Muscle signals from electromyography (EMG) are widely used to detect muscle contraction and intention to motion. By using these signals in real time in prosthetic control, a low signal to noise ratio is commonly found. Thus, it is necessary to have recursive methods, robust to noise and efficient algorithms, to generate commands in real time for the robotic actuator. In this research, stochastic system indentification techniques, Kalman filter, sensor fusion and stochastic predictive control techniques were investigated and applied to improve the measurement and processing of electromyographic signals to increase robustness in the control of biomechatronic prostheses. Thus, it is an improved process, less sensitive to noise and with minimal delays and phase lags. In this methodology, a four-stage distribution method is used: (1) features extraction by using an autoregressive model (AR), (2) data fusion with the Kalman filter, (3) motion estimation algorithm, and (4) predictive control with the generalized minimum variance controller applied to a servomechanism. The main objectives were: to enhance the signal-to-noise ratio of EMG signals, to have a low-cost real-time processing man-machine interface, to avoid measurement problems and to minimize energy consumption of the control system. A didactic plant was developed, which is a 4 channel EMG data acquisition and processing system with a servomechanism and its control system coupled in a robotic jaw. Practical tests were conducted with the prototype and the results show that it is possible to continuously estimate the intention of opening and closing movement of the hand and can confirm the good performance of the stochastic controller designed for the control of the electric prosthesis.
|
