Professional Documents
Culture Documents
doi: 10.14355/rame.2015.04.001
www.seipub.org/rame
StudyonMuscleFatigueMeasurement
DuringElbowFlexionExtensionUsingEMG
andOpenSim
DavidSangKunChun*1,YounJooLee2,SeongWookCho3,InchanYoun4
BiomedicalResearchInstitute,KoreaInstituteofScienceandTechnology,Seongbukgu,Seoul,Republicof
Korea
*1,2,4
*1,3
*1
SchoolofMechanicalEngineering,ChungAngUniversity,Dongjakgu,Seoul,RepublicofKorea
t13215@naver.com;2younjoo@kist.re.kr;3scho@cau.ac.kr;4iyoun@kist.re.kr
Abstract
Thispaperproposesamethodofsurfaceelectromyography(sEMG)basedmensurationofmusclefatigueduringelbowflexion
extension.Theproposedmethodhastwoparts.First,thejointangleandlocationsofbodysegmentsareestimatedusinginertial
measurementunits(IMU)andtheOpenSimbiomechanicssimulator.Second,musclefatigueismeasuredusingOpenSimanda
threecompartmentmusclefatiguemodel.Theexperimentalresultsshowthatmusclefatigueincreasesinaccordancewiththe
momentum.
Keywords
MuscleFatigue;EMG;IMU,Biomechanics;Opensim,MotionCapture;ElbowFlexionExtension
Introduction
Commonly used surface electromyography (sEMG)based methods for measuring muscle fatigue use the mean
power frequency or median frequency of the electromyogram (EMG) recorded while a subject performs static
contraction.However,usingthesemethods,itisdifficulttoaccuratelymeasuremusclefatigueindailyactivities,
exercise,andlaboractivitiesthatactuallyrepresentasubjectsmotionthatmaycauseoraggravatemusculoskeletal
diseases.Tosolvethisproblem,somepreviousstudieshaveadditionallyusedtheapostprocessingstepbasedon
patternrecognitiontechniques(RogersandMacIsaac2011,2010).
In this paper, we propose a muscle fatigue measurement method based on kinematic information. This method
connects muscle fatigue to motion, and can obtain biomechanics values, such as the muscle force and muscle
activation,andmomentfromtheusersmovements.Tomeasurethemotionfromwhichkinematicinformationis
derived, markerbased motion capture (MbMoCap) devices have been commonly used. However these are very
expensive,difficulttosetup,andhavespaceconstraints.Toovercometheselimitations,weproposeamethodto
use inertial measurement units (IMU). These are inexpensive, easy to use, and can be used outdoors (Daniel
Roetenberg2013).
The OpenSim biomechanics simulator (OpenSim, Stanford Univ., USA) and threecompartment muscle fatigue
model(XiaandFreyLaw2008)wereusedtocalculatemusclefatigue.BecauseinputformorextensionofOpenSim
isonlyMbMoCapbased,markerpositionswereestimatedusingtheIMUandinputtotheinversekinematicsin
OpenSimtocalculatethejointangle.ThisjointanglewasinputtothestaticoptimizationinOpenSimtocalculate
muscle activation. Then, the coefficients of the EMGmuscle activation transformation were calculated using the
muscleactivationandEMGtotransformtheafterwardmeasuredEMGintomuscleactivation.Finally,theEMG
basedmuscleactivationwasinputtothemusclefatiguemodeltocalculatemusclefatiguewhichisdimensionless.
ThisflowchartisshowninFig.1.
Mensurationofmusclefatigueduringelbowflexionextension(EFE),whichhas1degreeoffreedom(DOF),was
developed.
www.seipub.org/rame
FIG.1OVERALLPROCESSOFTHEPROPOSEDMUSCLEFATIGUEMEASUREMENTMETHOD.
FIG.2LOCATIONOFTHEMARKERSATTACHEDTOTHEUPPERARMANDTORSO.
