Understanding Human Decision Failures in Abnormal Situations using Dynamic Ontology
Operating procedures in nuclear power plants are designed based on ideal execution assumptions. However, in practice, human operators often rely on heuristics under conditions of limited information and time pressure. This leads to mismatches between prescribed procedures and actual human actions—an issue that directly impacts safety.
This research highlights that such mismatches are not merely the result of errors or lack of training, but stem from the structural distance between procedural logic and human cognition. To analyze this gap, we structure procedures as knowledge graphs based on dynamic ontologies and compare them with actual decision-making flows to quantitatively identify points of divergence.
In addition, we analyze regulatory documents from the U.S. NRC and Canada’s CNSC to identify cases where seemingly similar terms (e.g., Design Basis Accident vs. Postulated Initiating Event) actually reflect different regulatory assumptions. This is achieved through AI-driven semantic matching and ontology-based document structure analysis.
This research provides a foundation for practical and safety-oriented procedural improvements and for enhancing human trust and reliability in automated reactor systems, with consideration for procedure designers, educators, and operators alike.
Researchers
