A Data-Driven Method for Hazard Zone Identification in Construction Sites with Wearable Sensors

Hazard zone identification plays a significant role in designing construction site layouts and preventing construction accidents. The existing identification methods require laborious and empirical predefinitions. Despite the emergence of some automatic hazard zone identification algorithms, they still heavily depend on stagnant regulations and rules. Thanks to the development of real-time location systems and sensor technology, the trajectory data of construction workers and equipment can be precisely collected and stored. Such data can facilitate understanding workers' and equipment's activity patterns, thus further improving the dynamic recognition of hazardous behaviors and areas. In this work, we introduce a data-driven method to automatically identify and predict potential hazard zones in the construction site. The algorithm is implemented on a digital twin platform to retrieve location data and generate real-time hazard index maps. The method consists of the following parts: (a) construction site sensor data collection and processing, (b) worker and equipment data analysis (e.g., speed, acceleration, and trajectory), and (c) hazard zone identification algorithms development. For validation, we implement the method on one railway construction project in Karlsruhe and compare the result with the close-call incidents map. This real-life case study partially demonstrates the effectiveness and accuracy of our method under the constraints of currently limited project data. On the basis of this work, further study can be conducted on the aspects of workers’ behavioral patterns and prediction model selection.