Industrial safety management in the digital era: constructing a knowledge graph from near misses

Learning from incidents is instrumental for modern safety management. Over recent years, a positive safety trend proved a steady decrease in the number of high-consequences adverse events. However, accidents still happen, with various degrees of consequences. Learning should then be enlarged to include near misses, i.e. all those reported events that could have been resulted in an incident or accident but they did not. These events carry large sets of data, which should be investigated systematically to improve the current and future safety management era. Even though near misses are mandatory to be reported in the industrial sectors regulated by the Seveso regulation, the reports themselves are usually written in natural language. In this paper we propose a methodological solution to extract knowledge from near miss reports, relying on the development of a knowledge graph. The knowledge graph is grounded on a custom ontology developed to model information flows as contained in near miss reports. Out of the knowledge graph, an ontological explorative analysis is presented to get information of interest from large set of reports, otherwise difficult to analyze. The exploratory results provide evidence of the benefits of the proposed modelling solution to support safety monitoring and instructing safety interventions.