Transcript Lecture 13 and 14 - Electrical & Computer Engineering

Lecture 13
and 14
Dimitar Stefanov
Feedback in the prosthesis control
•Visual observation
•Transmission of force and pressure through the socket
•Indirect feedback from the prosthesis to the body (force - , slip- and
pressure-transducers mounted to the terminal device plus interface
devices attached to the residual limb)
Force sensors, slip
sensors, pressure
sensors
Terminal
device
Interface
devices
bladders,
plungers or
vibrotactile
actuators
Residual limb or other
intact area of the body
Prostheses controlled by myoelectric
signal (MES)
The MES, captured on the surface of the skin, has been exploited as a
control source of powered prostheses.
•In the 1950's the first powered prosthetic limb was built.
•By 1964, Robert Scott of the University of New Brunswick developed
the first control system for prosthetic limbs.
The myoelectric signal (MES)
The myoelectric signal (MES)
•A single motor unit
•A motoneuron activates a group of muscle fibres of
the motor unit
•A linear model:
•A series of excitatory impulses from a motor
neuron innervate a group of muscle fibres.
•The excitatory impulses are separated by the
intervals x1, x2,.., xn.
•The motor unit action potential (MUAP) is the
electrical response of each impulse.
•Potentials between the electrodes, placed in
proximity to the motor unit, depends on the MUAP.
•The speed of the pulse wave – from 2 to 6 m/s, in
inverse relation to the fibre diameter (Buchtal54)
•In order to sustain a voluntary muscle contraction,
the motor units must be repeatedly activated.
•Series of MUAPs – motor train.
MES during sustained contractions
Control of a skeletal muscle is achieved by:
1. Recruitment *– varying the number and composition of
activated motor units
2. Firing rate – varying the rate of activation of the individual
motor units. *Size principle of recruitment – the smallest motor units are being
recruited first (Henneman65).
•At a contraction, which is up to 30% from the maximal
contraction, the recruitment process in dominated.
•At a contraction, which is between 30 and 75% from the maximal
contraction, the firing rate in dominated in the force production.
•At a contraction, which is over 75% from the maximal contraction,
the force increasing is a result solely of the the firing rate
increasing.
Variants of prosthetic devices:
1. One MES** and one single device*
2. One MES and n single devices (at each moment
only one single device of the prostheses is
controlled)
3. n MES and n single devices***
4. n MES and m single devices (m>n) . At each
moment p single devices are operated. (
)
*single device – motor or lock belonging to the hand, elbow, or
wrist of the prosthesis
** In case of lock, only one channel of MES is required. In case of
motor control, two channels of MES are required.
*** At each moment, one or more single devices can be operated.
Control information from the MES, which relates to variant 2:
•Selection of the device, which will be controlled from the MES
(choice of a motor or a lock)
•Information about the speed of the selected single device.
Example: The length of the signals from one and the same MES
source can determine the device and its level of activation. Short
MES for selection of the single device and long lasting MES for
activation of the same device. The amplitude of the long signal
determines the level of activation of the device.
Simple algorithms of prosthetic control:
1. Control signal which relates to the estimated amplitude of the
MES (Dorcas, 1966)
2. Control signal which relates to the rate of change of the MES
(Childrese, 1969)
Speed control of the motor of the prosthesis:
1. Analogue control (speed, proportional to the amplitude of the control signal)
2. Binary control (The motor is either not activated or moves on a constant speed
when the amplitude of the control signal exceeds certain threshold).
The relationship between a user’s input signals and the movement of
the prosthesis’ motorized components is called a control strategy.
Analogue control of a simple device
Electrodes on agonist/antagonist muscle
The result signal determines the direction and the force in the controlled joint.
In the most commonly used myoelectric control strategy a flexor muscle signal is used to
close a hand and an extensor muscle signal is used to open the hand. We refer to this as
classic myoelectric control.
