Neurofuzzy Logic as a Control Algorithm for an Externally Powered Multifunctional Hand Prosthesis

INTRODUCTION We are developing a controller for a multifunctional hand prosthesis based upon multiple surface electromyograms (sEMG) using neurofuzzy logic technology. The sEMG signal is successfully used as a means of control in current commercially available myoelectric prostheses. However, these are either single degree-of-freedom (DOF) devices or sequential controlled devices with locking mechanisms to switch between DOFs. There have been several proposed algorithms of control extended to multiple DOF prostheses. Early myoelectric prostheses, such as the Sven Hand [1] and the Philadelphia Arm [2], involved the use of electrode arrays and used adaptive weighted filters to process the signals. More recently, Hudgins et. al [3] proposed extracting several parameters out of the first 200 ms of EMG activity in an effort to obtain a higher ratio of controllable functions to inputs. Other algorithms have also proposed control based upon pattern classification / feature extraction [4, 5] with the use of neural networks for system training [6], while others, specifically for hand control, have been based upon spectral analysis of the EMG signal [7]. These have met with varying degrees of success, but have for the most part been limited to laboratory success, and to our knowledge, have not been demonstrated as clinically practical solutions to the multifunctional control problem. We propose an algorithm based upon neurofuzzy technology. We believe that because of the inherent “fuzziness” of human activity, a control algorithm based on fuzzy logic may have advantages for multifunctional prosthesis control. We seek an acceptable compromise between the number of electrode sites used and processing complexity, and thereby desire not more than three to four control sites to control three to four DOF. This approach delivers more information to the system and, by using fuzzy logic, reduces the complexity of the processing.