeBionic, and DEKA Arm [18]. Most [2] Maier, S. (2008). Surface EMG suffices to classify the of these products are similar in design and features, motion of each finger independently. Proceedings of and focus on providing a complex hand actuator MOVIC 2008. which, nevertheless, is still activated via single EMG [3] Cipriani, C., Antfolk, C., Controzzi, M., Lundborg, G., sensor and resolves a single muscle activation, Rosén, B., Carrozza, M. C., & Sebelius, F. (2011). exactly as in our design. The cost of such prosthetics, Online myoelectric control of a dexterous hand prosthesis by transradial amputees. IEEE however, is immeasurably greater than ours. The Transactions on Neural Systems and Rehabilitation cheapest of the commercial myoelectric prosthetics Engineering, 19(3), 260-270. (Touch Bionic’s iLimb) costs close to 20,000 USD. [4] Resnik, L., Meucci, M. R., Lieberman-Klinger, S., The design described in this work has been Fantini, C., Kelty, D. L., Disla, R., & Sasson, N. implemented for under 350 USD, where 50-100 USD (2012). Advanced upper limb prosthetic devices: had been spent on electronics parts and 250 USD was implications for upper limb prosthetic the cost of the original conventional Otto Bock rehabilitation. Archives of physical and prosthesis. Nevertheless, our prosthetic design rehabilitation, 93(4), 710-717.

482 TEM Journal – Volume 6 / Number 3 / 2017. TEM Journal. Volume 6, Issue 3, Pages 479-483, ISSN 2217-8309, DOI: 10.18421/TEM63-07, August 2017.

[5] Borgia, M., Latlief, G., Sasson, N., & Smurr-Walters, [12] Ozkaya, U., Coskun, O., & Comlekci, S. (2010, L. (2014). Self-reported and performance-based May). Frequency analysis of EMG signals with outcomes using DEKA Arm. Journal of rehabilitation Matlab sptool. In Proceedings of the 9th WSEAS research and development, 51(3), 351. international conference on Signal processing (pp. [6] Cipriani, C., Segil, J. L., Birdwell, J. A., & ff Weir, R. 83-89). F. (2014). Dexterous control of a prosthetic hand [13] Tacar, B., Kaya, T., & Gulten, A. (2014, April). using fine-wire intramuscular electrodes in targeted Control of robotic hand simulator via human hand extrinsic muscles. IEEE Transactions on Neural motion analysis based on EMG. In Signal Processing Systems and Rehabilitation Engineering, 22(4), 828- and Applications Conference (SIU), 836. 2014 22nd(pp. 389-392). IEEE. [7] Fukuda, O., Tsuji, T., Kaneko, M., & Otsuka, A. [14] Grove EMG Detector, http://wiki.seeed.cc/Grove- (2003). A human-assisting manipulator teleoperated EMG_Detector, retrieved 19 June 2017. by EMG signals and arm motions. IEEE Transactions [15] Mainardi E., Urbano E., Davalli A. Design of a new on Robotics and Automation, 19(2), 210-222. EMG sensor for upper limb prosthetic control and real [8] Matrone, G. C., Cipriani, C., Carrozza, M. C., & time frequency analysis. In: Proceedings of the 2008 Magenes, G. (2012). Real-time myoelectric control of MyoElectric Controls/Powered Prosthetics a multi-fingered hand prosthesis using principal Symposium; 13-15 August 2008; Fredericton, New components analysis. Journal of neuroengineering Brunswick, Canada. and rehabilitation, 9(1), 40. [16] Ahmad, I., Ansari, F., & Dey, U. (2012). A Review of [9] Tenore, F. V., Ramos, A., Fahmy, A., Acharya, S., EMG recording technique. International Journal of Etienne-Cummings, R., & Thakor, N. V. (2009). Engineering Science and Technology, 4(2), 530-539. Decoding of individuated finger movements using [17] De Luca G. Fundamental concepts in EMG signal surface electromyography. IEEE transactions on acquisition. Natick, MA, USA: DelSys Inc; 2001 , 56(5), 1427-1434. [18] Resnik, L., Klinger, S. L., & Etter, K. (2014). The [10] Touch Bionics: Product Catalog. Istanbul, Turkey: DEKA Arm: Its features, functionality, and evolution Simay Medikal Distributor, 2016. during the Veterans Affairs Study to optimize the [11] 3M Red Dot electrodes application instructions. St. DEKA Arm. Prosthetics and Paul, MN, USA: 3M Inc; 2016. international, 38(6), 492-504.

TEM Journal – Volume 6 / Number 3 / 2017. 483