Deep learning-based fatigue monitoring of construction workers using physiological signals

Construction workers often suffer from physical fatigue, leading to health issues, quality compromises, and accidents. Previous research on fatigue monitoring using physiological measures has three main limitations: inappropriate benchmarking with the Ratings of Perceived Exertion (RPE) scale, which poorly correlates with actual field fatigue; data collection in controlled settings; and ignoring the time-series nature of physiological signals. These issues question the applicability of such measures for monitoring fatigue on active job sites. This paper introduces an approach leveraging deep learning models and physiological data, using appropriate benchmarks and comprehensive on-site data collection. The approach was evaluated using metrics such as accuracy, precision, recall, specificity, and the F1 Score. Results showed models like Bi-LSTM achieved up to 98.5 % accuracy, validating the effectiveness of physiological signals. This paper contributes to automation in construction by developing deep learning models for fatigue monitoring that can automate safety-related concerns for construction workers and managers. • Physical fatigue monitoring methods for construction workers were critically reviewed. • An approach that uses deep learning models and physiological data was evaluated. • The performance of three deep learning models was assessed. • The Bi-LSTM model achieved the best performance with an accuracy of 98.5 %.