Development and Benchmarking of RAVEN with TRACE for use in Dynamic Probabilistic Risk Assessment

Abstract

The identification of potential accident conditions for a nuclear power plant requires a systematic evaluation of postulated hazards, and accurate methods for predicting the behaviour of the system if these hazards were to occur. It is particularly important to identify scenarios which carry severe consequences (e.g., large radioactive releases to the environment), even if the conditions have a low probability of occurrence, so that preventative measures can be implemented. Dynamic probabilistic risk assessment (DPRA) is a field of analysis that aims to determine the failure pathways of complex systems while simultaneously analyzing the time-evolution of the proposed accident. By studying the dynamics of the system, DPRA methods are capable of analyzing the impact of impaired or late equipment response, human actions during the transient, and the inter relationship between different systems and failures. This approach promotes realistic predictions of the complex response of the system under accident conditions, and for the dynamics of the accident progression to unfold with timing that is not pre-determined by an analyst, thereby removing potential user bias from the results. The work that is outlined in this thesis was undertaken in order to demonstrate the DPRA software platform called RAVEN, and to leverage its application in the near-future probabilistic assessment of accident conditions applied to CANDU reactor simulation models. Features of the work include: • Demonstration of the capability of RAVEN to produce predictable results using the dynamic event tree (DET) approach; • The development of a code interface to allow RAVEN to drive DET simulations of TRACE simulation models; and • Demonstration of the capability of the developed RAVEN-TRACE interface to produce predictable results for systems that are well-understood.