3D Printed Prosthetic Hand withIntelligent EMG Control

Timothy Inglis1007168555

Supervisor: Dr. Leonard MacEachernApril 10, 2013

Contents

List of Tables 2

List of Figures 2

1 Introduction 7

1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.2 Prosthetic Hand Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 The Engineering Project 11

2.1 Health and Safety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2 Engineering Professionalism . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.3 Project Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3 EMG Signals and their Applications in Prosthetics 14

3.1 Fundamentals of Surface Electromyography . . . . . . . . . . . . . . . . . 14

3.2 Application of Electromyography in Prosthetics . . . . . . . . . . . . . . 15

4 EMG Acquisition System 18

4.1 Electrode Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4.2 Pre-Amplification Stage . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4.3 Signal Digitization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4.4 Microcontroller Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

5 Digital Signal Processing 22

5.1 Fast Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5.2 Advantages of DSP Filtering . . . . . . . . . . . . . . . . . . . . . . . . . 23

6 Grip Pressure Acquisition 24

6.1 Force Sensor Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

6.2 Force Sensor Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

1

7 User Interface and Hand Actuation System 27

7.1 Theory of Operation for Multi-Input System . . . . . . . . . . . . . . . . 27

7.2 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

8 Mechanical Hand Design and Manufacture 33

8.1 Mechanical Design Requirements . . . . . . . . . . . . . . . . . . . . . . 33

8.2 Preliminary design work . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

8.3 3D Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

8.4 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

8.5 3D Printing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

8.6 Motor Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

8.7 Assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

8.8 Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

8.9 Benefits of a 3D printed hand . . . . . . . . . . . . . . . . . . . . . . . . 43

2

List of Tables

1 Number of Discrete Functions Availible in MIUI System . . . . . . . . . 28

2 Active Channels and Their Corresponding Opcode Numbers . . . . . . . 29

3 Implemented Hand Functions and Their Corresponding Opcodes . . . . . 30

4 Turnigy TGY 1501 MG: Hobby Servos . . . . . . . . . . . . . . . . . . . 39

5 Gantt Chart Displaying the Project Schedule for the Fall Term . . . . . . 45

6 Gantt Chart Displaying the Project Schedule for the Winter Term . . . . 46

List of Figures

1 Top Level Overview of the Electronic Control System for the Prosthetic

Prototype . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Single Motor Unit Activation Clearly Visible . . . . . . . . . . . . . . . . 14

3 Multiple Motor Unit Activation Appears Extremely Noisy . . . . . . . . 15

4 Myoelectric Arm Control Loop . . . . . . . . . . . . . . . . . . . . . . . . 16

5 Single Channel Amplitude Control . . . . . . . . . . . . . . . . . . . . . 17

6 Duel Channel Amplitude Control . . . . . . . . . . . . . . . . . . . . . . 17

7 Hhalf-cell potentials for various commonly used electrode materials [4, p.196] 19

8 FSR Signal Processing Circuit . . . . . . . . . . . . . . . . . . . . . . . . 25

9 FSR Pressure Distribution Modification . . . . . . . . . . . . . . . . . . . 26

10 Push-Button Implementation of Multiple Impulse User Interface . . . . . 32

11 Basic Mechanical Hand Operation . . . . . . . . . . . . . . . . . . . . . . 34

12 Prototype Hand Assembly . . . . . . . . . . . . . . . . . . . . . . . . . . 35

13 Prototype Hand Printing in Progressy . . . . . . . . . . . . . . . . . . . 37

14 Illistration of PWM signal [1] . . . . . . . . . . . . . . . . . . . . . . . . 39

15 Dorsal Side of Assembled Prototype . . . . . . . . . . . . . . . . . . . . . 40

3

Abstract

Electromyography (EMG) is the most common control interface for modern, upper-

limb prosthetics. Prosthetics that make use of EMG interfaces are commonly referred to

as myoelectric arms. The hands used in these devices are known as myoelectric hands.

The costs of commercially available myoelectric hands are very high, ranging in price

from $15,000 to $50,000 [2]. Additionally, the repair of these hands usually requires

expensive proprietary components and in almost all cases, a trained professional must

conduct repairs. For many amputees, in Canada and else ware, this cost barrier makes

the use of a myoelectric prosthetic impractical or impossible.

At an approximate cost of $15,000, the most affordable and widely used myoelectric

hands suffer from a severe lack of functionality compared to a human hand. These

commonly used myoelectric hands afford the user only a single grip type: the pinch grip.

The EMG control system employed by these hands allows the user to control the hands

speed, but they must pay attention at all times when holding an object to avoid crushing

it through accidental hand actuation.

There are hands that offer dramatically superior functionality when compared to these

most affordable hands. For example, the I-Limb and BeBionic hands have individual

finger movement and the ability to switch between multiple grips. However, these hands

are prohibitively expensive for all but those amputees with the best funding.

Recognizing the need for an alternative to currently available technology for those with

limited resources, we were able to develop a prototype hand with similar functionally

to the more sophisticated myoelectric hands on the market while reducing the cost of

the hand to well below that of the most affordable myoelectric hands that are currently

available. Our team’s low-cost objective was achieved by developing an inexpensive EMG

control platform and a mechanical hand that is designed to be produced inexpensively

on a 3D printer. Our goal of achieving a high level of hand functionality was achieved by

implementing a user interface that allows many more hand functions to be called by the

user than current myoelectric hand control systems. This Multiple Impulse User Interface

also reduces the chance of a stray muscle impulse causing unintended actuation of the

4

hand; reducing the mental task load for the user of the device.

EMG signals are acquired from the residual limb of the user, amplified and then dig-

itized with a high resolution analog to digital converter. A serial peripheral interface is

employed to transfer the digitized signal from the analog to digital converter to a micro-

controller which identifies muscle impulses and actuates motors in the printed hand. The

function that the mechanical hand preforms is determined by the sequence of impulses

produced by the user in quick succession. If the control logic of the hand detects a func-

tion call, it will provide haptic feedback to the user alerting them to the device status.

A pressure sensor in the thumb of the prosthetic hand provides feedback to the hands

control logic, reducing the risk of dropping or crushing an object being held in the device.

5

Acknowledgements

I would like to thank my team mates Alim Baytekin, Alborz Erfani, and Natalie

Levasseur for all of the hard work and long nights they committed to this project over

the school year. I would like to thank Dr. Leonard MacEachern for agreeing to supervise

a project that we were all so passionate about and for all the encouragement and advice

he provided us. I would like to thank Mark Kilbanov and Nick Stupich who were always

willing to answer questions and provide support, even very late into the night and on

the weekends. I would like to thank the War Amps and Smiths Prosthetic Services for

providing background information and lending their support to the project. I would like

to thank the Carleton University Department of Mechanical Engineering, particularly

Stephan Biljan, for printing our prototype hand. I would like to thank all the faculty

members in the Carleton University Department of Electronics and the Department of

Systems and Computer Engineering for all of their collective advice and support. Lastly,

I would like to thank the Lab Manager in the Department of Electronics at Carleton

University, Nagui Mikhael. Nagui is an incredible individual who takes a personal interest

in the success of every student who works in his labs. We could not have completed our

prototype without his support.