MbmocapDataCalibration
The body segments of the model were scaled based on MbMoCap. In addition, the coordinates of the measured
datashouldbecalibrated.First,theMbMoCapdatawascalibrated.Thecalibrationequationisasfollows:
default
where
default
pi
default
exp
T * exp pi
(1)
pi is the ith marker position in OpenSim, exp pi is the ith measured marker position, and default
T is the
exp
transformation matrix that calibrates the coordinates from the experimental coordinates to the coordinates of
www.seipub.org/rame
OpenSim,calculatedbytheleastsquaresmethod(LSM)(Sorkine2009).
T argmin
default
exp
wi
T * exp pi
default
exp
default
pi
(2)
i=marker
wherewiisweight,whichisgreaterthantheothersifthecorrespondingmarkerisonabonyplaceorplacedatthe
centerofthebody.ThemarkerpositionsinOpenSimwerecalculatedusingthegeometryofthescaledmodel.The
transformationmatrixwascalculatedusingtheMbMoCapdatameasuredunderastaticposesimilartothedefault
in OpenSim (default pose). Finally, all measured marker positions were premultiplied by the transformation
matrix.
IMUDataCalibration
TheIMUdatawerecalibratedusingMbMoCap.IfIMUbasedmotioncaptureisaccurateasMbMoCap,inother
words,therotationaldifferencesbetweenMbMoCapandtheIMUareconstantthecalibratoris,
where
calibrator
calibrator
qj
MbMoCap
q j * IMU q j 1 (3)
IMU
q j 1 is the inverse quaternion of the jth joint measured with IMU. The rotational
differences between MbMoCap, IMU,and calibratorareshownin Fig.3. First,the rotational differencefrom the
defaultposetoMbMoCapwascalculatedusingtheLSMforeachbodysegment.
default R j argmin
exp
exp
default
R j * default pi exp pi
(4)
i=marker
exp
where default R j is the rotation matrix of the jth joint from the default pose to MbMoCap data,
default pose marker position in the local coordinate, and
coordinate.
exp
default
MbMoCap
exp
default
pi is the ith
calibratorforeachjointwaspremultipliedforeachofIMUdata.
calibrated
q j t
calibrator
q j * IMU q j t (5)
Themarkerpositionswereestimatedasfollows:
IMU
pi t
calibrated
FIG.3ROTATIONOFJOINTANGLEFOREACHMBMOCAP,IMU,ANDCALIBRATOR.
where Tj(t) is the translation vector ofthe jth joint in time t,and the vector wasadded to each of corresponding
markerstotunethechildjointtotheparentjoint.Finally,a*.trcfiletheinputofOpenSimwascreatedusingthese
www.seipub.org/rame
markerpositionstocalculatethejointangleusingOpenSim.
Muscle Fatigue Measurement
MusclefatiguewascalculatedusingthejointanglesandEMG.Muscleactivationwascalculatedbyinputtingthe
joint angle into static optimization in OpenSim. Using this muscle activation and EMGs, the coefficients of the
EMGmuscleactivationtransformationwerecalculated.Then,theafterwardmeasuredEMGwastransformedinto
muscleactivation.TheEMGbasedmuscleactivationwasinputintothreecompartmentmusclefatiguemodel(Xia
andFreyLaw2008),andmusclefatigueiscalculatedlast.
EMGMuscleActivationTransformation
The EMGmuscle activation transformation was derived using EMGneural activation and neural activation
muscleactivationtransformation.ThetransferfunctionoftheEMGneuralactivationisasfollows(Buchananetal.
2004,DavidG.Lloyd2002,Manaletal.2002):
u t
e (t )
z d
1 1 z 2 z
1
1 1 z 2 z 2
1
(7)
whereu(t)isneuralactivation;e(t)isEMG;and,1,and2arecoefficientsconditionedasfollows:
1 2 11 1 2 2 1 * 2 1 1,
2 1 (8)
Wheredisdelaytimeandisequalto0.04.Thesimplifiedtransformationofneuralmuscleactivationisasfollows
(Buchananetal.2004,DavidG.Lloyd2002,Manaletal.2002):
a (t )
A*u t
e 1
A
(9)
wherea(t)ismuscleactivationandAisanonlinearshapefactor.Thecoefficients1,2,andAdifferfromtheother
muscles and are valid until the sEMG sensor is removed (Criswell 2011). The coefficients were determined by
minimizingthegapbetweentheEMGandmotionbasedmuscleactivationwithsimulatedannealing(Sartorietal.
