Research of the Neural Network Control for Functional Electrical Stimulation

• Main Authors: Luh, Jer-Junn, 陸哲駒
• Other Authors: Lai, Jin-Shin
• Format: Others
• Language: en_US
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/01189381180466409163

博士 === 國立臺灣大學 === 電機工程學研究所 === 87 === The functional electrical stimulation (FES) has been demonstrated to be effective in restoring hand function in quadriplegia and locomotion in paraplegia and hemiplegia. In this dissertation, some important issues about FES, including the EMG analyzing system for user interface, the ability that the neural network can identify the information contained in a single joint movement, and a diagonal recurrent neural network based neuro-control system for joint position control, are investigated. An active electrode based on an instrumentation amplifier (AD620) ,a PC based human-machine interfacing with TMS320C31 and a portable system with a ADSP2101 are introduced and validated. The results of the experimental validation showed that the EMG analyzing system is useful as human-machine interface. Applications of the system described here benefit both the disable who wishes to use a myoelectric signal as a command resource and the engineers who want to validate their new identification algorithms. Feed-forward network with one layer of hidden nodes was constructed to learn the characteristics of the single joint movement. The experimental validation was achieved by rectified, low-pass filtered EMG signals from the representative muscles, joint angles and joint angular velocities and measured torque. Learning of the neural network allowed accurate prediction of isokinetic joint torque from novel EMG activities, joint position, and joint angular velocity. Predictions were well correlated with the experimental data (the mean root-mean-square-error and correlation coefficient in learning were 0.0290 and 0.998, respectively, and in 3 different speed testings were 0.1413 and 0.900, respectively). These results suggested that an ANN model could represent the human joint dynamics. A diagonal recurrent neural network (DRNN) based functional electrical stimulation (FES) system was designed to control the knee joint to move in accordance with the desired trajectory of movement through stimulation of quadriceps muscle. This system, which consisted of a DRNN controller and a DRNN identifier, could learn the non-linearity of the plant (knee joint) and control it both in on-line condition. The simulated knee joint angle was controlled with only small deviations along the desired trajectory with the aid of neural controller.