6

1 Introduction

A low-cost 3D printed prosthetic hand with intelligent EMG control was designed

in the department of electronics at Carleton University as a fourth year engineering

design project. The team members who worked on the project were Alim Baytekin,

Alborz Erfani, Natalie Levasseur, and myself. Dr. Leonard MacEachern supervised the

design and manufacture of the prototype prosthetic hand. The intention of this report

is to provide a technical overview of the design of the prototype that was developed.

Particular emphasis will be placed on the design of the mechanical hand, the intelligent

motor control logic, haptic feedback implementation, and the integration of all of the

individual system components into a functional prototype.

1.1 Background

The cost of a modern myoelectric prosthetic hand in Canada ranges from $15,000

to $50,000. The average arm amputee owns a more moderately priced device costing

about $15,000 [2]. The high cost associated with myoelectric hands creates a significant

economic barrier to ownership for many amputees in Canada and abroad. Additionally,

hand functionality is limited among these more moderately priced devices. Many of these

hands only have the ability to open and close in a single grip.

Meanwhile, the functionality afforded by high end, upper-limb prosthetics has never

been greater. Products such as the I-Limb and the BeBionic hand operate using inde-

pendently actuated fingers and are capable of many grips. Unfortunately, the prohibitive

costs associated with these devices prevent all except the best funded amputees from

taking advantage of this revolutionary technology.

There is a clear deficiency of low-cost, high functionality devices in the upper-limb

prosthetics market. The goal of this project is to address this deficiency by developing a

myoelectric hand that has similar functionally to the most expensive hands available at

a cost dramatically lower than the least expensive hands.

In order to accomplish this goal, two parallel systems had to be developed: an in-

7

expensive electromechanical hand that was a reasonable analog of a human hand and

an inexpensive EMG based control platform. In order to keep the costs down and has-

ten manufacturing, the team decided that the mechanical hand components should be

printed using a 3D printer. For the same reasons, the team decided that the EMG control

platform should be comprised of inexpensive, readily available components.

3D Printing technology has been around since 1986 [3]. The cost associated with the

technology and the fragility of the parts that were produced prevented the widespread

adoption of 3D printing as a manufacturing technique. Over the past few years, this

paradigm has begun to shift. Introduction of high quality consumer 3D printers, such

as the Makerbot Replicator and the Bits from Bytes RapMan, have made 3D printing

available to a much wider audience. There are now more than 24 consumer-grade 3D

printers available for under $5000 [4]. The majority of these printers use fused deposition

modeling (FDM) as their printing technique, which provides the most robust parts of any

3D printing technology.

EMG signal acquisition systems are traditionally composed of analog filters and gain

stages for each channel. In order to meet our goal of using fewer components than a

traditional EMG acquisition system, our team explored a newer approach to the problem.

A high resolution analog to digital converter can be used to capture the signal with a

resolution of only a couple hundred Nano Volts. This level of resolution allows all of the

filtering to be done in the digital domain.

All of the signal processing must be performed on a microcontroller. Fortunately,

the cost and accessibility of microcontrollers has been improving over the past few years.

Arduino microcontrollers are still a mainstay in the inexpensive microcontroller category,

characterized by their low cost and simple C++ based programing environment. Other

more powerful microcontrollers are also beginning to enter the market. The fastest Ar-

duino currently available, the DUE, runs at a clock frequency of 84MHz. Even more

speed can be found in the Beaglebone, a new, Arduino-sized microcontroller tha operates

at up to 700MHz.

As the quality of, and access to advanced components and manufacturing techniques

8

improves, many previously specialized applications will begin to see lower cost alterna-

tives. One of these areas of growth is prosthetics and our 3D-printed prosthetic hand

exemplifies that trend.

1.2 Prosthetic Hand Overview

The low-cost prosthetic hand developed for this project consists of four primary com-

ponents: a 3D-printed electromechanical hand, an EMG interface, a microcontroller capa-

ble of real-time signal processing, and a stable embeded control system. The 3D-printed

hand prototype was modeled, printed and assembled for less than $250. The hand con-

tains over 30 components, including 15 unique printed components. It is actuated with

high-torque hobby servos that are controlled by pulse width modulated (PWM) signals

regulated by the microcontroller, an Arduino Due. Figure 1 shows a basic overview of

the hand’s electronic control system.

Figure 1: Top Level Overview of the Electronic Control System for the Prosthetic Prototype

9

The EMG interface works by acquiring differential signals from muscle impulses in

the residual limb of the user. Those signals are then amplified, and pass them to a high

resolution analog to digital converter (ADC). The ADC then outputs the signals over a

serial peripheral interface (SPI) to the microcontroller.

The Arduino processes the signals by preforming a Fast Fourier Transform (FFT)

which converts the signals from the time domain to the frequency domain. Once a

signal is in the frequency domain, the magnitude of the relevant frequency bins can be

calculated. If the magnitude of these relevant bins exceeds a threshold value, a muscle

impulse is said to have been detected on that channel.

Control logic embedded in the microcontroller captures combinations of muscle im-

pulses across all of the available channels, henceforth known as opcodes. The control

logic analyzes sequences of opcodes, and actuates the motors in the hand in order to

perform the function dictated by the given opcode sequence. This novel method of using

sequences of opcodes to dictate prosthetic hand function is called a Multiple Impulse

User Interface (MIUI).

A pressure sensor that resides on the gripping surface of the prosthetic thumb provides

feedback to the microcontroller through an onboard ADC. This feedback is used to control

the pressure that is applied when the prosthetic hand grips an object. The control system

is also equipped with a haptic feedback system, that causes small motors to vibrate in

response to a successfully received command.

10

2 The Engineering Project

As participants in an engineering project, it was important that our team considered:

how our product might affect the health and safety of its users, what the role of ethics

and engineering professionalism was in the project, and the way that the project was

managed. The ways in which our device might affect the health and safety of users

must be addressed at every level of the design process. Myoelectric devices are classified

as class 2 medical devices in Canada because they are electrically powered [5]. Special

precautions, such as power source isolation must be taken to ensure that the risk of an

electric discharge is minimized. As engineering students, it is imperative that we adhere

to the professional standards and practices that are expected of professional engineers.

The design and manufacture of a prototype that met all of the project objectives prior

to the end of the school year would not have been possible without carful and deliberate

project planning. A carful division of labor, regular team meetings, and proper time

management were essential to the completion of the project.

