Perpustakaan Digital ITB

Musculoskeletal disorders are among the leading causes of disability worldwide and often affect the functionality of the upper extremity. Monitoring joint angles is important for understanding human movement, rehabilitation, and biomechanical analysis. Conventional measurement methods such as optical motion capture systems provide accurate results but require expensive equipment and controlled laboratory environments. Surface electromyography (sEMG) offers a more accessible alternative because muscle activation signals contain information related to joint movement. Therefore, this study aims to estimate the joint angles of the upper extremity using sEMG signals combined with machine learning techniques. The research methodology involves collecting sEMG and motion capture data from approximately ten participants performing several upper extremity movements, including elbow flexion–extension, shoulder flexion–extension, shoulder hyperextension, and composite movements. sEMG signals are processed through filtering, rectification, and envelope extraction, while joint angle data are obtained from optical motion capture recordings processed using OpenSim inverse kinematics. The processed data are then used to train a Long Short- Term Memory (LSTM) machine learning model to learn the relationship between muscle activation patterns and joint angles. The results show that the machine learning model can estimate upper extremity joint angles from sEMG signals, particularly when the predicted joint is actively involved in the movement. Composite movements produce better prediction performance due to stronger and more consistent muscle activation patterns. However, the model performance decreases when predicting joints not directly involved in the movement and when tested on participants outside the training dataset. These findings indicate that sEMG-based joint angle estimation using machine learning is feasible but requires further improvements in data diversity and model generalization.