estimation of joint torque and impedance by means of surface emg edward (ted) a. clancy department...

Estimation of Joint Torque and Impedance by Means of Surface EMG

Edward (Ted) A. ClancyDepartment of Electrical and Computer Engineering

Department of Biomedical EngineeringWorcester Polytechnic Institute, USA

Denis RancourtDepartment of Mechanical Engineering

Sherbrooke University, Canada

Supported by USA National Institute for Occupational Safety and Health grant R03-OH007829.

EAC07–130

Copyright Edward A. Clancy and Denis Rancourt, 2007. Some rights reserved. Content in this presentation is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 License. This license is more fully described at: http://creativecommons.org/licenses/by-nc-sa/3.0/.

http://creativecommons.org/licenses/by-nc-sa/3.0/

http://creativecommons.org/licenses/by-nc-sa/3.0/

Research Aims

EAC07–131

Introduction

Zelbow GZ

Raw EMG GT

Advanced EMG

processors

Elbow torque

Non-invasive advancedadvanced, EMG amplitude & torque-impedance

estimator

(Optimize each block)

Research Applications

• Myoelectric control of prosthesis

• EMG biofeedback for rehabilitation

• Ergonomic analysis / task analysis

• Biomechanical modeling

• Measurement in motion control studies

EAC07–132

Introduction

Presentation Overview1. EMG amplitude (EMGamp) estimation:

2. EMG-torque estimation:

3. EMG-impedance estimation:

ˆ s t

EAC07–133

Introduction

EMGampEstimator

tsAgˆEMGamp-

Torqueˆ T

EMGamp-Impedance Jbk ,,

tmAg

EMG Waveforms

Jbk ,,

ˆ T

EMGampEstimator

tmAn tsAnˆ

– Preliminary research

Electromyogram (EMG) Signal Model

“Interference pattern” EMG:Superimposition of multiple MUAPs plus measurement

additive noise.

EMG Signal Origin [Basmajian and De Luca, 1985]

EMGamp

EAC07–134

MUAP

EMG Amplitude (EMGamp) Estimation

EAC07–135

EMGamp

EMGamp

EMG amplitude 0–10 mV (~1.5 mV rms)Surface EMG frequency bandlimited at 2000 Hz

• EMG Amplitude (EMGamp): “Intensity” of recorded EMG– “Time-varying standard deviation of EMG signal”

• Original estimator: Inman et al. [1956]– Analog full-wave rectify and RC low pass filter

Electrode-Amplifier

(From Liberating Technologies)

– Varies with force

Improving EMGamp Estimation

• EMGamp improved by:– Removal of measurement noise– EMG signal whitening

• Adaptive whiteningTo reject measurement noise

– Multiple EMG channels (for large muscles)– Optimal detectors– Optimal smoothing (bias vs. variance error)

EAC07–136

EMGamp

Increases statistical bandwidth of EMG Reduces variance of amplitude estimate.

{

Single Channel, Including Noise

H etimej

w i ni ir

si

mi

v i

Zero Mean, WSS,CE, White Process of Unit Intensity

Shaping Filter

Zero Mean,WSS, CE,Additive NoiseProcess

Measured SurfaceEMG

EMG Amplitude

iaCable andmotion artifact

ii vr

Prakash et al., IEEE Trans Biomed Eng 52: 331–334, 2005.

EAC07–137

EMGamp

Phenomenological ModelEMGamp standard deviation of mi

Optimal EMG Processor — Single Site

Clancy and Hogan, IEEE Trans Biomed Eng 41: 159–167, 1994.

mi Htime1 je 1

N i k

2ˆ v k0

N 1

12

ˆ s i ˆ v i

Whitening Filter Moving Average RMS

Note: Does not account for noise/interference.

EAC07–138

EMGamp

EMGWaveform

EMGamp

Whitening Example


EAC07–139

EMGamp

Single-Channel EMGamp Processor

...

ksRelinearize

(·)1/dSmoothmk Detect

|·|dWhitenNoise Reject/ Filter

......