2.1 Health and Safety

There are three primary factors to consider when assessing the electrical safety of

any medical device: the magnitude of the potential current, the duration of current flow,

and the pathway the current takes. A myoelectric arm carries a large battery capable of

driving 3 Amps of DC current in order to power the motors at maximum torque. This is

more than sufficient current to cause respiratory paralysis, pain and fatigue. It may also

be enough current to trigger a sustained myocardial contraction. A sustained myocardial

contraction, though unplesent, is not necessarily fatal. The heart stops beating during

the contraction, but a normal heart beat usually returns when current is no longer applied

[6, p. 640].So, as long as the current is not sustained for an extended period, lethality is

unlikely.

The chance of a 3A discharge into the body is small, however, much less current

than this, applied to a vulnerable part of the body can still be extreamly dangerous.

11

The path of the current has a significant effect on the seriousness of any electric shock.

A microshock occurs if current passes through a persons cardiac tissue. Microshocks

are much more likely than macroshocks to be lethal, even with an applied current as

small as 10µA [6, p. 653]. Although the magnitude of current that is required to cause

fibrillation of the heart is far larger when applied through the skin, due to the potential for

harm as a result of small leakage currents passing through cardiac tissue, it is important

that proper precautions are taken to prevent leakage currents in production devices. The

primary solution for mitigating electrical risk in medical devices is to use an isolated power

supply system. Though we did not include proper isolation in this first design iteration,

an optical or capacitive isolation stage should be implemented in the next iteration. The

isolation stage should be placed between the electrodes on the users residual limb and

the microcontroller. This will limit the possible current that the user would be exposed

to if there was an electrical discharge into the body.

2.2 Engineering Professionalism

Canadian Engineers are bound by legal and Ethical responsibilities, outlined In On-

tario by the Professional Engineers of Ontario (PEO) the regulating body for professional

engineering in Ontario. The PEO defines these responsibilities in its Code of Ethics, lo-

cated in section 77 of the Professional Engineers Act[7]. In accordance with the principles

outlined in the Code of Ethics, our team acted with courtesy toward one another, other

teams and the Carleton University faculty.

Our Team also acted professionally by being honest and forthcoming in the face of

uncertainty. Team members did not impart a false impression of competence which could

have led to delays, mistakes or even dangerous situations for team members or the public.

A concerted effort was made to keep records of design decisions and the responsibilities

of each team member. These records were developed in the form of the project proposal,

the progress report, and countless email interactions. These records will be useful if there

is ever any inquiry surrounding the work performed over the course of the project.

12

2.3 Project Management

To ensure that all the project objectives were met, the project was sub-divided into

four areas of responsibility at the beginning of the school year. Due to the multi-

disciplinary nature of this undertaking, a carful division of labor was essential to ensure

that each team member was contributing in a way that would optimize their skill set.

Although there was significant cooperation between all of the group members on every

component of the project, the division of labor devised at the projects outset was quite

representative of the roles carried out by each of the group members throughout the

course of the project. Alborz Erfani developed the EMG interface. This included circuit

design of the pre-amplification stage and the ADC stage of the system. It also included

the implementation of an SPI driver in the microcontroller to facilitate data acquisition

from the ADC. Natalie Levasseur developed the pressure sensor array for grip pressure

control. This included selecting a force sensor and modifying it to ensure a consistent

response in varying use cases. Natalie also implemented the pressure acquisition stage

of the hand control logic; programing the Microcontroller to read the onboard ADC and

return actual force values. Alim Baytekin implemented the digital processing of the EMG

signals once they were acquired. This included transferring functions from time domain

to frequency domain by preforming a real time FFT. Additionally Alim filtered the sig-

nal and characterized the muscle impulses on the various channels. The element of this

project that I focused on was the design of the mechanical hand, the intelligent motor

control logic, haptic feedback implementation, and the integration of all of the individual

system components into a functional prototype.

In addition to a well-defined division of labor, at the beginning of the project, the

group decided on a schedule for the achievement of various project milestones. These

included several design and test periods. This schedule was very helpful in ensuring we

met the overarching goals of the project on time. A detailed description of the project

timeline is given in the Gantt charts given in Table 5 and Table 6 in Appendix A.

13

3 EMG Signals and their Applications in Prosthetics

The process of recording and interpreting changes in electrical potential produced

by contracting skeletal muscles is called Electromyography or (EMG) [8, p. XV]. EMG

signals can be used to detect when a muscle, or group of muscles have contracted. This

makes EMG a very good candidate for upper limb prosthetic device control. EMG allows

myoelectric hand users to manipulate their artificial limbs using muscles on their residual

limb that would otherwise remain unused.

3.1 Fundamentals of Surface Electromyography

The basis of skeletal muscle organization is the motor unit (MU). Each MU is com-

posed of a motor nerve fiber attached to a bundle of muscle fibers. A MU is the smallest

muscle component that can be activated voluntarily. The activation of a single MU man-

ifests itself as a distributed biopotential due to the superposition of action potentials of

all of the muscle fibers in the bundle. Depending on the size of a MU, the amplitude

of a measured EMG signal also known as a myoelectric signal is between 20µV and

2000µV [6, p. 145]. The frequencies of interest when recording EMG signals are about

30Hz to 300Hz. When intramuscular probes and proper signal filtering are employed, it

is possible to contract the muscle in such a way that the distinctive activation of a single

MU can be observed as illustrated in Figure 2.

Figure 2: Single Motor Unit Activation Clearly Visible

As the force of contraction increases, it becomes impossible to distinguish an individual

MU in the time domain signal. The reason for this is a combination of increased MU

recruitment (more MUs firing) and increased MU activation rate [8, p. 2]. The result is

a time domain signal that resembles stochastic noise as illustrated in Figure 3.

14

Figure 3: Multiple Motor Unit Activation Appears Extremely Noisy

This noisy, distributed signal increases the difficulty of isolating the EMG signal from

a single muscle. The problem is compounded when a surface EMG(SEMG) acquisition

strategy is employed. Unlike intramuscular, needle based myoelectric signal acquisi-

tion, SEMG requires a relatively large superficial muscle or pronounced muscle group

[8, p. 455]. A closely spaced electrode pair positioned along the muscle fiber is required

to acquire a highly localized signal, suitable for a single muscle control channel. If the

bipolar electrode pair is spaced a little further apart, then the signal that is recorded will

be a spatial and temporal superposition of the electrical activity of all of the active MUs

in the muscle group [8, p. 455].

The interpretation of these complex signals is further complicated by electrical noise.

The primary source of electrical noise in all bioinstrumentation applications results from

capacitive coupling between the human body and power delivery lines. In North America

electricity is supplied at about 60Hz, so this is the primary frequency at which noise

occurs [9]. The noise is also observed at the harmonic frequencies, integer multiples of

60Hz, including 120Hz and 180Hz.

3.2 Application of Electromyography in Prosthetics

A myoelectric hand is an artificial hand whose control system is based on Electromyo-

graphy . Myoelectric prosthetic devices are controlled by purposeful muscle contractions,

usually in the users residual limb. Tight fitting prosthetic sockets hold dry electrodes

against prominent superficial muscles or groups of muscles in the users residual limb.

