Universität der Bundeswehr MünchenFakultät für InformatikInstitut für Technische InformatikUniv.-Prof. Klaus Buchenrieder, Ph.D./OSUWerner-Heisenberg-Weg 3985577 NeubibergDimensionality Reduction for the Control of PoweredUpper Limb ProsthesesECBS 2007ECBS 20072Götz von BerlichingenGottfried von Berlichingen 1480 - † 23. July 1562 was a German knight (Deutscher Ritter), Soldier ofFortune and robber baron. He was born around 1480 to a noble family at the Schloss Jagsthausen inWürttemberg.June 23 1504, Berlichingen and his company fought for Albert IV, Duke of Bavaria. During the siegeof the city of Landshut, he lost his right arm to an enemy cannon fire. He had a prosthetic ironreplacement made, which is still on display at the Schloss Jagsthausen today.ECBS 20073Ernst Ferdinand SauerbruchChirurgische Universitätsklinik in the Nußbaumstraße in Münchenhttp://www.deutsches-museum.de/dmznt/ersatzteile/hand/sauerbruchtechnik/index.html1916ECBS 20074Surface Controlled ProsthesesECBS 20075System ElectrohandsMYOBOCK System Electrohandhttp://www.oandp.com/edge/issues/articles/2005-11_05.aspPhoto courtesy of LIVINGSKINECBS 20076Origins and Nature of MESThe innervation process The neuro-muscular junctionModified from: E.N. Marieb, Human Anatomy and Physiology, 4th edn. Benjamin-Cummings, 1998ECBS 20077Origins and Nature of MESThe propagation of the depolarized region.The propagation action potential and theassociated external potential.Modified from: E.N. Marieb, Human Anatomy and Physiology, 4th edn. Benjamin-Cummings, 1998ECBS 20078ECBS 20079Origins and Nature of MESModified from: E.N. Marieb, Human Anatomy and Physiology, 4th edn. Benjamin-Cummings, 1998123ECBS 200710Skin ElectrodesECBS 200711Myoelectric SignalsECBS 200712Wrist FlexionPalmer View of Forearm and Hand1. Flexor carpi radialis2. Flexor carpi ulnarisECBS 200713Wrist FlexionPalmer View of Forearm and Hand1. Flexor carpi radialis2. Flexor carpi ulnarisECBS 200714Wrist ExtensionPalmer View of Forearmand Hand1. Extensor carpi radialis longus2. Extensor carpi radialis brevis3. Extensor carpi ulnaris4. Extensor digitorumECBS 200715Palmer View of Forearmand Hand1. Extensor carpi radialis longus2. Extensor carpi radialis brevis3. Extensor carpi ulnaris4. Extensor digitorumWrist ExtensionECBS 200716MES Capture PointsFlexor carpi radialisExtensorcarpi ulnarisExtensor carpi radialis longusExtensor carpi radialis brevisFlexor carpi ulnarisExtensor digitorumECBS 200717Englehart & Hudgins’ Ctrl. ConceptAmplification &PreprocessingFeatureExtractionMESClassLabelsDimensionalityReductionClassification• Time Domain• Wavelet • Wavelet Packet• Short-Time Fourier• Feature Selection• Feature Projection• MLP• LDA• StatisticalDimensionality Reduction:Important information for best class discrimination is retained and superfluous information is discarded.ECBS 200718UniBw MES-Based ControlAmplification &PreprocessingAmplification &PreprocessingFeature ExtractionNeural NetClassifierMAVRMSWFLMAVSLPWAMPSSCVARIZCMESMESData SetsMLP TrainingWeightsMotor ControlGlove Amplifier& PreprocessingTraining GloveDynamic Hand∆wij(t+1) = λ⋅δi⋅xi + α⋅ β ⋅∆wij(t)•The variable momentum term αreduces the problems related to« steep valley drops » and « flatplateaus » (Zell 1994).•The training term β controls therate of weight advancement dueto the expressiveness of thegesture (Buchenrieder 2006).ECBS 200719Selected Time-Domain Features€ RMS =1Nxi2i =1N∑.€ WFL = xi− xi −1()i =1N∑.€ ZC = gzcxk( )k=1N∑withgzcx( )=1 if sgn −xk× xk+1( )∧ xk− xk+1( )≥ S,0 otherwise     wherebysgn x( )=true if x ≥ 0,false otherwise   .Root Mean Square (Quadratic Mean):Statistical measure of the magnitude or strength of the muscle contraction. The RMS for N samples {x1,…,xN} is:Waveform Length:Information about the frequency range and the amplitude of the signal.Zero-Crossings:Number of sign-changes along a signal within a predefined time-window. The threshold S is introduced to reduce noise effects. ZC provides information about some frequency related properties. For two consecutive samples xk and xk+1, over the whole time frame holds:ECBS 200720Traditional Cluster SelectionSensor placed at theFlexor Carpis RadialisSensor placed at theExtensor Digitorum100 Points each shown for:Wrist FlexionWrist ExtensionFist“Center of Mass” Calculation MethodECBS 200721Dimensionality ReductionDimensionality ReductionECBS 200722Normal DistributionsRed: Wrist extension, Flexor carpi radialis, RMSRed: Wrist extension, Flexor carpi radialis, RMSGreen: Radial flexion, Flexor carpi radialis, RMSBlue: Wrist flexion, Flexor carpi radialis, RMSPurple: Fist, Flexor carpi radialis, RMSECBS 200723Features of 4 Hand PositionsECBS 200724Combination of Two Featureszc1, rms1:ECBS 200725Combination of Two Featuresnzc1, nrms1:ECBS 200726Distance CalculationIntersection point : 4.35424, Probability: 0.000198829The area corresponding to a probability of p=0.001 is: 0.0454993Eucledian distance: 6.26128, Angle: 4.19443ECBS 200727Distance CalculationIntersection point : 1.95424, Probability: 0.000198829€ CombinedNvd xint+, xint−,a1,d1,a2,d2( )=max12πd1( )exint +−a1( )22d12+12πd2( )exint +−a2( )22d22,12πd1( )exint −−a1( )22d12+12πd2( )exint −−a2( )22d22                    ;€ Inter sec tion a1,d1,a2,d2( )= xint+, xint−=1d12− d22a2d12− a1d22±a12d12d22− 2a1a2d12d22+ a22d12d22−2d14d22logd2d1        + 2d12d24logd2d1                                      ;ECBS 2007282-D Feature Worst-Case DistanceWrist-FlexionWrist-ExtensionRadial-Flexionrms31-wfl31rms41-wfl41rms61-wfl61RMS1-WFL10.258554.6172 • 10-10.00336685Fistrms21-wfl21.000029509.00535947rms31-wfl31.0671984rms41-wfl41ECBS 200729Selection Matrix RMS-WFL-ZCNormalized Result ZC1-WFL2: Normalized Result ZC1-WFL1:RMS2WFL1WFL2ZC1ZC21.3952.258550.0201346.00440842.0210817RMS1.208677.0501229.00391864.00391293RMS2.0405795.00352604.02322640WFL1.00335150.00446167WFL2.00453696ZC1EUROCAST 2007ECBS 200730Improved UniBw-Hand ControlAmplification &PreprocessingAmplification &PreprocessingFeature ExtractionGuilin HillsFeature SelectionMAVRMSWFLMAVSLPWAMPSSCVARIZCMESMES•A minimal set of features isdetermined to