1 32

4 65

1 2 3 4 5 6

Clancy, Farina, Fillogoi, 2004

EMGamp

EAC07–140

Multiple-Channel, Including Noise

EAC07–141

EMGamp

Clancy Ph.D., 1991.

Independent, Zero Mean, JWSS, JCE, White Processes of Unit Intensity

Spatial and Temporal

Shaping Filters

MeasuredSurfaceEMG

Independent, Zero Mean, JWSS, JCE, Additive Noise Processes

wL ,i

w3,i

w2,i

w1, i

xL ,i

x3,i

x2,i

x1, i n1, i

n2,i

n3,i

nL ,ir L ,i

r 3,i

r 2,i

r1,i

v1,i

v2, i

v3, i

vL, i

mL, i

m3, i

m2, i

m1,i

si

si

si

si

Hspace

Htime ,1 ej

Htime ,2 ej

Htime ,3 ej

Htime ,L ej

Phenomenological Model

Six-Stage EMG Amplitude Estimator

EAC07–142

EMGamp

Clancy et al., J Electromyo Kinesiol 12:1–16, 2002.

EMGamp

Experimental Apparatus

EAC07–143

EMGamp

Clancy, IEEE Trans Biomed Eng 46:711–729, 1999.

EMG Electrode Sites

EAC07–144

EMGamp

EMG Flexion 3

EMG Flexion 2

EMG Flexion 1

EMG Flexion 0

EMG Extension 3

EMG Extension 2

EMG Extension 0

EMG Extension 1

Experiment Brief Description– Subjects seated in exercise machine

– Active bipolar electrode-amplifiers applied to both biceps and triceps

Subjects performed various static or dynamic contractions, tracking a target.

EMGamp

EAC07–145

Single-Channel Whitening


EAC07–146

EMGamp

• Constant-posture, constant-force contraction

s

sSNRˆ

ˆ

Multiple-Channel Whitening

EAC07–147

EMGamp



Multiple-Channel Whitening — Average Results


EAC07–148

EMGamp


SNR vs. Window Length

EAC07–149

EMGamp

• Theory:

TLBSNR s 22

Window Length (T), ms

St-Amant et al., IEEE Trans Biomed Eng 45: 795–799, 1998.

SN

R


Adaptive Whitening Problem

EAC07–150

EMGamp

EMGamp

EMG signal

Clancy Ph.D., 1991.

RawAdaptive

WhiteningWhitened

Power Spectrum of Amplitude-Modulated EMG

EAC07–151

EMGamp

Clancy Ph.D., 1991.Frequency (Hz)

Lo

g P

ow

er S

pec

tru

m

Adaptive Whitening Solution

EAC07–152

Adaptive Whitening Filter: HA ej ,si Htime 1 e j HW e j , si d si

yi

Fixed Whitening

Filter: H e

s

S e s S e

timejw

Cal

mmjw

Cal vvjw

1

,

~ ~ ~m s n vi i i i

Adaptive Wiener

Filter: WH e s

s

s S e

ji

i

i v vj

,

~~

2

2

d si

m s n vi i i i

StageStage 2 StageStage 3StageStage 1

si

EMGamp

Sample Adaptive Whitening Filters

EAC07–153

Mean value (solid line) plus one standard

deviation (dash line)

EMGamp

EMGamp: Myoelectric Control of Prosthesis

EAC07–154

Prosthetics

• Use remnant muscle EMG to command electric hand, wrist, elbow– Some lower limb prosthetics research

Major current effort by U.S. military to improve prosthetic limbs

•Improved myoelectric controls

•Embedded neural/myo sensors•To give more control•To provide feedback

•Directly connect limb to boneBoston Elbow

Opportunities in Myocontrol

• Newer prosthetic limbs — circa 2002– Microprocessor– Digital control– Simultaneous control of multiple motors

• Advanced EMG processing now feasible– Not feasible previously in production arms

EAC07–155

Prosthetics

EMGamp: Gait Biofeedback

EAC06–xxx

Gait

EAC07–156

Re-training after stroke, traumatic

brain injury.