15

These electrodes register surface EMG signals. By carefully controlling muscle contrac-

tions, myoelectric arm users are able to open and close basic myoelectric hands and

switch between active functions in multi-function hands. The control loop employed in a

myoelectric arm is shown in figure 4.

Figure 4: Myoelectric Arm Control Loop

Myoelectric control systems come in single and multi-channel configurations. Typi-

cally, children and those with little muscle in their residual limb are fitted with single

channel control systems, while most adults are fitted with two-channel control systems.

Using a single muscle per control channel is impractical because it requires precise

placement of an electrode pair over the muscle being observed. Muscle group signal

acquisition is much more forgiving in the exact placement of the electrodes, which is a

major consideration given the tendency of prosthetic sockets to shift while in use. For

this reason muscle group signal acquisition is a much more common control approach

[8, p. 455]. It is simpler to obtain and process more complex EMG signals than it is to

attempt to maintain single muscle channel control in a real world prosthetic.

In in order to control a myoelectric prosthesis, features that indicate the intention

of the user must be extracted from these complex signals. Most commercially available

devices use signal amplitude to elicit a control system response. This control technique

is commonly employed in single and multi-channel devices as is shown in Figure 5 and

Figure 6 respectively.

Pattern recognition control systems have been employed in the past to operate multi-

16

Figure 5: Single Channel Amplitude Control

Figure 6: Duel Channel Amplitude Control

function control systems. These systems are not normally used do to their lack of intu-

itiveness and the basic functionality of most limbs in use today. A limb with only two

functions requires a simple control system.

17

4 EMG Acquisition System

There are four stages in the EMG acquisition circuit: the Electrodes, the pre-amplification

stage, the signal digitization stage, and the interface between the analog to digital con-

verter and the micro controller. This chapter gives a brief overview of these stages,

however, a more detail technical examination of the EMG signal acquisition system can

be found in the report submitted by Alborz Erfani.

4.1 Electrode Selection

An interface is required to acquire biopotential signals generated from voluntary mus-

cle impulses in the residual limb of a myoelectric arm user. Biopotential electrodes are

used to meet this requirement. A biopotential electrode is a kind of transducer, convert-

ing current carried by ions in the body to current carried by electrons in the electronic

circuitry [6, p. 189]. The core process that permits this transduction at the electrode

electrolyte interface is a redox chemical reaction that takes between the electrode and

electrolyte. The general reaction that occurs is given below.

C ⇀↽ Cn+ + ne− (1)

Am− ⇀↽ A+me− (2)

These reactions occur continuously when the electrode is in contact with the elec-

trolyte, even when there is no current passing across the interface. In the case of no net

current, the oxidation and reduction reactions balance. The oxidation reaction dominates

when there is net current flow from the electrode to the electrolyte, and the reduction

reaction dominates when the net current flow is from the electrolyte to the electrode [6,

p. 190].

When an electrode is placed against an electrolyte, the local concentration of ions

around the electrode in the electrolyte will change, so the electric potential of the elec-

trolyte in this region will differ from the electric potential in other parts of the electrolytic

solution. This potential difference is known as the half-cell potential of the interface. The

18

half-cell potential of an interface is determined by the metal in the electrode, the concen-

tration of ions in the electrolyte solution, and the temperature [6, p. 191].

It is not possible to experimentally measure the half-cell potential of an interface

without using another electrode which carries its own half-cell potential. So, as a way of

objectively comparing the half-cell potentials of different metals, the standard practice is

to measure the half-cell potential of an electrode against a standard hydrogen electrode

that has a defined half-cell potential of zero.

Figure 7: Hhalf-cell potentials for various commonly used electrode materials [4, p.196]

These half-cell potentials are representative of conditions where there is no net current

crossing the interface. When there is a current and the observed half-cell potential differs

from the zero-current half-cell potential, the electrode is said to be polarized. This differ-

ence between the observed half-cell potential and the the zero-current half-cell potential

is known as the over-potential [6, p. 192].

There are two possible categories of electrode that can be employed in EMG acquisi-

tion. The industry standard for prosthetic control systems is to use use dry electrodes,

whose interface is composed of silver. These reusable electrodes are fit into the wall of a

prosthetic socket. In our design we employed disposable, pre-gelled, silver/sliver-chloride

19

electrodes. The advantage of these electrodes is that they are much less expensive, readily

available and are nearly ideally non-polarizable. This greatly simplifies the modeling of

the electrode skin interface.

4.2 Pre-Amplification Stage

The amplitude of the electrical signal emitted by a single motor unit is in the range

of 20µV to 2000µV [6, p. 145]. These signals must be amplified in order to make use

of the entire dynamic range of the ADC, the Texas Instruments ADS1298. In order to

achieve this pre-amplification; an instrumentation amplifier with a gain of about 100 was

employed for each EMG channel. The instrumentation amplifier that was used was the

single supply AD627.

4.3 Signal Digitization

Digitization of the signal was preformed by the TI ADS1298, a bioinstrumentation

front end, which includes a 24-bit high resolution ADC. The purpose of pre-amplification

of the Analog EMG signal is to optimize the utilization of the ADS1298’s dynamic range.

The 24-bit ADC more than 16.7 million quantization levels spanning a 5.6V input range.

This gives the ADC roughly 341nV of accuracy on each EMG channel.

The ADS1298 has many options that are easily programmed by sending opcodes to the

device over the Serial Peripheral Interface. This allows the device to be easily customized

to suit many different applications. For our application, each analog channel had a gain

of 1, full 24-bit resolution was employed and a sampling occurred at a rate of 1kHz.

4.4 Microcontroller Interface

Once the EMG signals are digitized, they need to be delivered to the microcontroller

for processing. This occurs over a Serial Peripheral Interface (SPI). SPI is a four line

interface that allows high frequency serial transmission of digital data. The four lines of

an SPI interface are, Master In Slave Out (MISO), Master Out Slave In (MOSI), Chip

Select (CS), and Serial Clock (SCLK). The MISO is the line on which serial data is fed

20

into the microcontroller. The MOSI line allows commands to be sent to the ADS1298

from the microcontroller in the form of opcodes. These can be used to change attributes

of the ADS1298 operation, such as: the gain applied to each of the analog channels, the

resolution of the ADC, and the sampling rate of the analog channels. The CS line must

be low in order for the SPI system to function with the ADS1298. In applications where

multiple ADS1298s are employed, this line is used to select which device is currently being

accessed. In our application there is only one ASD1298, so we are able to permanently

tie this line low. The SCLK line dictates the rate of data transfer over the SPI interface.

The master device, in our case the microcontroller, supplies a clock signal on the SCLK

line. One bite of data is transferred on the MISO and MOSI lines on each clock cycle.

This means that SPI communication occurs in full duplex for every bit sent, another bit

is received simultaneously. In addition to these four core lines, a fifth line is used when

interfacing the ADS1298. The Data Ready (DR) line is used to tell the microcontroller

that the ADS1298 has a set of data in its buffer that is ready to be sent.

