A quantitative accident analysis model on nuclear safety culture based on Bayesian network

Young Gab Kim

DOI: https://doi.org/10.1016/j.anucene.2021.108703

IF: 1.9

2022-02-01

Annals of Nuclear Energy

Abstract:As technology advances over time, systems and processes in industries have been becoming increasingly complex, and the likelihood of errors has been increasing in those work activities that workers perform without a proper understanding of them. In such an environment, a minor error in combination with weak barrier systems could lead to an unexpected accident. If a human error induced accident has failed to be prevented from occurring despite multiple barriers in place, then we should look for problems underlying in our organization. In the nuclear industry, however, accident cause analyses have been mostly focused on failed equipment, improper designs, or personal mistakes. In order to prevent recurrence, detailed cause analysis and timely corrective actions are crucial. It is also necessary to disseminate the lessons learned (i.e. operating experience) from accident analyses throughout the organization. High-reliability industries have experienced numerous near-misses and minor problems that might have otherwise resulted in major accidents. Due to organizational culture, however, lessons learned from such issues were often overlooked or considered insignificant to ensure that weaknesses could be identified in terms of safety culture (or organizational management). Therefore, methods are necessary to help analyze accidents in terms of safety culture. This paper proposes an accident analysis model based on a Bayesian network that can be used to quantitatively analyze root causes of an accident (or event) occurring at a nuclear power plant from the perspective of nuclear safety culture rather than equipment. For the purpose of this study, a case study was conducted on those actual events that had occurred at nuclear power plants. Sensitivity analysis was performed on the safety culture attributes categorized into individual, manager and system layers, which discovered more effective attributes for preventing accidents at nuclear power plants. If the analysis results of accidents (or events) at nuclear power plants based on this model are collected in an organization's database, the data may be used to monitor nuclear safety culture degradation signs and contribute to enhanced plant safety.

nuclear science & technology