Spatio-temporal probabilistic methodology and computational platform for common cause failure modeling in risk analysis

This research paves the way for a paradigm shift in Common Cause Failure (CCF) analysis. CCF, and in general dependent failure, modeling is one of the most important topics in Probabilistic Risk Assessment (PRA). In classical PRA, CCF is addressed by data-driven parametric approaches and is based on historical CCF data. There are two major problems with the existing parametric methods: (i) their accuracy relies on the availability and quality of historical CCF data, and (ii) they lack an �explicit� connection between the root causes of dependencies, embedded in the failure mechanisms and CCF probabilities in PRA. Therefore, the parametric CCF approaches are limited in providing �cause-specific� quantitative insights that would be helpful in (a) reducing CCF occurrence, and (b) making design decisions for new systems (e.g., new reactors). This research develops a simulation-based CCF analysis that integrates explicit and implicit (stochastic) modeling of root causes and coupling associated with CCFs. This methodology is grounded on numerical outputs from a spatio-temporal probabilistic model of underlying failure mechanisms, which capture chains of causal factors leading to CCF events. The theoretical foundation of this proposed approach is described, focusing on how two key dimensions of CCF, i.e., shared root cause(s) and coupling mechanism(s), are captured in the simulation environment. The computational procedure is developed to operationalize the simulation-based CCF methodology in an Integrated PRA, which is a hybrid PRA framework that connects deterministic simulation models to classical PRA. The implementation of this new CCF methodology is conducted using a case study for a fire scenario at a nuclear power plant. This methodology, not only fills the two major gaps (i.e., relying solely on historical data and lack of explicit connection to root causes of failures) in the existing CCF methods, but also provides two additional features for CCF analysis: (1) modeling of �spatio-temporal� dependencies, and (2) treatment of the �uncertainties� in root causes of dependencies and in the associated failure mechanisms. Ongoing research by the authors is focusing on the incorporation of causal models for social root causes of CCFs (e.g., human and organizational deficiencies in maintenance) into this new methodology to connect them with the physical failure mechanisms associated with dependent failures in PRA.