Electromyography (EMG) Based Control for Shoulder Muscles

Long Doan Thanh1, Nhon Phan Nguyen Quy1

1

Faculty of Electrical Engineering, UTP University, 31750, Perak, Malaysia doanthanhlong777@gmail.com, dumber89@gmail.com The proposed TDANN system consisted a static, two-layers artificial neural network, with nonlinear tansig neurons in the hidden layer and linear neurons in the output layer. By further development of proposed method, Au and Kirsch (2001) suggested that the TDANN system would be trained using backpropagation with Matlab Neural Network Toolbox (The MathWorks, Inc.). Alsmadi et al. (2009) explain that a typical back-propagation network has an input layer, an output layer, and at least one hidden layer, each layer is fully connected to the succeeding layer as in Figure 1. Further details will be described in Theory/Techniques.

AbstractWe study previous researches and This electronic document is a live template. The various components of your paper [title, text, heads, etc.] are already defined on the style sheet, as illustrated by the portions given in this document. Abstract is limited to 200 words. Keywords-EMG-based control, shoulder muscles, artificial neural network,

I.

INTRODUCTION

Electromyography (EMG) is the technique of testing the electrical properties of muscles. In recent decades, EMG has been used to help people to diagnose of nerve compression or injury and other muscles problems. According to Losier (2007), it is very reliable to derive the control source from the myoelectric signal for the prosthetic equipments. For the clinical treatment nowadays, acquiring the electromyography signals from particular muscles and using that to identify intended movements for some prosthetic control becoming more and more popular. In this paper, we mainly focus on the use of shoulder movement and EMG as combined input sources for the control of prosthetic limbs. We process with two researches which are Time-Delay Artificial Neural Network (TDANN) and the Surface EMG based robotic rehabilitation system. Since this is only the preliminary study, a lot more research need to be carefully done to widely bring this technology to the medical treatment system in the future. II. LITERATURE REVIEW In order to develop a system that anticipate kinematic movement from shoulder muscle EMG signals, many works have been done with various types of approaches. Among them, Kirsch and Au (1997) proposed a significant method of utilizing Time-Delay Artificial Neural Network (TDANN) with positive results. In their experiment, the authors acquired EMG signals from 6 shoulder muscles as anterior deltoid, middle deltoid, posterior deltoid, biceps, triceps, and the clavicular portion of the pectoralis major. These raw signals then were digitally rectified and lowpass filtered and became input to TDANN system. According to Sinha et al. (2000) TDANN is a feed forward network that forces input signals through intermediate tapped delay line to process time series data by converting the ternporal sequence into a static pattern by unfolding the sequence over time.

Figure 1. Typical back-propagation network

Figure 2. Structural diagram of the EMG time-delayed artificial neural network (TDANN). A hidden layer with a nonlinear (tansig) transfer function was followed by a linear output layer. The b and w parameters associated with each layer indicate biases and weights, respectively.

The outputs from TDANN system are 4 joint angles: shoulder horizontal flexionextension (HFE), shoulder internalexternal rotation (IER), elbow flexionextension (EFE), and shoulder elevation-depression (ED), each joint angle possesses its own parameters as: angle, angle velocity and angle acceleration. Figure 3 illustrates these angles and positions of EMG electrodes.

recorded and amplified 1000 times using bipolar noninvasive surface electrodes. After being acquired from muscles, sEMG signals were digitally process the signals using 3 techniques: full wave rectification, feature extraction and normalization with respect to the maximum voluntary contraction (MVC) value. The amplitude of MAV (mean absolute value) can be regarded as indications of the power of signals and the representation of the muscle activation level. It can be used for the feature extraction for joint torque estimation. From there, control input for the robotic rehabilitation system was gained. III. MATERIALS AND METHODS

3.1 Control Block Diagram

Figure 3. Definition of output joint angles and EMG electrodes position.

As shown in Figure 3, Au and Kirsch (2001) utilized OptoTrak (Northern Digital, Inc.) motion analysis system to record the joint angles in order to test the efficiency of EMGto-kinematic prediction from TDANN system. With the results from an arbitrary subject, they proved that TDANN system is able of anticipating shoulder/arm movement from collected EMG signals with acceptable errors. In addition to shoulder movement, there have been other researches to estimate the shoulder force utilizing EMG signals. Laursen et at. (1998) proposed such a system with an equation to calculate shoulder muscle force based on EMG signals from 10 positions, abduction and flexion glenohumeral joint moments and physiological cross-sectional areas for the shoulder muscles. The results from Laursens estimation could be used to regulate the intensity of force in kinematic movement. According to Song et al.(2007), most of the patient who participate in the recovery experiment with the assistance of the myoelectric controlled robotic system had the improvement in the range of motion (ROM) . Moreover, the increasing in assistance of the system results in the decrease in the amplitude of agonist EMG signal, which means less effort was required for the patients to perform the movement. By inducing active muscle contraction of the patients, Park et al. (2011) introduced the sEMG based robotic rehabilitation system that helps recover the neural or the musculoskeletal system of paralyzed patients. The assistive rehabilitation system which can interpret motion intention of the patient by sEMG can simplify the recovery process through the voluntary stimulation of nerve and muscle. Among six muscles related to shoulder flexion motion, there are 2 muscles named as anterior deltoid and posterior deltoid generally dominate shoulder flexion/extension motion and are chosen as target muscles. The activities of muscles were
Figure 4. Proposed control block diagram

3.2 System Modelling 3.3 Theory / Techniques 3.3.1 Back propagation network technique The number of hidden layers and the number of processing element per layer in the back-propagation network are dependent on designers. Feed forward back propagation network applies training inputs into input layer and the desired outputs are compared with actual outputs at output layer. The difference between the output of the final layer and the desired output is back-propagated to the previous layer(s), usually modified by the derivative of the transfer function, and the connection weights are normally adjusted using the Delta Rule. The Delta Rule varies connection weights proportionally to the error (between actual and desired outputs) times a scaling factor for global accuracy. 3.3.2 Mean absolute value In robotic rehabilitation system by Park et al. (2011), the equation of MAV is expressed as follows:

MAV

1 N

x
k 1

for k=1,2,..N.