Block diagram of analogue proportional control
Nonlinear integrator
Example:
+Vcc
v
v
t
T1
T2
R
output
Zener
-diode
C
t
Binary control
One-channel control
of a terminal device
•In the past, the control strategy of a powered prosthesis
was a built-in feature of the electronic circuit that was sold
with the prosthesis.
•Changing the control strategy meant changing the
electronics.
Programmable myoelectric control
The electronics can be easily programmed to implement the selected
strategy.
MyoMicro TM
Software product of Variety Ability Systems Inc., Ontario, Canada
http://www.vasi.on.ca/prosthetic/myomicro/myomicro.html
Software product for easily programming of the electronic control
systems from Variety Ability Systems Inc. or Liberty Technology.
SPM from
VASI Inc.
MyoMicro TM
VARIGRIP modules from Liberty
Technology.
MyoMicro system
Hardware module + software package
•The MyoMicro hardware module
is connected both to the
computer’s parallel port and
computer’s serial port.
•The programmable prosthesis
controller is connected also to the
same hardware module.
The MyoMicro software allows selection of a control strategy and programming
the circuit board with the selected control strategy using "strategy wizard" .
Flexible programming - One strategy can be initially used and later the control
strategy can be changed.
In MyoMicro, control strategies are described graphically as a
collection of primitive building blocks.
Three basic types of building blocks:
1. Input devices - correspond to electrodes, touchpads, and
switches
2. Signal processors - represent the required modification of the
input signals (comparator, differencer, etc.).
3. Motor Controls – power amplifiers which operate the physical
effectors (hand, elbow or wrist).
Supported control strategies:
1. Proportional Control strategy
2. Digital Control strategy
3. "doing more with less“ strategy.
Some "doing more with less“ control strategies:
A./ Level-sensitive control
Input Signal
Level Sensitive System
Open
Digital control only
Close
time
delay
Time
•The input signal is compared to two preset thresholds.
•Typically, if the input signal is in the middle region the hand will
close.
•If the input signal is in the high region, i.e., above the upper
threshold, the hand will open.
•When the input signal is in the middle region the output is slightly
delayed to avoid inadvertent hand closing while the signal is in
transition between the high and low regions.
B./ Rate-sensitive control
Input Signal
Rate Sensitive System
Proportional or
digital control
Open
Close
time
delay
time
delay
Time
•When the input signal crosses the lower threshold, the system waits
for a preset period of time (tens of milliseconds) and then compares
the signal to the upper threshold.
•If the signal is still below the upper threshold after the preset time
delay, it is considered to be a slowly rising signal and the hand closes.
•If the signal exceeds the upper threshold before the end of the delay, it
is considered to be a quickly rising signal and the hand opens.
C./ Mode switching control
Control a hand and a wrist with two input signals only.
•At first the control signals operate the hand (flexors to
close, extensors to open).
•Cocontraction of the flexors and extensors switches
control to the powered wrist. A flexor signal will then
supinate the wrist and an extensor signal will pronate
the wrist.
•To switch control back to the hand, the user can
cocontract again.
•If no signal is received for a preset period of time,
control automatically reverts to the hand.
•The prosthesis controller is adjusted and preprogrammed via the computer.
•MyoMicro possesses option for monitoring of the user’s input signals from
electrodes during the adjustment.
•User can try a variety of different strategies and the best suitable strategy can be
found easily.
Menu examples:
Other attempts to increase the number of states
from the surface MES
•Usage of many sites (or channels) of amplitude coded information
(Schmiedl77, Wirta78, Almstrom81). A vector of features is subject
of some pattern recognition. Disadvantage: Requirement of many
electrode sites, problems in locating and maintenance of the integrity
of patterns.
•Usage of time-series model (Graupe82, Doershuk83).
Disadvantage: sensitivity to the signal amplitude.
All methods for incensement the number of signals,
commented until now, use steady-state analysis to the
MES (signals that are produced during constant muscle
effort).