The ADS1298 captures eight channels of data in parallel. With 24 bits of data per

channel, this amounts to 192 bits of signal data in each transfer. When the onboard data

buffer is full, the ADS1298, sets the data ready pin high. An edge triggered interrupt

in the microcontroller begins an ISR which initiates the SPI transfer. At this point the

microcontroller begins outputting a serial clock at 2MHz and duplex data transfer begins.

There are no opcodes to send to the ADS1298 in normal operation, so zeros flow out of the

MOSI line on every clock cycle, while data is being received by the microcontroller on the

MISO line. Each time a data transfer is initiated, a 24 bit header containing information

about the ADS1298 settings is sent prior to the 192 bits of sample data. So, for every

transfer 216 bits of data are sent. Data to be transferred over SPI is broken up, and

transferred as a series of 8bit words that can each be interpreted as 2 digit hexadecimal

numbers [10].

Once the data has been received by the microcontroller, it is separated by channel

and stored in a buffer for signal processing.

21

5 Digital Signal Processing

Once an EMG signal has been acquired and digitized by the EMG signal acquisition

system, it needs to be processed so that intentional muscle impulses can be detected. This

process involves converting the signal information from the time domain to the frequency

domain where the meaning of the data can be more easily ascertained. This section gives

an overview of the signal processing system, a more detailed description of the system

can be found in the report submitted by Alim Baytekin.

5.1 Fast Fourier Transform

When the data enters the microcontroller via SPI it needs to be divided into buffers;

one for each of the eight channels. The first three eight-bit words that arrive belong to

the header and can be ignored. The next three words belong to channel one, the following

three to channel two, and so on. Once the buffers are populated with a sufficient number

of samples, a Fast Fourier Transform (FFT) is preformed, which converts the data from

the time domain to the frequency domain.

The Cooley-Tukey FFT algorithm is applied for the conversion. The conversion uses

rectangular windowing and takes place for each of the eight channels with a window size

of 128 samples. This means that at a sampling frequency of 1000Hz, 128ms are required

for the buffers to fill before the first FFT can be calculated. Each time the FFT is

calculated, the buffers are purged and new samples repopulate it, and the next FFT is

computed. The resulting size of the frequency bins is 7.84576Hz and the latency is 128ms.

This latency can be reduced computing the FFT more than once every 128 samples. If

each sample that is added to the buffer replaces the oldest sample in the buffer, the FTT

could be preformed more often, thereby reducing latency.

When an FFT is performed, it returns real and imaginary components which together

make up the complex frequency domain signal. In order to extract useful information

from this signal, complex values in relevant bins must be converted into magnitudes.

These magnitudes can then be compared to a threshold value to determine if muscle

22

contraction has occurred.

5.2 Advantages of DSP Filtering

One of the major advantages of preforming signal processing in the digital domain

rather than in an analog circuit is that filtering becomes largely unnecessary. As dis-

cussed previously, the major source of noise in an EMG signal is power line noise. In

North America, power is delivered at a frequency of about 60Hz, so a band of frequency

spectrum between 55Hz and 65Hz can simply be ignored when analyzing the EMG signal.

This method of signal processing greatly simplifies the hardware design and allows for

modification of the filtering system in software. In other parts of the world the frequency

of power line noise may differ, but this system easily adapts to these differences by simply

ignoring a different part of the spectrum.

23

6 Grip Pressure Acquisition

One of the core goals of this project was to develop a system that would regulate

the pressure that the prosthetic hand applies to an object it is gripping. In order to

accomplish this, a pressure sensor needed to be selected, modified, characterized, and

implemented on the hand. The following section gives a brief overview of this process.

For a more comprehensive description, please see the final report of Natalie Levasseur.

6.1 Force Sensor Selection

Many force sensors were candidates for use in this project. The active component

of each of these sensors is a force transducer that converts mechanical pressure into an

electrical signal. The type of electrical signal varies for each type of load cell. The

two foremost options were a load cell whose resistive characteristic changed with applied

pressure, and one for which capacitance characteristics changed with applied pressure.

The decision was made to use a Force Sensing Resistor (FSR) for several reasons.

First, it is a robust system that can withstand repeated use. Second, it is a commercially

manufactured component that is very inexpensive and readily available. Third, the re-

sponse of an FSR is straightforward to understand and characterize, so integration with

the 12-bit ADC on our microcontroller presented no complications.

6.2 Force Sensor Integration

Once the pressure sensor was chosen, it was implemented using a simple opamp feed-

back circuit as was recommended by the manufacturer. This optimized the pressure

sensor output for the dynamic range of the 12-bit ADC on the microcontroller. The

circuit that was employed is illustrated in figure 8 below.

The position of the hand chosen for the force sensor was at the very tip of the thumb.

This location was chosen because most of the hand functions that involve holding an

object will activate the sensor. In future design iterations, many more force sensors will

be incorporated into the device, giving the hand much more control over how an object

24

Figure 8: FSR Signal Processing Circuit

is held and what fingers are supplying the pressure.

An important consideration in the implementation of the force sensor was how differing

force distributions across the surface of the FSR would affect its response. After some

experimentation, it was determined that the FSR requires a fairly evenly distributed force

across its surface to provide a consistent response. So, the force sensor was modified in

an attempt to fulfill this requirement. The modification involved sandwiching a soft piece

of rubber between the active surface of the force sensor and a piece of hard plastic as

illustrated in figure 9. The result was a sensor that would provide a consistent output

when a point force was applied to different areas of the device.

25

Figure 9: FSR Pressure Distribution Modification

26

7 User Interface and Hand Actuation System

Once a muscle contraction has been identified by the EMG acquisition system, that

information must be processed further to determine what, if any action should be taken

by the prosthetic hand. Two factors are considered when determining if an actuation

should occur. The first is determining if a multi-input command has been recognized.

The second is considering what kind of pressure is being applied at the thumb sensor.

7.1 Theory of Operation for Multi-Input System

Our hand uses a pattern based recognition system, herein referred to as the Multiple

Impulse User Interface (MIUI). MIUI works by receiving and interpreting combinations

of very short EMG impulses and carrying out a pre-determined sequence of commands.

For example, if the user had a two electrode channels, Channel A and Channel B, then

a possible code would be A-B. This could be the code for the ”Soft-Grip” function that

closes the hand around an object until it has a firm, but non-crushing grip.

The integrated pressure sensor provides feedback about grip strength and the motor

control logic maintains a firm grip, even in the case where there is some slippage. Once

the object has been successfully grasped, the user is notified by haptic feedback and they

are free to stop paying attention to what their prosthetic hand is doing, feeling confident

that it will not let go unless instructed to do so by another code.

This ”Soft Grip” example is only one of many functions that can be requested with

user customizable codes. The use of the ADS1298 as a front end allows more advanced

users to add up to eight channels to various muscle groups on their residual limbs to

improve signal acquisition accuracy and increase the number of available functions.

