Experimental comparison of emg-based control methods used in bionic prosthetic hands and hand angle estimation from surface EMG signals

The number of physically disabled individuals is increasing rapidly with the increasing and ageing population in the world. Unfortunately, the number of amputees who can access prosthesis does not increase in parallel with this number. While economic reasons prevent amputees from accessing prostheses, a large proportion of users who can access prostheses abandon the use of prostheses. Although prosthesis developers use many sensory and biological cues to reliably control the prosthesis, the output falls far short of natural limb capabilities. In the most general form, this thesis analyses the current control methods used in myoelectrically controlled prostheses for upper limb amputees and aims to guide efforts to develop new control methods. The thesis includes three different applications. In the first application, the myoelectric prosthesis control methods used in today's academic and commercial studies are comparatively analysed and in the light of the results obtained, original suggestions are presented for the development of more comfortable and more humane prosthesis control methods for the user. Within the scope of the second application, a regressionbased application that can be applied in real time, which includes the estimation of the prosthesis hand opening angle from the surface EMG signals obtained from the relevant muscles of the individual, has been realised. This application was performed using two different estimators (MLP and MLP-NARX), two different hand/wrist movements (hand opening and wrist extension) and three different forearm sEMG data acquisition points. As a result of the application, it was observed that the MLP-NARX model was able to predict the hand/wrist movement angles with very high accuracy (>99%) depending on the EMG input. In the third application, a simulator was designed to facilitate the adaptation process of amputees to prostheses and to make amputee training more effective. The results obtained from the thesis applications reveal the importance of approaches to develop prosthesis inspection methods that are customised, differentiable (using hybrid inspection methods) and capable of mimicking the intact limb (regression-based). The proposed methods will be used in future studies to design myoelectric prosthesis controllers that can simultaneously perform multiple joint movements with surface EMG data collected from multiple muscles.