2012).
1 , 2 , A argmin
m muscle
EMG
am t
motion
am t (10)
Theinitialvaluesofthecoefficientsare1=0.5,2=0.5,andA=0.1(Buchananetal.2004).Aftercalculatingthe
coefficients, the afterward measured EMG was transformed into muscle activation without the motion based
information.
MuscleFatigueCalculation
Musclefatiguewascalculatedusingathreecompartmentfatiguemodel(XiaandFreyLaw2008).
RC t M A M R 1 M F (11)
where RC is residual capacity; MA is the activated compartment; MR is the resting compartment; and MF is the
fatigued compartment, which denotes muscle fatigue. The values of the compartments ranged from 0 to 1. As a
muscle activates, the fatigued compartment increases, and the others decrease. Fig. 4 shows the muscle fatigue
model.MA,MR,andMFarecalculatedusingthefollowingequation:
dM R
dt
C ( t ) R * M F t
dM A
dt
C t F * M A t
dM F
dt
F * M A t R * M F t (12)
where C(t) is the activationdeactivation driving controller, F is the fatigue coefficient, and R is the recovery
coefficient.C(t)isthesameasshowninthefollowingequation(MiguelT.Silva2011):
LR * TL t M A
C (t ) LD * TL t M A
LD * M R
www.seipub.org/rame
TL t M
A
M TL t M M (13)
M M TL t
A
whereLDistheforcedevelopmentfactor;LRistherelaxationfactor;andTListhetargetloadwhichis(XiaandFrey
Law2008):
TL BE
(14)
BE TL
RC * BE
TL
whereBEisthevoluntarybraineffort.As M F increases,RCdecreases,andBEincreases.Duetosubstitutinga(t)
intotheTL,theresultofthesimulationwasthesameasthatofXias(XiaandFreyLaw2008).Therefore,C(t)can
berewrittenas,
LR * a t M A
C (t ) LD * a t M A
LD * M R
a t M
A
M A a t M A M R (15)
M R a t
Itispostulatedthatallmusclefibersareslowmusclefibers,soFis0.01,Ris0.002,andLRandLDare10(Pereira
2009).Also,MRis1beforethemuscleactivates.
FIG.4THREECOMPARTMENTFATIGUEMODEL.
Experimental Results
DatabaseandSensorDevices
A nondisabled male subject (27 years old, 181 cm, weighing 72 kg) was tested. Markers, IMUs, and the sEMG
sensor were attached to the subjects body. Quaternion data were received from two wireless IMUs (50 Hz)
(EBIMU24GV2,e2box,SouthKorea).Akalmanfilterbasedembeddedalgorithmcalculatedthequaternion,andthe
erroroftheIMUwasknowntobewithin1indailylifewristmotions(I.W.Jung2014).Becauseonlythestartand
end times could be recorded, it was assumed that the data were recorded equidistantly. Lost data were
interpolated with spherical linear interpolation (Slerp) (Eberly 2010). Retroreflective markers placed on the
subjects body were recorded (60 Hz) using eight motion capture system cameras (Motion Analysis, Motion
Analysis,USA).ThetimedataoftheIMUsandMbMoCapweresynchronizedusingtheLSMaftermeasurement.
AdryandbipolartypesEMGsensorwasused(2000Hz)(DelsysTrigno,Delsys,USA).Thereceivedsignalwas
20450 Hz bandpass filtered to remove baseline noise and artifact noise, full wave rectified, normalized by
maximumvoluntarycontraction(MVC),and6Hzlowpassfiltered(Sartorietal.2012).
