HAZOPCT: A HAZOP analysis completeness tool based on knowledge graph reasoning

Abstract

Hazard and Operability (HAZOP) analysis is widely recognized as a cornerstone of risk assessment in the chemical industry. However, the reliance on manual processes makes HAZOP analysis time-consuming, labor-intensive, and subject to variability, leading to inconsistent report quality and increased accident risks. To address these challenges, this paper introduces a HAZOP analysis Completeness Tool (HAZOPCT), which helps to review process designs for enhancing the completeness of HAZOP analysis. Initially, a HAZOP knowledge graph is constructed automatically to integrate multi-source process safety information based on a predefined ontology. Subsequently, HAZOPCT is employed to perform reasoning over the knowledge graph, thereby generating recommended conclusions from similar cases for completing HAZOP analysis. A case study focusing on the sulfur recovery process demonstrates the ability of HAZOPCT to efficiently identify hazards and operability issues that may have been overlooked in original HAZOP analysis tables. Additionally, ablation studies validate the critical role of semantic information and graph attention mechanisms in improving reasoning performance. The generated attention weight network further enhances the interpretability of HAZOPCT, offering insights into the automated reasoning process and contributing to a more transparent and reliable HAZOP analysis.