Predicting the movement of the human arm by electromyography signals

Predicting the movement of the human arm by using electromyography (EMG) signals has been the subject of interest in many fields of science, including biomechanics, medicine and robotics. However, studies in the literature generally focus on estimating joint angles in the arm in a single degree of freedom. In this thesis, it is aimed to create an artificial neural network model that predicts the rotation angles of the shoulder and elbow joints of the arm at 6 degrees of freedom. In order to train the artificial neural network, data sets were formed with EMG signals obtained from the muscles in the arm and shoulder region (biceps, triceps, anterior deltoid, posterior deltoid, pectoralis major and trapezius) of subjects without limb loss and the angles calculated by the optical motion capture system of the movements during the experiments. Neural network models in multilayer perceptron and long short-term memory (LSTM) structure were trained with obtained datasets. It is found that, LSTM artificial neural network models overperform the multi-layer artificial neural network models.