This digital-twin system flags construction hazards in under a second, with 98% accuracy
A construction site accident often gets analyzed only after it happens. Researchers have built a digital-twin monitoring system designed to flag hazards on high-rise construction sites as conditions change. That approach differs from waiting for an incident report to identify what went wrong after the fact.
The study combines three distinct techniques into a single monitoring system. The researchers tested the system across 90 cases, reporting recall rates between 96 and 99 percent, risk-factor identification accuracy near 98 percent, and processing times under one second.
Three techniques, one system
The system begins with an improved form of gray relational analysis, a statistical method capable of working with incomplete data. Missing data is a common problem on active construction sites, where sensors and reports rarely capture every variable in real time. That technique helps identify which factors are most closely correlated with hazardous conditions even when some data points are missing.
Next, the researchers applied local linear embedding to reduce the complexity of that data. The technique compresses large, tangled datasets down to their essential structure without breaking the relationships between variables, making the information easier for the system to process quickly.
Those two techniques feed into a digital twin, a continuously updated virtual model of the physical construction site linked to live measurements pulled from the real environment. That connection lets the system compare current site conditions against the underlying data patterns in near real time.
Catching hazards that interact
Conventional safety monitoring approaches often examine hazards individually, evaluating a fall risk or an electrical risk in isolation. Researchers argue that approach misses how hazards frequently interact and compound on active construction sites. Accounting for those interactions, they say, could catch risks that isolated monitoring overlooks entirely.
The hazard categories addressed include falls, struck-by incidents, structural collapses, mechanical injuries, and electric shocks. All are common on high-rise construction sites, where multiple crews and machinery frequently operate close together across different elevations at once.
Speed matters on-site
The system's sub-second processing time carries particular significance for active job sites. Conditions there can shift within minutes as equipment moves, weather changes, or workers relocate between tasks. A monitoring system that takes too long to process data risks flagging a hazard only after the dangerous window has already passed.
By pairing statistical hazard identification with a continuously updated digital twin, the researchers say their approach could shift high-rise construction safety monitoring in a meaningful direction. Instead of a reactive process, the system aims to catch emerging risks as they develop.
The system's near-98 percent accuracy across the tested cases suggests the approach could reliably flag genuine hazards without generating an unmanageable volume of false alarms. Further real-world deployment would still be needed, however, to confirm how the system performs under the full range of unpredictable conditions found on an actual construction site.
This study was published in the International Journal of Critical Infrastructures .
