Utilization of Artificial Intelligence in HAZOP Studies and Reports

Abstract

The Hazard and Operability (HAZOP) study is the most widely used hazard analysis technique in the process industry, aimed at enhancing safety and preventing accidents. It identifies potential hazards and operability malfunctions by dividing the design into small sections, called nodes, and analyzing each node's process separately. This paper briefly describes HAZOP studies and their role in enhancing operational safety across process industries. It also describes the advancements made to facilitate the performance of these studies and their challenges. To address these challenges, a new framework integrating BERTopic into HAZOP studies is proposed for enhanced efficiency and accuracy. The framework leverages historical data to categorize HAZOP elements into topics and extract node process and equipment descriptions to generate an intelligent pre-populated HAZOP analysis table. This paper focuses on categorizing causes into main risk factors for each HAZOP node and prioritizing them based on the likelihood of occurrence for each factor. The BER Topic model, incorporating embedding generation, dimensionality reduction, clustering, and topic representation, was applied to 1,574 HAZOP records from an oil pump station. The model achieved coherence and diversity scores of 80% and 92.4% respectively, outperforming Latent Dirichlet Allocation (LDA) model at 45.4% and 88.8%. It identified 13 topics, validated against hazard causes in oil pump stations and pipelines from literature. This model can be extended to categorize consequences and countermeasures, prioritizing them by severity and risk levels to generate a prepopulated table. This table can guide participants during sessions, significantly reducing the time required for the final HAZOP report.