Common-cause failures (CCFs) refer to events in which multiple redundant or multiple components or systems fail simultaneously or in a correlated manner due to a single common cause. CCFs are an important consideration in probabilistic risk assessment (PRA) to assess realistic risk. While conventional PRA has primarily focused on hardware-related CCFs, increasing attention has recently been directed toward the impact of human‑induced CCFs—such as those arising from maintenance and testing activities—on core damage risk. These human‑induced CCFs may remain latent as potential failure causes shared across multiple actions and may not surface until the actions are initiated and manifest as operational failures. Consequently, they have been identified as factors that can contribute to the simultaneous loss of safety system functions. Inadequate planning, insufficient knowledge, design deficiencies, and deviations from procedures have been identified as key contributing factors.
To address these human‑induced CCFs, it is effective to perform human reliability analysis (HRA) that explicitly accounts for CCFs. In this study, we conducted a human error probability (HEP) analysis that incorporates human‑induced CCFs, based on actual trouble events that occurred at nuclear power plants in Japan. The analysis scenario consists of recovery actions performed by operators in response to trouble events caused by failures during maintenance activities that involve repeated, similar actions. In this analysis, to consider human‑induced CCFs, failure factors common to repeated operations during maintenance were identified, as well as failure factors that affect all recovery actions following the maintenance failure. These factors were incorporated into the HEP analysis. The analysis employed Phoenix, a human reliability analysis method developed by UCLA. Phoenix is a model‑based HRA method that integrates crew response trees (CRTs), fault trees including crew failure modes (CFMs), and Bayesian networks representing the relationships between PIFs and CFMs. Therefore, Phoenix enables scenario‑specific HEP estimation and allows for the treatment of dependencies among human failure events—i.e., correlations based on common causes—via PIFs.
This paper summarizes the impact of considering human‑induced CCFs on HEP results, based on the outcomes of the trial analysis, and discusses approaches to handling CCFs within HRA.