EMG-Based Fatigue Detection System for Robotic Collaboration: A Deep Learning Approach

This paper explores the field of biomedical signal processing and deep learning by developing an Electromyography (EMG)-based muscle fatigue detection system. The proposed system leverages deep learning methods as a foundational step toward future integration with robotic control. EMG signals are collected from a small number of participants performing a fatigue-inducing task (holding a 1kg load), resulting in a small yet balanced dataset. To address this limitation, data augmentation techniques (e.g., noise injection, time-warping) are applied to expand the dataset. A Convolutional Neural Network (CNN) is then trained to classify muscle fatigue states, achieving high classification accuracy. The trained model demonstrates robust performance in distinguishing fatigue from resting conditions, validating EMG as a reliable indicator for real-time monitoring. While this study focuses on model development, the system is designed for future deployment in human-robot collaboration(HRC), where detected fatigue could trigger adaptive robotic responses (e.g., reduced speed or safety pauses).