Dissertation Defense: Integrated Framework for Representing Recoveries Using the Dynamic Event Tree Approach

Committee:

• Tunc Aldemir, nuclear engineering faculty member
• Carol Smidts, nuclear engineering faculty member
• Marat Khafizov, nuclear engineering faculty member
• Valentin Rychkov
• Andrea Alfonsi

Abstract

Traditionally, probabilistic risk/safety assessment (PRA/PSA) uses Boolean logic event-tree (ET)/fault-tree (FT) formalism to quantify the risk associated with nuclear power plant (NPP) operation. The PRA process aims at identifying accident scenarios, quantifying their likelihood and evaluating their consequences. For this purpose, it often makes use of conservative assumptions and relies on expert judgment. The need for more realistic analysis and the need to reduce the resort to experts has led to the development of dynamic methodologies. These methods, such as the dynamic event tree (DET) approach, integrate probabilistic analysis with simulation of plant behavior and explores possible path of plant evolution when given accident initiators occur. While many DET applications to NPPs have appeared in the literature, the effect of repair/recovery of systems has not yet been fully addressed. This dissertation proposes a framework to systematically integrate system recoveries within DET as possible branching conditions. The framework applies also the possibility to model multiple failures and recoveries (i.e., first failure, first recovery, second failure, second recovery, etc.) for a given system within the DET. The modeling of failures and recoveries is not quite straightforward for two reasons: (i) the thermal-hydraulic (TH) model has to explicitly account for all the phenomenological dependencies among different systems and (ii) from probability perspective, failure and recovery distributions may be correlated. To address these two issues, the framework uses a modeling strategy that labels TH events, particularly those that may be similar phenomenologically and may occur under different system configurations and/or times as separate (but possibly correlated) events. Multidimensional distributions are used to account for such correlations among failure and recovery distributions. While this approach allows to model the NPP evolution in a more general and systematic way, it makes more complicated to control the TH model since the number of coded statements increases. In order to verify the model correctness, the use of a graphical tool is proposed. The capability to visualize the control logic evolution via the use of a graphical tool allows the analyst to check if the TH model control logic is working correctly. The DET requires, as input data, system/subsystem failures and recoveries probability distributions to define the times of occurrence of the branches that form the tree structure. These probability distributions may not always be available, particularly in case of multiple failures and recoveries for which interdependencies have to be accounted and need to be extracted. These distributions refer to systems as whole and reliability data (e.g., failure rate, recovery rate) related to individual components cannot be usually directly integrated within the DET analysis. An approach is then proposed in this dissertation to use component reliability data in DETs and extract system failures and recoveries distributions. Different software is used in the proposed framework to systematically integrate system recoveries within DET: RAVEN and MAAP5-EDF (an electricité de France derivative of MAAP5) codes are coupled to generate the DET, YAKINDU StateChart Tools (YSCT) is used to create a graphical model for verifying the MAAP5-EDF TH model. Finally, the software PyCATSHOO is used to generate realistic distributions for a given system, starting from the component reliability data. The different features of this framework have been applied separately to two case studies. A DET with few branching conditions arising from a loss of offsite power (LOOP) accident have been generated to show the use of the TH model verification process and the generation of system probability distributions. Finally, a realistic industrially relevant application has been developed for the case of a LOOP accident.