Recent tendency – reduction of the number of electrodes and reduction
of the sensitivity to the electrode placement.
Multifunction Myoelectric Control Systems
(Control of Artificial Limbs using
Myoelectric Pattern Recognition)
Electrodes placed over a set of muscles
Multiple control signals
Control system's ability to control more than one device,
such as an elbow and a wrist by one MES.
Control of prostheses for high-level amputations
Structure of the transient MES
MES which coincide with the onset of rapid contractions
Hudgins (1991) placed a single pair of surface electrodes over the biceps
and triceps and investigated the signals between the electrodes during
small isometric and anisometric contractions.
Patterns of transient MES, corresponding to flexion/extension
of the elbow and pronation/supination of the forearm:
Elbow flexion
Forearm pronation
Elbow extension
Forearm supination
Basmajian (1985) The motor unit recruitment order appears stable for a
given task, once the task has been learned.
Two channel transient MES patterns
•Kuruganti (1995) The performance of pattern recognition Hudgins’
myoelectric scheme can be enhanced by using two channels (localized
MES) instead of one channel (global MES).
•The localized activity of the biceps and the triceps can be better
analyzed by two sets of closely spaced bipolar electrode pairs.
Patterns of transient MES using two sets of
electrode pairs:
Elbow flexion
Elbow extension
Biceps
Biceps
Triceps
Triceps
Forearm pronation
Forearm supination
Biceps
Biceps
Triceps
Triceps
Localization of the signals provides a greater distinction between the
classes than one channel.
Hudgins (1991, 1993) investigated the information contents of the transient burst of
myoelectric activity accompanying the onset of sudden muscular effort.
The Hudgins’ multifunction control scheme for powered upper-limb
prosthesis
•Time-domain features of the MES are used (zerocrossing, mean absolute value, mean absolute value
slope, and trace length)
•Simple multilayer perception (MLP) ANN is used
as a classifier.
•The controller identified four types of muscle
contractions using signals measured from the biceps
and triceps.
•Multifunction control from a single site
•Output control signals derived from natural muscle
contractions
Results from the experiments:
Average classification performance – 89% for a group of
15 subjects (9 normally-limbed subjects and 9 persons
with limb deficits).
Two ways for further improvement of the
classification accuracy:
1. Improvement the classifier (type and
characteristics)
2. Improvement the means of the signal
representation (improvement of the feature set).
1. Although some classifiers possess obviously better
characteristics, the effect of the improved classifier to
the recognition characteristics is not so significant.
2. The signal representation most significantly affects the
classification performance (Bishop 96).
The problem of the signal representation
for pattern classification can be broken
down into following 3 tasks:
1. Feature extraction
Procedure of
2. Dimensionality reduction
signal representation
3. Classification.
}
•The transient MES patterns have their structure both in
the time and the frequency domain.
•It is supposed that the better recognition methods should
be based on time-frequency representation (timefrequency based feature set).
Feature sets based upon the short-time Fourier
transformations, the wavelet transform
Dimensionality reduction – very important
Relating Surface EMG Amplitude to Joint Torque
Ongoing project at the Worchester polytechnic Institute, Dept. of EECS,
http://www.ece.wpi.edu/~ted/research.htm
Relation between the surface EMG amplitude to the tension (or force)
produced by individual muscles.
non-linear relationship
Difficulties:
1.
EMG from muscles other than that which the experimenter intends
to record may be included in the signal ("cross-talk“).
2.
Relating EMG to individual muscle tension requires independent
verification via direct mechanical measurement of individual
muscle tension. At present, there is no practical method for reliably
making such in-vivo measurements.
Easier task is finding relations between the surface EMG amplitude and
the joint torque.
EMG-torque relationship can be modeled as a polynomial relationship in
case of constant-posture, nonfatiguing contractions about the elbow.
Example of cable driven, body powered prostheses.