摘要
Musculoskeletal fatigue is a major cause of injuries in manufacturing, but most monitoring systems only detect injuries after they have occurred, which is particularly problematic when workers share workspaces with collaborative robots. This paper proposes a wearable sensing platform and a learning algorithm that continuously monitor operator fatigue during human-robot collaboration. The system uses eight surface electromyography (sEMG) sensors, six inertial measurement units, and four force-sensitive resistors to collect data, which are then processed into a 127-dimensional feature set. Eight classifiers were trained using data from 1536 simulated work hours involving 24 diverse individuals. With leave-operators-out cross-validation, a 1D-CNN–LSTM hybrid model achieved 89.3% accuracy and 87.1% balanced accuracy with just 73.9 kiloparameters. A CNN–BiLSTM–attention model with slightly more parameters improved accuracy by 0.3 percentage points. Alerts issued before severe fatigue reduced shoulder strain by 43% while maintaining 94% of baseline productivity.
