Industrial internet of things and unsupervised deep learning enabled real-time occupational safety monitoring in cold storage warehouse

Occupational safety and health (OSH) has always been a big concern in the labor-intensive warehouse industry, especially under peculiar circumstances like a low temperature. Accordingly, this paper aims to propose a framework of a smart system using the Industrial Internet of Things (IIoT) and digital twin (DT) technologies to realize real-time occupational safety monitoring in the warehouse and ensure synchronized cyber-physical spaces for information traceability and visibility. The unsupervised deep neural structure of stacked auto-encoder (SAE) is designed to identify abnormal stationary from human motion status, which is perceived as a sign of potential accident. The model is developed to automatically update online by cooperating with calibration samples so as to keep in accordance with the evolution of surroundings. The Bluetooth low energy (BLE) and a log-distance path loss model are used to fulfill indoor localization in order for managers to promptly respond to an incident on site. Besides, some intelligent services are enabled to promote the efficiency of safety management. A real-life case study is carried out in an air cargo cold storage warehouse to illustrate the viability and rationality of the proposed system and methods. The elaboration of the implementation is envisioned to facilitate replication and reproduction effectively. The impact of learning features concerned with distance and vibration on the performance of anomaly detection has also been analyzed by experiments. The insights and lessons gained in this study hold the promise of providing a reference or sparking new ideas for researchers and practitioners to meet similar needs in practice.