A discrimination system using neural network for SMA prosthesis

The electromyographic(=EMG) signal of active muscle is observed on the surface of living body. It has been considered that the EMG signal is useful for controlling an externally powered upper extremity prosthesis. In these researches, it is reported that the signal processing system using a neural network system are effective for generating commands to the prosthesis because this system is robust on the location error of electrodes. In this paper, a new simple discrimination system using a neural network for EMG and SMA controlled prosthesis are proposed. The discrimination system is composed of two processes: rectifying and integrating EMG signals; and neural network process. The former process was adopted for faster calculation and simpler device than FFT processing. This new process will contribute to the portability of a prosthesis. Moreover SMA (shape memory alloy) actuators were used for designing a prosthesis in order to make it light and compact. As a result, a SMA prosthesis which has 3 degrees of freedom and weight 400 gr was designed