This paper proposes a fusion model for developing a data-driven risk association network, based on historical inspection records. The fusion model first re-categorizes the hazards based on the similarity in their occurrence patterns. Second, spatial and temporal heterogeneity of the hazard occurrence is examined, after which site-specific records as outliers are removed from the database. Third, a structured learning approach is used to investigate the causal relations between safety risks and the weight of each relation is calculated based on the association rules. Finally, the causal relations and weightings are fused to form the hazard association network, based on which critical hazards can be identified for safety management strategy planning. Safety management for an elevator installation and maintenance is used as a domain to validate the proposed fusion model, which develops the hazard association network using a dataset with 110,698 safety inspection records on 25,729 sites (with elevator installation or maintenance) managed by an elevator company. Using the developed network, critical hazards on the sites are identified for proactive construction management.

safety inspection; data mining; real-time association rules; hazard association network; data fusion; hazard pattern; proactive safety


No posts