Ahmad, Siti Anom
Moving approximate entropy and its application to the
electromyographic control of an artificial hand
University of Southampton, School of Electronics and Computer Science,
A multiple-degree-of-freedom artificial hand has been developed at the University of Southampton with the aim of including control philosophies to form a highly functional prosthesis hand. Using electromyographic signals is an established technique for the control of a hand. In it simplest form, the signals allow for opening a hand and subsequent closing to grasp an object.
This thesis describes the work carried out in the development of an electromyographic control system, with the aim to have a simple and robust method. A model of the control system was developed to differentiate grip postures using two surface electromyographic signals. A new method, moving approximate entropy was employed to investigate whether any significant patterns can be observed in the structure of the electromyographic signals. An investigation, using moving
approximate entropy, on twenty healthy participants' wrist muscles (flexor carpi ulnaris and extensor carpi radialis) during wrist exion, wrist extension and cocontraction at different speeds has shown repeatable and distinct patterns at three
states of contraction: start, middle and end. An analysis of the results also showed differences at different speeds of contraction. There is a low variation of the approximate entropy values between participants. This result, if used in the control
of an artificial hand, would eliminate any training requirement. Other methods, mean absolute value, number of zero crossings, sample entropy, standard deviation, skewness and kurtosis have been determined from the signals. Of these
features, mean absolute value and kurtosis were selected for information extraction. These three methods: moving approximate entropy, mean absolute value and kurtosis are used in the feature extraction process of the control system. A
fuzzy logic system is used to classify the extracted information in discriminating the final grip posture. The results demonstrate the ability of the system to classify the information related to different grip postures.
Actions (login required)