The number of functions that can be implemented with this system is dependent on

the number of channels being employed and the number of opcodes required to call a

function. The number of available opcodes is related to the number of channels by the

relationship:

O = 2n − 1 (3)

27

Where, O is the number of opcodes, and n is the number of channels. This is just the

number of possible channel activation combinations minus one. The minus one term

simply represents an all null (no contractions detected) opcode which should not be

included in any function calls to preserve system stability. The maximum number of

functions that can be called is simply the number of possible opcodes raised to the power

of the number of opcodes required to call a function:

F = (2n − 1)m (4)

Where, F is the maximum number of functions that can be called and m is the number of

opcodes required for each function call. The ADS1298 provides up to eight input channels

and there is no theoretical limit to the number of sequential opcodes that can be used

to call a function. However, there are significant trade-offs to consider when using more

channels and longer sequences of opcodes. Increasing the number of channels increases

the complexity of each opcode, while increasing the number of required opcodes increases

latency. A comparison of the number of possible functions relative to the number of

channels and required opcodes is given in table 1.

Table 1: Number of Discrete Functions Availible in MIUI System

28

A major advantage of MIUI over traditional prosthetic control schemes is the flexibility

it offers to the end user. As is evident from table 1 above, the MIUI control system is

customizable for each user. A child who receives their first myoelectric arm at the age of

three will use a very simple control scheme that requires one or two electrodes and only

a single opcode. As the child grows older, they will be able to incrementally increase the

number of channels and the number of opcodes they use, increasing the functionality of

their hand.

62% of child amputees who register with the War Amps are upper-limb amputees,

but only 11% of adult amputees are missing upper-limbs [2]. This suggests that many

arm amputees use myoelectric limbs for the vast majority of their lives, which gives them

decades to gradually improve their hand’s functionality. This makes a flexible system

that can grow with the user an ideal choice for those just beginning to use a prosthetic

arm.

7.2 Implementation

The control system that was implemented in the prototype hand is a two-channel,

two-opcode control system. Figure ?? in Appendix C shows an overview of the prototype

MIUI implementation. The MIUI system works by polling the EMG acquisition system to

see if any channels have experienced muscle contraction. If no contractions have occurred,

then the system continues to poll, waiting for the first contraction. Once a contraction

is detected, the system checks to see what channels have been activated and it registers

the corresponding opcode. Every possible combination of channel activations is given

an opcode number. The opcode numbers assigned for our prototype implementation are

given in table 2.

Channels with Contraction Opcode NumberA 1B 2

A & B 3

Table 2: Active Channels and Their Corresponding Opcode Numbers

Once the first opcode has been registered, the system begins polling for the second

29

opcode. If no opcode is registered within one second, there is a timeout and the system

begins polling for the first opcode again. If a second opcode is registered, the system sets

the digital pin that controls haptic feedback to high causing a small motor to vibrate. The

control system then activates hand function that corresponds with the opcode sequence.

A complete list of hand functions implemented in this prototype and their corresponding

opcode sequences are given in table 3.

Opcode Sequence Hand Function1,1 Fast Flat Palm2,2 Fast Fist3,3 Fast Index Finger Point1,2 Two Finger Pinch2,1 Three Finger Pinch3,1 Slow Grip With Pressure Sensor Feedback

Table 3: Implemented Hand Functions and Their Corresponding Opcodes

The the final position of each servo, the speed of the hand movement, and the max-

imum pressure that the hand will apply are all pre-defined components of the hand

functions. The angle of displacement of each servo is controlled by Pulse Width Modula-

tion. If the hands motors are not in their final position as dictated by the function that

has been called, they will move to that position. Some functions change the pulse width

of the PWM signal in a single step, causing the servos to move at their top speed; these

functions have the prefix ”Fast” in their names. Other hand functions ramp the pulse

width up or down in many steps resulting in slower hand movement.

With regard to the maximum pressure a function can apply, the prototype hand has

two types of functions: those that do not respond to pressure sensor feedback and those

that do respond to pressure sensor feedback. The only currently implemented command

that responds to pressure sensor feedback is the ”Soft Grip” function mentioned above.

In the case of this function, the motors are actuated by slowly ramping down the pulse

width of the PWM signal. As the hand slowly closes the pressure sensor is polled. As the

pressure on the sensor increases, the resistance of the FSR decreases causing the voltage

applied to the ADC to increase. Once this voltage reaches a desired threshold, motor

actuation ceases. Finally, after completeng this function, the hand is placed in ”pressure

30

sensitive mode”.

While in this mode, the control system will poll for the first opcode again. However,

because the last function to be called was pressure sensitive, the hand will continue to

poll the pressure sensor approximately every 100ms. If the pressure applied to the sensor

drops, it is indicative of a slipping object, so the control system will increase tension on

the motors to attempt to increase the pressure applied the pressure sensor and to get a

stronger grip on the object. The control system will leave ”pressure sensitive mode” as

soon as a function that is not pressure dependant is called.

This control system was fully implemented prior to the end of the project; however,

stability in the EMG acquisition system was not sufficient to interface it with the control

system. In order to prove the functionality of the control system, a simple push-button

implementation was constructed. Figure 10 shows the push-button implementation.

In this implementation, push-buttons were used to simulate confirmed muscle con-

tractions on two channels. In the figure above, the red button is a power button which

primes the system for operation. The black button represents channel A and the green

button represents channel B. These buttons directly replace the inputs that would oth-

erwise be supplied by the EMG acquisition and signal processing stages. The pushing of

a button is equivalent to the contraction of a muscle on the corresponding channel.

This push-button implementation worked flawlessly, demonstrating the effectiveness

of a simple Multiple Impulse User Interface.

31

Figure 10: Push-Button Implementation of Multiple Impulse User Interface

32

8 Mechanical Hand Design and Manufacture

8.1 Mechanical Design Requirements

In order to meet the functionality objectives our team set out for the prosthetic hand,

we had to set several design requirements for the mechanical hand component. First,

the hand had to be printable using a fused deposition modeling (FDM) 3D printer.

FDM provides the most robust physical product of all 3D printing technologies and it is

also the technology most widely available in inexpensive consumer grade printers. This

makes FDM the most accessible type of 3D printing. Second, the hand must be capable

of preforming multiple functions, such as pinch, bat grip, and index point. To accomplish

this, at least the thumb and index finger must be actuated independently of any other

fingers. Ideally, each finger should be actuated independently. Third, the design should

strive to accurately reproduce the look and style of a human hand. There should be four

parallel fingers, at least one of which is directly opposed by the thumb, so they can come

together in a pinch.

8.2 Preliminary design work

The concept of wire actuated fingers came from a simple toy that we were introduced

to at the start of the project. Cables running down the palmer side of each finger in the

toy caused the fingers to curl toward the palm as tension was applied. When tension on

a finger cable was released, the finger associated with that cable was extended due to the

elastic force of the bent plastic which was now free to return to its original shape.