With Paolo Bonato, Spaulding Rehabilitation Hospital, Boston

EMGamp: EMG-Torque Uses

EAC07–157

EMG-torque

• Non-invasive torque measurement for scientific studies

• Study/evaluation of worker safety– Repetitive, high-force tasks can lead to

cumulative trauma injuries

Surface EMG Signal and Joint Torque

EAC07–158

EMG-torque

Time in Seconds

Time in Seconds

Joint Torque

Surface EMG

Clancy Ph.D., 1991.

Simple Elbow Mechanics Model

EAC07–159

EMG-torque

Clancy Ph.D., 1991

TNET = TF+TE

(www.mrothery.co.uk)

EMG-Torque Block Diagram

ExtensorEMG

AmplitudeEstimator

ExtensorEMG

Amplitude toTorque

Estimator

mE1,i

mE2,i

mELE,i

FlexorEMG

AmplitudeEstimator

FlexorEMG

Amplitude toTorque

Estimator

mF1,i

mF2,i

mFLF,i

ŝE

ŝF

E

Ext

F

_

+

……

• OptimallyOptimally relate biceps, triceps EMGamp to elbow torque – Calibrate for each subject

• Compare conventional vs. advanced EMGamp processors

EAC07–160

EMG-torque

• Slowly force-varying No dynamics

• Constant Posture

• Polynomial model

EMGamp-Torque: Quasi-Static Contraction


Tor

qu

e in

A/D

Un

its

Tor

qu

e in

A/D

Un

its

Time in Seconds

Time in Seconds

Unwhitened, Single-Channel Processor

Whitened, Four-Channel Processor

EAC07–161

EMG-torque

EMG-Torque: Constant-Posture, Force-Varying

5 10 15 20 25-400

-300

-200

-100

0

100

200

300

400Extension (Triceps M/C)

Time (sec)

Ra

w E

MG

in

%M

VE

E

5 10 15 20 25-400

-300

-200

-100

0

100

200

300

400Flexion (Biceps L/C)

Time (sec)

Ra

w E

MG

in

%M

VE

F

5 10 15 20 25-50

-40

-30

-20

-10

0

10

20

30

40

50Real Torque and Predicted Torque Normalized in %MVC

Time (sec)

To

rqu

e A

mp

litu

de

%M

VC Positive Torque - Flexion

Negative Torque - Extension

Real Torque: MeasuredPredicted Torque, Unwhitened EMGPredicted Torque, Whitened EMG

Clancy et al., J Biomechanics, 2006.

EAC07–162

EMG-torque

EMG-Torque: Force-Varying, Results0 5 10 15 20 25 30

20

30

40

50

60

70

80

90 Median of %VAF, Fast Tracking Speed

System Identification Order, nb

%V

AF

0 5 10 15 20 25 305

10

15Median of MAE, Fast Tracking Speed


MA

E (

Pe

rce

nt)

0 5 10 15 20 25 3020

30

40

50

60

70

80

90 Mean of %VAF, Fast Tracking Speed


%V

AF

0 5 10 15 20 25 305

10

15Mean of MAE, Fast Tracking Speed


MA

E (

Pe

rce

nt)

S-CH-UNWHITS-CH-WHITM-CH-UNWHITM-CH-WHIT




55%

80%

10%

7.3%

Clancy et al., J Biomechanics, 2006.

EAC07–163

EMG-torque

• Linear (moving average) model

EMG-Torque Summary

• Future directions– Applications in ergonomics

• Constant-posture tasks– Relax postural constraints– Multi-joint systems

• Goal: Calibrate EMG-torque in apparatus; then estimate torque in unconstrained tasks

Better EMGamp Better EMGamp Better Torque Better Torque

EAC07–164

EMG-torque

EMG-Impedance•Rigorous representation of muscular co-activation

•Want optimized EMG-impedance relationship

– Constant-posture, quasi-static conditions– Preliminary work

•Goal: Calibrate EMG-impedance in apparatus; then estimate impedance in unconstrained tasks

EAC07–165

EMG-impedance

EMG-Impedance Block Diagram

EMGAmplitude toImpedanceEstimator

K (stiffness)

ExtensorEMG

AmplitudeEstimator

mE1,i

mE2,i

mELE,i

ŝE…

FlexorEMG

AmplitudeEstimator

mF1,i

mF2,i

mFLF,i

ŝF…

B (viscosity)

I (inertia)

Higher-quality EMG amplitude estimates should give better impedance estimation.

