Advancing Real-Time Physiological Health Monitoring in Construction: A Feasibility Study of Fatigue Estimation of Construction Workers on Smart Devices

In the dynamic field of construction health monitoring, effectively addressing real-time physiological concerns encountered by on-site workers is a pressing challenge. While recent strides in wearable sensor technology and machine learning have demonstrated success in detecting various health parameters such as physical fatigue, mental stress, and heat stress, implementing these technologies on construction sites poses substantial hurdles. Privacy concerns associated with cloud-based data transmission and the impracticality of deploying large CPU and GPU devices underscore the need for lightweight personalized alternatives, such as mobile phones, for on-site computation. This study addresses this need by developing a physical fatigue estimation framework based on physiological sensing and machine learning algorithms, optimized for speed and performance in lightweight devices. The proposed framework leverages Photoplethysmography (PPG), Electrodermal Activity (EDA), and Skin Temperature (ST) signals as input for a deep neural network to assess the physical fatigue of the subject. The network is optimized using the quantization technique. The framework is evaluated using metrics of accuracy and latency, and the results demonstrate that the proposed framework achieves significant performance enhancements. The framework was able to achieve an accuracy of 81% with a latency of 0.02 ms. By developing an optimized framework for physiological sensing, this study not only responds to the challenges of construction health monitoring but also sets the stage for the practical, privacy-conscious deployment of health monitoring systems within the construction industry.