Prior to modeling the mechanical hand in a solid modeling suite, several cardboard

mock ups of the final hand were developed to get a sense of how the final mechanical

system would behave. In these mockups, plastic zip-ties were employed to act as both the

palmer cables and the elastic recoil mechanism. In a final mock-up, elastic bands were

run the length of each finger on the dorsal side of the hand. This is the configuration that

was employed in the final prototype design. Figure 11 illustrates the design principles

that were employed in this final mock up and in the prototype hand.

33

Figure 11: Basic Mechanical Hand Operation

During the course of assembling these mock-ups, it appeared as though it would be

possible to implement an antagonistic force scheme for finger actuation. In this scheme,

two cables would run the length of each finger; one down the palmer side of the finger

and a second down the dorsal side of the finger. Both of the cables would be secured to

the same servo, on opposite ends of the actuator arm. When the servo rotated, tension

would be applied to the palmer cable while the dorsal cable would be supplied with slack.

The advantage of this scheme is that the motors would not be required to work against

an elastic force that was constantly working to retract the fingers. The final prototype

hand design did include the holes on the dorsal side of each finger that are required to

configure this type of actuation system. However, in order to be effective, an additional

34

cable suspension system would have been required as the displacement of the palmer

cable was not equal to the displacement of the dorsal cable. This or a similar system of

antagonistic actuation will be incorporated into the next design iteration of the prosthetic

hand.

8.3 3D Modeling

Solid modeling of the 15 unique hand components was performed in Auto Desk In-

ventor. The index, middle, ring and pinky fingers are each made up of a tip segment, a

middle segment and a base segment, which is connected to the palm of the hand. The

thumb is composed of just two segments but sits on a raised platform away from the

palm, allowing it to directly oppose the other fingers when contracted. Figure 12 shows

the assembly of the printed hand as well as the hands major design features.

Figure 12: Prototype Hand Assembly

35

The axel holes for each of the joint is 3mm, except for where finger segments interface

with the palm, in which case the axel holes were 5mm. The joint channel width differed

from joint to join and ranged in size from 1cm to 1.2cm. Each joint key was 1mm narrower

than its corresponding joint channel to reduce friction between the channel wall and the

channel key.

8.4 Simulation

Once initial models of all of the components had been completed, they were virtually

assembled inside Autodesk Inventor. This was accomplished by constraining the parts in

the model to each other about the axis of revolution of each of the joint that connects

them. Virtual assembly of the hand allowed for the examination of the entire system to

ensure its mechanical function was adequate. If any feature of a component did not mesh

with another component, then modifications to one or both components were performed.

There were several features of the mechanical system that had to be examined prior

to dynamic simulation in order to ensure basic functionality. First, the holes for the joint

axels had to be of equal size and line up for each interfacing part. Second, the width of

each joint canal must be slightly larger than the width of the joint key of the interfacing

part. I chose to make this difference 1mm. Third, the depth of the joint canal must be

large enough to contain the entire joint key of the interfacing part.

Once the basic geometry of the components meshed correctly, the model could be

simulated dynamically. To accomplish this, a dynamic constraint solver was employed.

This function prevents the parts, which are constrained about their respective joint axes

from moving through an area that is occupied by another physical object. Simply stated,

it prevents the model from moving in a way that would not be possible were it a physical

object. If a finger stopped as it was moved through its desired range of motion, it

was clear that there was a physical conflict. In this case one or both of the interfering

parts would be modified and the motion would be re-simulated. This was an iterative

process, requiring dozens of simulations and modifications before a satisfactory model

was generated.

36

8.5 3D Printing

Upon completion of virtual mechanical simulations, the hand was ready for manufac-

ture. The method of manufacture selected was rapid prototyping on a Fused Deposition

Modeling (FDM) 3D printer. Fused Deposition Modeling, also known as Fused Filament

Fabrication is a type of additive manufacturing where a thread of molten plastic is used

to trace out a layer of a part in the X-Y plane. Once an entire layer is traced, the print

platform is lowered and the next layer is printed.

The prototype hand model was converted into stereolithography (STL) files in Au-

todesk Inventor. These STL files were loaded into the printer’s software, arranged for

printing, and converted to G code. G code is the control code that provides the printer

with instructions regarding the velocity of the print head, extruder temperature and the

filament extrusion velocity. The hand parts were printed in 18 hours on a Dimension SST

3D printer. The plastic used to print the prototype hand was Acrylonitrile Butadiene

Styrene (ABS) Plastic. The Dimension SST has a layer thickness of 0.254mm.

Figure 13: Prototype Hand Printing in Progressy

37

Figure 13 shows the hand about one third of the way into the print. The white

material being printed is the ABS plastic that will constitute the finished parts and the

black printed material is support material that dissolves when exposed to a weakly basic

solution. After the print is complete, parts are submerged in a bath of basic solution and

the support material takes about one hour to dissolve away.

8.6 Motor Selection

Hobby servos are DC motors that have internal gearing, integrated motor drivers and

internal feedback systems. Hobby servos were selected for use in this first prototype of

our prosthetic hand because their inclusion greatly simplified the design and helped to

ensure that there was a working prototype for the end of the project. The servos used

in this project are single-rotation servos, meaning they are unable to continuously rotate

and have maximum and minimum angular displacements. The drive shaft of each servos

motor is attached to a potentiometer. This provides feedback to the motors internal

driver, indicating the current angle of displacement. The motors angular displacement is

set by the microcontroller using Pulse Width Modulation (PWM). A PWM signal is a

periodic signal with two possible amplitudes 0V and 5V. The two parameters of PWM

that can be modified are the frequency of the signal and the duty cycle - the percentage

of each cycle where the signal is high. The manufacturer of the servo usually publishes

the optimal frequency at which to run the PWM signal. In this case the frequency was

50Hz. Altering the duty cycle alters the pulse width. This is how the motor is directed.

An illustration of a PWM signal can be seen in Figure 14.

A servo must have a PWM signal to provide torque; even holding torque. The signal

provided to a servo has a pulse width of between approximately 1ms and 2ms. One end

of the input signal range directs the servo to its minimum angular displacement, while

the other end of the range directs the servo to its maximum displacement. Every servo

has different tolerances, and it is possible for a servo to strip its internal gears if it is

directed to go beyond its minimum or maximum tolerance.

The servos that were chosen for this project were Turnigy TGY1501 MG: Hobby

38

Figure 14: Illistration of PWM signal [1]

Servos. These operating specifications for this servo are given in Table 4. The primary

reason for selecting this servo was the high torque that they can supply at a low price-

point. When all five servos were at stall, the maximum current draw was approximately

3A.

Motor Feature ValueSpeed 0.14s/60 @ 6VTorque 17kg-cm @ 6VWeight 60g

Current Draw 300mA @ Stall TorqueCost $12 each

Table 4: Turnigy TGY 1501 MG: Hobby Servos

8.7 Assembly

After the printed hand components were manufactured, they were assembled and

integrated with the other mechanical components of the hand. The first step in the

process was to assemble the printed components. 22 gauge solid copper wire was used as

39

the axel for each of the joints. A single, straight piece of wire was required for each of the

3mm holes and three twisted wires of the same gauge were used in each of the 5mm holes.

