Using Artificial Intelligence to Analyse Incidents Involving Falling Outfitting Structures

Abstract This paper presents a feasibility study to investigate the value of modern search methods, how can the use of ontologies (knowledge graphs), artificial intelligence (AI) and text analysis, provide insight into trends and learning for falling objects and structural elements. The study leveraged the Havtil incident database, which contains all incidents reported operators in the Norwegian petroleum sector (onshore and offshore). The aim was to enhance risk awareness and identify trends related to secondary and tertiary structural components beyond the main structure. The methodology involved extracting, structuring and contextualizing free-text data with the assistance of a knowledge graph, integrated with an SQL database, to facilitate advanced query capabilities. Large Language Models (LLMs) were employed to convert natural language queries into SQL, enabling efficient data retrieval and analysis. The study demonstrates the effectiveness of combining ontologies with LLMs to improve data quality, consistency, and integration, ultimately providing deeper insights and supporting informed decision-making. The results highlight the potential of AI-driven analysis to enhance safety and risk management in the petroleum industry, with a long-term vision of sharing insights across the industry to foster a collaborative approach to incident prevention and safety improvement.