Modeling Real-Time Bicep Muscle Fatigue With IMU Sensors Using LSTM Regression
Muscle fatigue is a leading contributor to musculoskeletal disorders (MSDs), which cause significant workplace disability and absenteeism. The biceps brachii are especially susceptible in manual labor due to prolonged use during tasks like lifting and carrying. Early detection of fatigue can help prevent MSDs by enabling timely interventions, such as workload adjustments or the use of assistive devices. This study presents a real-time, noninvasive, and cost-effective system for modeling biceps muscle fatigue during dumbbell curl exercises, which simulate industrial lifting movements. Two inertial measurement units (IMUs) captured arm kinematics, and a long short-term memory (LSTM) neural network was trained to predict fatigue levels on a 0–7 scale using participants’ perceived exertion ratings. After offline training and optimization, the model was implemented in real-time and evaluated across 18 experiments, achieving a root-mean-squared error (RMSE) of 1.2. Accuracy was assessed using RMSE thresholds (±5, ±3.3, and ±2), reaching 100% at ±5 and exceeding 87% for offline and online implementations. Compared to previous studies, this approach achieved higher accuracy with fewer sensors and demonstrated reliable real-time performance, highlighting its potential for workplace fatigue monitoring and injury prevention.