Evaluating impact of occupational exoskeletons on physical fatigue using wearable sensors and deep learning

Abstract Musculoskeletal disorders remain a leading source of occupational injury in manufacturing and logistics industries, often resulting from repetitive lifting and load transfer tasks. While back-support exoskeletons have shown promise in alleviating physical strain, their biomechanical impact under realistic task conditions and their integration with intelligent monitoring systems require further exploration. This study investigates the efficacy of a passive lumbar-support exoskeleton in reducing muscle fatigue during manual material handling tasks by combining high-resolution surface electromyography (sEMG) data with a deep learning-based classification framework. Ten participants performed lifting and rotational (twisting) tasks with and without exoskeleton assistance, while EMG signals were collected from the lower back and thigh muscles. A total of 32 EMG features, extracted via a rolling window approach, were used to train a feedforward neural network (FNN) to classify four task-exoskeleton conditions. The model achieved a classification accuracy of 99% with perfect class separability (AUC = 1.00), out-performing traditional statistical techniques in detecting nuanced biomechanical differences. Physiological analyses confirmed significant reductions in RMS muscle activation during exoskeleton-assisted lifting and twisting tasks, particularly in the lumbar region, demonstrating the exoskeleton’s role in fatigue mitigation and postural stabilization. These findings highlight the advantages of combining wearable biosensors with interpretable AI to support real-time fatigue monitoring and adaptive ergonomic interventions. The proposed framework contributes to the development of intelligent, human-centered exoskeletal systems that align with Industry 4.0 goals for safer, more sustainable industrial work environments.