ComparativeExperimentsforJointAngleEstimation
The accuracy of the joint angle using IMUs was tested by comparing the results with the MbMoCapbased joint
angle.EFEwasperformed10timesunder5differentfrequencies(0.2,0.25,0.33,0.5,and1Hz[1,2,3,4,and5sfor
eachcycle]).Theresultsweremeasuredtostudythecorrelationbetweentheerrorandangularvelocity.IMUswere
www.seipub.org/rame
attachedtothepronatortuberosityandthenearbyHUM,andthemarkersetdescribedpreviouslywasattached.
ThecorrespondingmarkersandIMUsforeachsegmentareprovidedinTable.1.
TABLE.1LOCATIONOFMARKERSANDNUMBEROFIMUSATTACHEDTOEACHBODYSEGMENT
segment
IMU
marker
thorax
none
CLAV,STRN,C7,T10,SHO
humerus
#1
HUM,LAT
forearm
#2
OLC,ULN,RAD
At the calibration stage, the default pose was measured. After the calibration stage, full range of motion (ROM)
(0~140)EFEwasmeasured.Wholestepsweremeasured10times.
OneofthetheexamplesoftheIMUandmarkerbasedjointangleunder1Hzfrequencyisthesameasshownin
Fig.5.ThesolidbluelineistheIMUbasedjointangle,andtheblackdottedlineisthemarkerbasedjointangle.
TheerrorsincreaselargelyafterthesubjectstartstomoveevenelbowjointangleexceededtheROMsometimes.
Errorswerenotrecoveredafterthesubjectfinishedmoving.Becausethemethodusedinthisstudyisnotusedto
improveaccuracybutinsteadconvertsIMUdatatoMbMoCapdata,thismightbecausedbytheinnersideofthe
IMUlikehardwareandembeddedalgorithm.
FIG.5COMPARISONOFEFEJOINTANGLEESTIMATIONRESULTSAT1HZ(BLUELINE:IMU,BLACKDOTTEDLINE:MBMOCAP).
TABLE.2MEANABSOLUTEERRORSBETWEENJOINTANGLESESTIMATEDBYIMUANDMBMOCAP
Meanabsoluteerror(deg.)
Error
Hz
Mean
Standarddeviation
0.2
47.35
13.42
0.25
44.3
7.31
0.33
41.36
8.75
0.5
39.22
8.23
36.55
8.84
ThemeanabsoluteerrorsofthemeanandstandarddeviationforeachfrequencyaregiveninTable.2.Themean
absolute error under different frequencies using the ttest shows that there are no significant differences under
different frequencies. A large motion or large acceleration which instantaneously occurs might generate an
www.seipub.org/rame
irreversibleerrorwhichwouldnotbereducedevenafterasubjectstopsthemotion.
MuscleFatigueMeasurementResults
Tomeasureandtestmusclefatigue,athreestepmeasurementwasperformed.First,tonormalizeEMG,theMVC
of the EMG was measured while making the forearm perpendicular to the humerus and keeping the humerus
vertical.Second,tocalculatecoefficientsoftheEMGmuscleactivationtransformation,afullROMEFEwitha3
kgdumbbellwasmeasuredfor30swhilekeepingthehumerusvertical.Third,thesameexercisewasmeasured,
buttodeterminethecorrelationbetweenmomentumandmusclefatigue,thesubjectperformedEFE10,20,and30
times.Thesubjectrestedbetweenmeasurements,andwholestepsweremeasuredtentimes.
AnexampleofMusclefatigueforthethreecompartmentsinBIClongisshowninFig.68.TheblacklineisMA;the
bluelineisMR;andtheredlineisMF,whichdenotesmusclefatigue.Themeanandstandarddeviation(SD)ofthe
accumulated fatigue for each muscle are the same as given in Table. 3. The first column means performed EFE
times.AsthesubjectincreasedthenumberofEFEsinameasurement,fatigueincreasedlinearly.Meanwhile,asthe
totaltimesinameasurementincreased,fatigueincreasedgeometrically.Theaccumulatedfatigueunderdifferent
EFEtimesusingthettestshowsthatthedifferencesbetweentimesarevalid.
Because simulated annealing was used, the coefficients of the EMGmuscle activation transformation changed
everytimecalculatingwiththesamemotionandEMGdata.Therefore,themuscleactivationandmusclefatigue
alsochanged.
