2010 IEEE EMBS Conference on Biomedical Engineering & Sciences (IECBES 2010), Kuala Lumpur, Malaysia, 30th November - 2nd December 2010. Estimation of Muscle Forces and Joint torque from EMG using SA process ArifWicaksana Oyong, S. Parasuraman, Veronica Lestari Jauw School of Engineering, Monash Universit (Sunwa Campus) Jalan Lagoon Selatan, Bandar Sunwa, 46150, Selangor Darul Ehsan, Malaysia Abstract- This paper is motivated by works done in the area of robot-assisted stroke rehabilitation. The use of electromyographic (EMG) signal brings a new way of communication interface between user and robot. However, the EMG signal has to be transferred into useful information that serve as robot input. This paper presents a novel methodology for conversion of electromyographic (EMG) signal into estimated joint torque. Investigation of the proposed methodology covers human upper limb movement: shoulder fexion-extension, shoulder abduction-adduction, and elbow fexion-extension. Simulated annealing (SA) is implemented to obtain optimum model that maps EMG into estimated joint torque. General principle, design, and the implementation of SA for the problem are discussed in this paper. Experimentation was carried out to investigate the feasibility of the proposed algorithm. The results show that the algorithm is able to find optimum model that enables EMG to joint torque conversion. Keywords- fexor tendon, tendon repair, suture pull-out, suture rupture. I. INTRODUCTION With advancement in robotic technology, it is now possible for human to work together with robots. Combination of human intelligence and advance technology offered by robot provides huge advantages in various areas, such as military, nuclear plant, medical, etc. One of the applications that motivate the works done in this paper is the use of robot in stroke rehabilitation area. The application of exoskeleton robot in rehabilitation programs will bring signifcant advantages in rehabilitation process. In this particular application, exoskeleton robot can be used to assist patient in performing exercise program. However, one of the key issues is the communication interface between robot and patient. The use of electromyographic signal is seen as a great solution to the problem. EMG is a measurement of human muscle activity. The activation of muscle generates small electrical signal. EMG system takes advantages of this phenomenon to measure muscle activity. The use of EMG signal provides advantages over conventional method [feischer]. EMG signals directly related to effort initiated by user, thus the intended movement of human operator can be recognized by the robot. By recognizing user intended movement the system can provides necessary assistance. EMG signal can still be detected even though no movement is performed. This feature of EMG provides signifcant advantage in stroke rehabilitation area. EMG-driven robot would provide great beneft for hemiplegia or hemiparesis patient. Furthermore, the EMG signals are 978-1-4244-7600-8/10/$26.00 ©2010 IEEE 81 generated unconsciously by patient, therefore there is no additional mental or physical load require activating robot motion. In such application, the assistive motion provided by robot has to be synchronized with human motion. If the system is lagging behind, the motion provided becomes resistive motion. EMG signals appear approximately 20-80ms before muscle contraction is performed [1][2][3]. The time delay provides the system with sufcient time to process the signal and provide the supporting torque. Many researchers have been carried out to investigate the use of EMG to control robotic system. A key issue in implementation of EMG in robotic application is the control strategy. Mulas, M. has developed an EMG-exoskeleton for hand rehabilitation program [4]. They system employs a threshold value, above which, the motion is activated. However, the system acts only as an on-off switch, without the capability of predicting the motion torque. Various researches have investigated the use of neural network to predict joint torque corresponding to EMG signal [5][6]. Different approach of neural network application was investigated by DaSalla [7]. In his work, neural network was used to estimate posture of human wrist and forearm based on the measured EMG signals. Although results fom neural network applications show positive results, there are still several issues regarding computational load associated with neural network. In large training data, the computational time might affect the feasibility of neural network for rehabilitation applications. Different approach was carried out by Rosen by constructing biomechanical model of human arm [8]. The biomechanical model was developed by using Hill-Based muscle model to construct mathematical representative of muscle mechanics. The use of EMG together as input to the biomechanical model was able to predict joint torque of the corresponding motion. Similar approaches of the use of biomechanical model to predict joint torque have been implemented in various studies [9][10]. However some of these methods use many anatomical and physiological parameters, which are diffcult to be obtained. In addition, these anatomical data varies for different individual. Some of the above methods use maximum voluntary contraction (muscle activity during maximum contraction), which is fne for healthy subjects but not for stroke affected patient. This project presents a novel methodology to establish a mathematical model that converts EMG signal into estimated joint torque. The system is based on Simulated Annealing (SA) process. The problem for fnding the optimum mathematical model that maps EMG signal into joint torque is rather