Wire was used rather than a proper fastener due to the lack of space between fingers.

Neither the head of a bolt nor the nut used to secure it would fit in the available space.

Unfortunately, there was no room in the design for countersunk bolt holes or embedded

nuts. Instead, the wire was secured inside the axel holes with high strength tape.

Figure 15: Dorsal Side of Assembled Prototype

Once the printed components of the hand were assembled, the dorsal elastics were

installed so the fingers would stay extended unless forcefully bent. The elastics were

secured to each finger segment and the palm segment with plastic zip-ties. Once this was

completed, it was time determine where to place the servos.

Many tests were performed with the servos in various locations to see how their

40

configuration effected the actuation of the fingers. It was quickly determined that at

least three of the servos would not fit inside the hand itself. To accommodate these

motors, a mounting platform was installed just below the palm segment. The mounting

platform consisted of a piece of Acrylic secured with a small block of medium density

fiber board (MDF). The MDF was secured to the palm segment by two 8cm lengths of

18 gauge steel wires. The mounting platform was further supported by a 1cm long M4

screw connecting it directly to the palm segment.

The next step was the installation of the motors. Each servo was secured with M4

screws. Two servos were installed in the body of the hand and the other three servos were

installed on the mounting platform. 50lb braded fishing line was used as the palmer cable

that would bend each of the fingers. This line was chosen due to its very high tensile

strength, its resistance to abrasion and its resistance to stretching. A 30cm length of line

was threaded through the palmer channels of each finger. Each line was secured at the

palm segment end to a motor and at the fingertip to the zip-tie holding the dorsal elastic

in place.

8.8 Testing

The prototype hand has been through many rounds of rigorous mechanical testing.

Each finger was fully actuated hundreds times, and as a result, many positive and negative

aspects of the mechanical design were discovered.

We have determined that the cable based actuation system is very effective. This

was observed in the early cardboard mockups of the hand; however, the system was

not expected to scale to a larger device as effectively as it did. The pressure imparted

through the fingers when the motors were fully engaged was considerable. The hand could

easily lift a 600g bottle of water and maintain a solid grip despite aggressive agitation.

The diverse mechanical functionality of the final product was another positive result.

The prototype met the required design goal of effectively preforming several grips with

impressive precision.

Despite the project being a resounding success, there were many design flaws that

41

will need to be addressed in future design iterations. Firstly, the single-cable-actuated

mechanical design requires the motors to work against the dorsal elastics at all times

when the fingers are curled. This requires a continuous high current draw by the motors.

This is simply unacceptable for a practical prosthetic limb. The next design iteration of

the mechanical hand must address this issue first and foremost. The antagonistic force

scheme discussed at the beginning of this chapter is a promising candidate to solve this

problem.

An additional deficiency encountered in this design is the routing of the cables from

the finger tips to the motors. The complexity of the cable routing can be seen in figure 15.

Abrasion encountered by the high strength fishing line as it rubs against the plastic of the

printed components and motors causes frequent line failures. To address this problem,

a better cable routing strategy must be developed. Strategically placed bearings or low

friction pylons will help to correct this issue, but moving away from fishing line in favor

of aircraft cable might be a better long term solution.

This solution might prevent the cables from failing, but it may exacerbate another

problem the hand is facing. Repeated actuation of the fingers by the high strength

motors has begun sawing through the printed parts. Using wire rather than synthetic

fishing line will almost certainly increase the rate at which the printed parts degrade. A

simple solution to this problem has presented itself during testing. By lining the abrasive

surfaces of the hand with stronger and less abrasive materials, the ware on both the hand

and the cables has been significantly reduced. Metal sheaths should be used to cover

contact points in the future to extend the life of printed components. Additionally, the

printed layers of the components should be oriented so that the weakest z-axis layers are

not parallel to the actuation cables. This has also played a substantial role in the quick

degradation of these parts.

Though there are many deficiencies in this first mechanical design, all of these prob-

lems can be overcome in future design iterations by applying systematic testing and

experimenting with creative solutions. All things considered, this mechanical design has

been a definite success.

42

8.9 Benefits of a 3D printed hand

There are two major advantages to producing prosthetic hands with 3D printers.

First, it is very inexpensive. The accelerating adoption of consumer grade printers and

the decrease in the price of feed stock means that replacement and upgraded parts can

now be printed in many materials for only tens of dollars. Second, 3D printing allows for

mass customization, a sought after quality in prosthetics. Using this technology, every

hand can be designed to meet a specific users needs, including the size of the hand and its

complexity. The size of each hand can inexpensively be modeled to match the proportions

of the opposite hand. Also, some people lack the fine motor skills, experience or need

for an advanced hand with many degrees of freedom and a myriad of functions. Due to

the modular nature of the hand design, these users can use a simpler hand and upgrade

individual components as they grow and their abilities or preferences change.

43

References

[1] Arduino. (2008). pwm [online]. available: http://www.ti.com/lit/ds/symlink/mf10-

n.pdf.

[2] ”national amputee center”. War Amps, March 2013.

[3] Carl R. Deckard. ”method and apparatus for producing parts by selective sintering”.

4,863,538, September 1989.

[4] Brian Heater. ”the shape of things to come: A consumer’s guide to 3d printers”.

http://www.engadget.com/2013/01/29/3d-printer-guide/.

[5] Health canada. (2006). guidance for industry- keyword index to assist

manufacturers in verifying the class of medical devices [online]. available:

http://ec.europa.eu/enterprise/newsroom/cf/ getdocument.cfm?doc id=4761.

[6] Michael R. Neuman et al. Medical Instrumentation, Application and Design. John

Wiley and Sons Inc., 111 River Street, Haboken, New Jersey, 4th edition, 2010.

[7] K. p. nearing. ecor 4995. class lecture. topic:“professional engineers: Responsibilities

of regulated profession”, faculty of engineering, carleton university, ottawa, ontario,

january 24, 2013.

[8] Roberto Merletti et al. Electromyography Physiology, Engineering, and Non-Invasive

Applications. John Wiley and Sons Inc., 111 River Street, Haboken, New Jersey,

2004.

[9] Delsys. (2002), surface electromyography: Detection and recording [online].

available: http://www.delsys.com/attachments pdf/wp semgintro.pdf.

[10] Texas intruments. (2000). low-power, 8-channel, 24-bit analog front-end for biopo-

tential measurements [online]. available: http://arduino.cc/en/tutorial/pwm.

44

Appendix A

Gantt chart showing the breakdown of projected tasks throughout the academic year:

Fall 2012

Table 5: Gantt Chart Displaying the Project Schedule for the Fall Term

45

Winter 2013

Table 6: Gantt Chart Displaying the Project Schedule for the Winter Term

46