Wellness Monitoring of Industrial Operators Using Sparse Autoencoder Analysis of Wearable Data
Abstract
New possibilities for real-time operator health and safety monitoring have emerged with the incorporation of wearable technologies into industrial settings. Industries that deal with high levels of stress, physiological instability, and weariness must implement continuous wellness assessments to ensure the safety and efficiency of their workers. When it comes to multimodal wearable data, however, traditional monitoring methods that rely on threshold-based metrics or human supervision often miss subtle, nonlinear trends. With these restrictions, it is more difficult to identify signs of wellness decline early, which delays care and increases dangers in the workplace. The new framework, Sparse Autoencoder-based Wellness Monitoring of Industrial Operators using Multi-modal Data (SAWIM), is designed to fill this need. Wearable data such as skin temperature, heart rate variability, and accelerometer readings are compressed and learned by SAWIM using sparse autoencoder neural networks. Anomalies are detected with high sensitivity and a low number of false alarms by SAWIM through the modeling of individual baseline health profiles. Through experimental assessment on a simulated industrial dataset, SAWIM outperforms traditional PCA and thresholding approaches in anomaly identification, while also enhancing recall and accuracy in recognizing signs of stress and fatigue. Finally, SAWIM offers a data-driven, efficient method for proactively monitoring operators' well-being, enabling more robust and secure industrial operations.