FIG.6MUSCLEFATIGUEMEASUREDINBICLONG
UNDER10TIMESREPETITION.
FIG.7MUSCLEFATIGUEMEASUREDINBICLONG
UNDER20TIMESREPETITION.
UNDER30TIMESREPETITION.
TABLE.3MEANANDSTANDARDDEVIATIONOFACCUMULATEDMUSCLE
FATIGUEFORBICLONG,BICSHORT,ANDBRA
FIG.8MUSCLEFATIGUEMEASUREDINBICLONG
BIClong
BICshort
BRA
Times
mean
SD
mean
SD
mean
SD
10
0.0637
0.0012
0.0381
0.0006
0.0318
0.0013
20
0.0836
0.0028
0.0523
0.0008
0.0466
0.0007
30
0.1204
0.0033
0.0826
0.0021
0.0743
0.0016
Discussion
To improve this study, first, IMUbased biomechanics simulator should be developed. This can reduce the
calculation burden, avoiding several stages that are needed to estimate the marker positions and create the
www.seipub.org/rame
MbMoCapbased input file described in this paper. Second, the EMGmuscle activation coefficients should be
calculatedusingthemomentinsteadofthemuscleactivation(Sartorietal.2012).Ifforwarddynamicsissimulated
withmuscleactivationbasedcoefficients,nosolutionisfoundorlargeerrorsaregeneratedoften.Third,toexpand
the DOF or measure more joint motions, the mensuration of spaculohumeral rhythm using IMUs should be
developed.Thisrhythmcannotbeneglectedbutisverydifficulttomeasureormodel(EdwardK.Chadwick2014).
IfanIMUisattachedtothesternum,whichistheonlysensorintheshouldertorsosegment,itisunderdetermined
tomeasurethespaculohumeralrhythm.Itmightbeimpossibletomeasurethisshouldermotioninseveralyears
because the current IMU has low sensitivity to translational and rotational small changes. Fourth, the muscle
fatiguemodelshouldbetestedandadjusted.Itwasdevelopedrecentlyandhasbeennotverifiedyet(Ackermann
2007).Thecoefficientsshouldbetestedandadjustedfirst.
This studydoes have limitation. First,deeply positioned muscles such as thepsoasand iliacus, which cannot be
measured with sEMG, create a gap between measured and EMGdriven motion (Sartori et al. 2012). Second,
becausetheproposedmethodisbasedonperipheralmusclemodels,centralnervesystemfatigueisneglected(Xia
andFreyLaw2008).
Conclusion
Inthisstudy,musclefatigueduringEFEmotionwasmeasuredusingEMGandkinematicinformation.Itcanbe
usedinnotonlystaticcontraction,whichismeasuredwithconventionalmethodsusingmeanpowerfrequencyor
medianfrequency,butalsodynamicmotion.Forconvenientmotioncapturing,weproposedusingIMUswhichare
easy to be used and less space restrictive in comparison with MbMoCap. The joint angles were calculated by
estimatingthemarkerpositionswithIMUsandOpenSim.Inaddition,EMGbasedmusclefatiguewasestimated
usingOpenSimandathreecompartmentmusclefatiguemodel(XiaandFreyLaw2008).Theexperimentalresults
showedthattheerrorofthejointangleandtheEFEfrequencydidnothaveacorrelation,butthatalargeerroris
generatedwhenthesubjectmakesalarge,instantaneousmovement.
Infuturework,first,theaccuracyofthejointangleestimationbasedonIMUswouldbeimprovedbydeveloping
hardware or algorithms to fulfil the requirements for biomechanics study. Second, IMU and sEMG based on BS
wouldbedevelopedtoreplaceOpenSim.Third,toexpandtheDOFormore,themensurationofspaculohumeral
rhythm using IMUs would be studied. Forth, the measurement of muscle fatigue should be verified or
comparatively studied with a more objective experiment tools like force/torque sensors. Lastly, the proposed
methodshouldbetestedonmoresubjects.
