Abstract
In major-accident hazards plants, regulated by Italian Legislative Decree 105/2015 (implementing European Directive 2012/18/EU), all significant events (accidents and near misses, but also anomalies and non-conformity) occurred in the last decade are reported in a specific legislative format called “operational experience”. Since 2016, these documents have been collected by National Institute for Insurance Against Accidents at Work (Inail) into a specific repository named EsOpIA (Operational Experience and Artificial Intelligence) that applies Machine Learning (ML) techniques to automatically extract and classify concepts and relations and makes them available for sharing among different stakeholders. The archive contains over 6000 reports, used for research purposes. Starting from the data collected in EsOpIA, this study presents a first Italian characterization based on accident, near misses and anomalies occurred in upper-tier Seveso establishments since 2000. The type of event, the critical equipment and barriers implemented are considered. This analysis is not merely statistical; rather, a new approach to studies at different levels of detail, taking advantages by NLP (Natural Language Processing) and ML capabilities. Results contribute to support establishments’ operators and Seveso inspectors and to improve plants’ safety by developing specific methodologies and guidelines.
