A MODEL-BASED HUMAN RELIABILITY ANALYSIS METHODOLOGY (PHOENIX METHOD)

Abstract

Despite the advances made so far in developing human reliability analysis (HRA) methods, many issues still exist. Most notable are; the lack of an explicit causal model that incorporates relevant psychological and cognitive theories in its core human performance model, inability to explicitly model interdependencies between human failure events (HFEs) and influencing factors on human performance, lack of consistency, traceability and reproducibility in HRA analysis.

These issues amongst others have contributed to the variability in results seen in the application of different HRA methods and even in cases where the same method is applied by different analysts. In an attempt to address these issues, a framework for a model-based HRA methodology has been recently proposed which incorporates strong elements of current HRA good practices, leverages lessons learned from empirical studies and the best features of existing and emerging HRA methods.

This research completely develops this methodology which is aimed at enabling a more credible, consistent, and accurate qualitative and quantitative HRA analysis. The complete qualitative analysis procedure (including a hierarchical performance influencing factor set) and a causal model using Bayesian Belief network (BBN) have been developed to explicitly model the influence and dependencies among HFEs and the different factors that influence human performance. This model has the flexibility to be modified for interfacing with existing methods like Standard-Plant-Analysis-Risk-HRA-method.

Also, the quantitative analysis procedure has been developed, incorporating a methodology for a cause-based explicit treatment of dependencies among HFEs, which has not been adequately addressed by any other HRA method. As part of this research, information has been gathered from sources (including other HRA methods, NPP operating experience, expert estimates), analyzed and aggregated to provide estimates for the model parameters needed for quantification. While the specific instance of this HRA method is used in nuclear power plants, the methodology itself is generic and can be applied in other environments.