REFERENCES
[1]
Ackermann, Marko. 2007. Dynamics and energetics of walking with prostheses. Institute of Engineering and
ComputationalMechanics,UniversityofStuttgart.
[2]
Buchanan, T. S., D. G. Lloyd, K. Manal, and T. F. Besier. 2004. Neuromusculoskeletal modeling: estimation of muscle
forcesandjointmomentsandmovementsfrommeasurementsofneuralcommand.JApplBiomech20(4):36795.
[3]
Criswell,Elenaor.2011.Cramsintroductiontosurfaceelectromyography.jonesandbartlettpublishers.
[4]
DanielRoetenberg,HenkLuinge,andPerSlycke.2013.XsensMVN:Full6DOFHumanMotionTrackingUsingMiniature
InertialSensors.
[5]
David G. Lloyd, Thor F. Besier. 2002. An EMGdriven musculoskeletal model to estimate muscle forces and knee joint
momentsinvivo.JournalofBiomechanics36:765776.
[6]
Eberly,David.2010.QuaternionAlgebraandCalculus.GeometricTools,LLC.
[7]
EdwardK.Chadwick,DimitraBlana,RobertF.Kirsch,AntonieJ.vandenBogert.2014.RealTimeSimulationofThree
DimensionalShoulderGirdleandArmDynamics.IEEETransBiomedEng.7:19471956.
[8]
Holzbaur, K. R., W. M. Murray, and S. L. Delp. 2005. A model of the upper extremity for simulating musculoskeletal
surgeryandanalyzingneuromuscularcontrol.AnnBiomedEng33(6):82940.
[9]
www.seipub.org/rame
I.W.Jung,H.S.Park,Y.J.Lee,I.C.Youn.2014.MeasurementofWristJointAnglesUsingIntertialSensor.International
BioMedicalEngineeringConference2014,GwangjuKorea.
[10] Manal, K., R. V. Gonzalez, D. G. Lloyd, and T. S. Buchanan. 2002. A realtime EMGdriven virtual arm. Computersin
BiologyandMedicine32(1):2536.doi:Doi10.1016/S00104825(01)000245.
[11] Miguel T. Silva, Andr F. Pereira, Jorge M. Martins. 2011. An efficient muscle fatigue model for forward and inverse
dynamicanalysisofhumanmovements.ProcediaIUTAM2(2):262274.
[12] Pereira, Andr F. 2009. Development of a HillType Muscle Model With Fatigue for the Calculation of the Redundant
Muscle Forces using Multibody Dynamics. The 1st Joint International Conference on Multibody System Dynamics,
Lappeenranta,Finland.
[13] Rettig, O., L. Fradet, P. Kasten, P. Raiss, and S. I. Wolf. 2009. A new kinematic model of the upper extremity based on
functional joint parameter determination for shoulder and elbow. Gait Posture 30 (4):46976. doi:
10.1016/j.gaitpost.2009.07.111.
[14] Rogers, D. R., and D. T. Macisaac. 2010. Training a multivariable myoelectric mapping function to estimate fatigue. J
ElectromyogrKinesiol20(5):95360.doi:10.1016/j.jelekin.2009.11.001.
[15] Rogers, D. R., and D. T. MacIsaac. 2011. EMGbased muscle fatigue assessment during dynamic contractions using
principal component analysis. Journal of Electromyography and Kinesiology 21 (5):811818. doi: DOI
10.1016/j.jelekin.2011.05.002.
[16] Sartori, M., M. Reggiani, D. Farina, and D. G. Lloyd. 2012. EMGDriven ForwardDynamic Estimation of Muscle Force
andJointMomentaboutMultipleDegreesofFreedomintheHumanLowerExtremity.PlosOne7(12).doi:ARTNe52618
[17] DOI10.1371/journal.pone.0052618.
[18] Sorkine,Olga.2009.LeastSquaresRigidMotionUsingSVD.
[19] Xia,T.,andL.A.FreyLaw.2008.Atheoreticalapproachformodelingperipheralmusclefatigueandrecovery.JBiomech
41(14):304652.doi:10.1016/j.jbiomech.2008.07.013.