Where

xk is the raw sEMG signal in segment k.

N is the number of samples (i.e. window length of moving average). It is known that the normalized MAV with respect to MVC represents muscle force (or joint torque) at each elbow angle configuration (Dipietro et al., 2005). Control input from the sEMG is the value of maximum torque at MVC multiplied by normalized MAV of posterior deltoid subtracted by anterior deltoid with 200 samples of window length (i.e. N = 200) (Park et al.,2011). IV. DISCUSSION

Consequently, our proposed system of combing Au and Kirschs TDANN system, Laursens force prediction and Park et al. movement actuator is supposedly a complete system from detecting EMG signals to generating corresponding movement for arm/shoulder. For the mechanical aspect, the actuator has considerable impedance which will affect the operation of the system. It requires further improvement to compensate this impedance for a better functional system. For the Surface EMG based robotic rehabilitation system

Figure 5. Performance of TDANN in predicting 4 joint angles of different movement conditions for one subject with C5 tetraplegia. Thin traces are experimentally recorded joint motions while thick traces are EMG-based predictions from the TDANN

Table 1. Laursens model calculation results for 6 subjects named A to F with correlations (r), slopes () and mean residual error (m.r.e.)

Figure 6.The results of assistive shoulder flexion in case of using sEMG control signal

By the results of Figure 5, Au and Kirsch (2000) together with Kirsch et al. (2001) have proven that TDANN can efficiently predict the angular directions of shoulder/arm movement. However, their work did not include the mechanic control scheme to translate the predicted angular motion into any devices spacial movement. Park et al. (2011) offered such a control system called motion actuator set with geared DC motor, encoder and magnetic brake. Nevertheless, this actuator set attempted to image the right hand movement to the paralyzed left hand. The drawback of this system is that the sEMG signal generated from the subjects is very weak so it needs more precise technique and research to extract required assistive inputs. Hence, we proposed to replace the signals for movement image by the output signals of TDANN angular prediction. In order for the actuator set of Park et al. (2011) to work, not only the angular input is required but also the torque input. From Laursens work of force prediction, this input torque can be calculated and fed into movement actuator set.

According to the result in Fig 6, The rehabilitation system using sEMG is also can generate the torque and the angle from the shoulder muscle and apply that as control for the prosthetic limbs. For stroke survivors, process of stimulating nerve and muscle from voluntary movements by this sytem are really important. Although considered as a preliminary study before applying in the real clinical treatment, the robotic system with sEMG is expected to be a perspective solution in recovering functionality of neural/musculoskeletal system. As state before, the drawback of this system is that the sEMG signal generated from the patients is very weak compared to healthy person. As the result, more precise technique and research need to be done to extract required assistive torque. For the mechanical aspect, the actuators in the robot system have the considerable impedance which will affect the operation of the system. By compensating the impedance, people can improve the performance of the system.

V.

CONCLUSION

For the development of EMG applications and comfortable of the patients, some researchs have proposed a compact size and a wireless surface electromyography measurement system. The developed system not only has a simple form but also amazingly produce EM signals with better signal-to-noise ratio (SNR) compared to commercially available EMG system (Youn&Kim,2009). ACKNOWLEDGMENT Avoid the stilted expression, One of us (R. B. G.) thanks . . . Instead, try R. B. G. thanks. REFERENCES
[1] G. Eason, B. Noble, and I. N. Sneddon, On certain integrals of Lipschitz-Hankel type involving products of Bessel functions, Phil. Trans. Roy. Soc. London, vol. A247, pp. 529551, April 1955. (references) J. Clerk Maxwell, A Treatise on Electricity and Magnetism, 3rd ed., vol. 2. Oxford: Clarendon, 1892, pp.6873. I. S. Jacobs and C. P. Bean, Fine particles, thin films and exchange anisotropy, in Magnetism, vol. III, G. T. Rado and H. Suhl, Eds. New York: Academic, 1963, pp. 271350. K. Elissa, Title of paper if known, unpublished. R. Nicole, Title of paper with only first word capitalized, J. Name Stand. Abbrev., in press. Y. Yorozu, M. Hirano, K. Oka, and Y. Tagawa, Electron spectroscopy studies on magneto-optical media and plastic substrate interface, IEEE Transl. J. Magn. Japan, vol. 2, pp. 740741, August 1987 [Digests 9th Annual Conf. Magnetics Japan, p. 301, 1982]. M. Young, The Technical Writer's Handbook. Mill Valley, CA: University Science, 1989.

[2] [3]

[4] [5] [6]

[7]

Browse Conference Publications > Industrial Technology 2000. P ... Prev | Next
Page Help

I.

DYNAMIC NEURAL NETWORKS: AN

OVERVIEW

This paper appears in: Industrial Technology 2000. Proceedings of IEEE International Conference on Date of Conference: 19-22 Jan. 2000 Author(s): Sinha, N.K. McMaster Univ., Hamilton, Ont., Canada Gupta, M.M. ; Rao, D.H. Volume: 1, On Page(s): 491 - 496 vol.2 Product Type: Conference Publication