Welcome to the PSAM 18 Abstract Status page.
Abstract TI259Full Paper + Presentation
Dependency analysis of human errors: Phoenix HRA methodology and a case study
Dependency among human actions remains a critical source of uncertainty in Human Reliability Analysis (HRA), particularly in complex accident recovery scenarios such as station blackout, where operators must perform multiple time-ordered actions under escalating stress, workload, and resource constraints. Traditional HRA approaches typically treat dependency using predefined levels or adjustment multipliers applied after base human error probabilities (HEPs) are estimated. Such treatments often lack explicit causal representation and do not capture the dynamic evolution of performance influencing factors (PIFs) across sequential tasks.
This paper presents a dependency analysis methodology integrated within the Phoenix HRA framework and demonstrates its application to recovery operations for a station blackout accident in an analog main control room of a nuclear power plant. Phoenix models dependency through integrated Crew Response Trees (CRTs), Fault Trees (FTs), and Bayesian Belief Networks (BBNs), enabling explicit representation of causal and probabilistic relationships among human errors and PIFs. Both direct dependency (shared PIFs across tasks) and indirect dependency (state evolution of PIFs across recovery actions) are incorporated into the quantification process.
The case study evaluates operator recovery actions under two modeling assumptions: (1) independence among tasks and (2) dependency-aware modeling incorporating PIF degradation over time (e.g., stress and fatigue accumulation) as well as shared contextual influences (e.g., team training and procedures). Results show that dependency-aware modeling leads to progressive HEP escalation across recovery actions and produces notable shifts in end-state probabilities compared to the independence assumption.
The study demonstrates that Phoenix HRA provides a mechanism-based and traceable approach to dependency treatment and highlights the importance of integrating dependency analysis early in the HRA quantification process. Implications for improving the identification of dependent actions, ensuring consistent causal modeling, and supporting risk-informed decision-making in PRA applications are discussed.
← Check another abstractThis paper presents a dependency analysis methodology integrated within the Phoenix HRA framework and demonstrates its application to recovery operations for a station blackout accident in an analog main control room of a nuclear power plant. Phoenix models dependency through integrated Crew Response Trees (CRTs), Fault Trees (FTs), and Bayesian Belief Networks (BBNs), enabling explicit representation of causal and probabilistic relationships among human errors and PIFs. Both direct dependency (shared PIFs across tasks) and indirect dependency (state evolution of PIFs across recovery actions) are incorporated into the quantification process.
The case study evaluates operator recovery actions under two modeling assumptions: (1) independence among tasks and (2) dependency-aware modeling incorporating PIF degradation over time (e.g., stress and fatigue accumulation) as well as shared contextual influences (e.g., team training and procedures). Results show that dependency-aware modeling leads to progressive HEP escalation across recovery actions and produces notable shifts in end-state probabilities compared to the independence assumption.
The study demonstrates that Phoenix HRA provides a mechanism-based and traceable approach to dependency treatment and highlights the importance of integrating dependency analysis early in the HRA quantification process. Implications for improving the identification of dependent actions, ensuring consistent causal modeling, and supporting risk-informed decision-making in PRA applications are discussed.