This paper presents a semi-automated HAZOP knowledge graph construction method using natural language processing and deep learning techniques for named entity recognition in HAZOP documents. The research addresses the critical problem that vast amounts of existing HAZOP information remain in paper form and is not effectively shared or reused across the industry. By using deep learning for entity extraction from historical HAZOP reports and constructing a knowledge graph based on ontology frameworks, the system enables better knowledge management and reuse. This is highly relevant to PSM as it leverages organizational learning from past safety studies. The knowledge graph approach allows facilities to query historical hazard scenarios, learn from similar past analyses, and identify potential gaps in current assessments. This supports the RBPS principle of learning from experience and continuous improvement in hazard identification practices.