Assume second-order, linear model; constant operating point.

EAC07–166

EMG-impedance

Preliminary Work ONLY !!!

Elbow Mechanical Impedance Measurement

Subject seated, shoulder 90o abducted, elbow 90o flexed.

Right hand immobilized in cuff (2), connected to actuated joystick (1)

Medio-lateral pseudo-random FORCE perturbations (3)

Measure medio-lateral movement through joystick encoders

Assume second order linear system: K, B, I

Estimate I separately

K, B vary with operating pointK, B vary with operating pointTop View

EAC07–167

EMG-impedance

S. Martel, M.S. Thesis, U. Sherbrooke, 2007.

Impedance Calibration: Ramp Contraction Profile

• Slow ramp from extension-dominant to flexion-dominant

EAC07–168

EMG-impedance

Time (s)

Per

turb

atio

n F

orc

e

S. Martel, M.S. Thesis, U. Sherbrooke, Fig. 3.15, 2007.

• Also test at fixed operating points

Impedance Models• Fixed operating point (“constant torque”), :

• T: Torque perturbation, : Angular perturbation• K: Stiffness, B: Viscosity, I: Inertia• Also measure EMG

• Slowly varied operating point :

• Polynomial basis gives:

– Fit parameters: k0, kE,1, kF,1, b0, bE,1, bF,1

EAC07–169

EMG-impedance

iiii IBKT

FE ss ˆ,ˆ iiFEiFEi Is,sBs,sKT

iiF,FE,EiF,FE,Ei IsbsbbskskkT 110110

EMGamplitudes

System ID Protocol

EAC07–170

EMG-impedance

0–15 Hz, pseudo random torque

EMG sampled at 4096 Hz

Force, displacement sampled at 400 Hz, 1.5–8 Hz band pass

Estimation of K and B

1. 10%, 20%, 30%, 40% MVC flex

2. 10%, 20%, 30%, 40% MVC ext

Estimation of K and B versus time for slowly varying ramps (40% extension to 40% flexion MVC)

16 subjects

20 seconds

20 seconds

20 seconds

Target torque, perturbation superimposed

Time (s)

Time (s)

Time (s)Time (s)

Normalized

For constant-torque flexion tests

Constant Torque Data: EMG Amplitude

FE ss ˆ,ˆ

Es

Fs

EAC07–171

EMG-impedance

Normalized

For constant-torque extension tests

EF ss ˆ,ˆEF ss ˆ,ˆ

Constant Torque Model Test: Torque Prediction

% VAF = 89% on estimated torqueTime (sec)

Actual torque vs Predicted torque

To

rqu

e (N

m)

EAC07–172

EMG-impedance

But, stiffness, viscosity parameters NOT NOT PHYSIOLOGICPHYSIOLOGIC

… We’re still working!!??!!

Ramp Torque Data: EMG Amplitude

Fs

EAC07–173

EMG-impedance

Es

Time (s)

Ramp Torque Issue at 0% MVC

• Force perturbations– At 0% MVC Position drifts

• Solutions:1. Do not pause at 0% MVC

2. Position perturbations (Requires stronger motor)

EAC07–174

EMG-impedance

Time (s)

Impedance Study Status• 0–15 Hz perturbation band sufficient• Post-filtering (T, ) 1.5–8 Hz essential to

remove low freq modes• Better torque prediction results obtained when

remove inertial forces prior to ID

• First order model for EMG-impedance or H.O.?• May use relaxed test to estimate k0, b0 instead?• Ramp calibration may require position control?• Will improved EMGamp estimators help?

EAC07–175

EMG-impedance

Questions?

EAC07–176

Zelbow GZ

Raw EMG GT

Advanced EMG

processors

Elbow torque

Non-invasive advancedadvanced, EMG amplitude & torque-impedance

estimator

(Optimize each block)

estimation of joint torque and impedance by means of surface emg edward (ted) a. clancy department...

Documents