Author: Joy Shen
Date/Time: October 2nd, 2025 at 12:30 PM EST
Location: EGR-1179, Glenn L. Martin Hall
Committee Members:
- Dr. Michelle Bensi, Chair
- Dr. Mohammad Modarres, Co-chair
- Dr. Nathan Siu
- Dr. Katrina Groth
- Dr. Gregory Baecher
- Dr. Al-Sheikhly, Dean’s Representative
Title of Dissertation: A Monte Carlo Augmented Bayesian Network Approach to External Flood Probabilistic Risk Assessments
Abstract:
Nuclear power plant (NPP) sites are vulnerable to external flooding, which can impact the safe operation and shutdown of reactors. Flood protection features mitigate flood risks by preventing floodwater from infiltrating areas housing safety-related systems, structures and components (SSCs). Flood-related operating experience has highlighted the frequency and severity of external flood impacts on NPP sites. The current state of practice of analyzing external flood risks is predominately deterministic. However, there is a growing interest in developing a probabilistic risk assessment (PRA) approach to complement the deterministic method to better capture more complex risks such as external floods.
The U.S. nuclear industry employs PRAs to analyze risks by modeling risks to a plant’s operation and ability to safely shutdown. In external flood PRA applications, there are three elements: (1) probabilistic flood hazard assessment, (2) flood fragility evaluation, and (3) plant response. This framework is structured around event trees (ETs) and fault trees (FTs). These tools have supported PRAs for decades. However, the assumptions in the binary state, static treatment of time, and spatial dependencies provide challenges in PRA modeling that are highlighted in external flood applications. Additionally, computational abilities and PRA methods have improved. These tool assumptions and improvements prompted a questionnaire to the international PRA community to understand the needs and trends of performing PRAs and supporting analyses. Given this motivation, there is a research need to explore hybrid external flood PRA frameworks to leverage extensive knowledge in ETs and FTs while strategically integrating tools better suited for external flood modeling.
This dissertation leverages the inherent characteristics of a Bayesian network (BN) to address the limiting assumptions in external flood risk modeling and incorporates questionnaire insights. To mitigate the computational and memory demands that come with BNs, a Monte Carlo simulation integrates out relevant nodes through physical relationships. A hybrid framework then strategically integrates the BN to complement the ETs and FTs. Multiple configurations in which the BN is interfaced with the ETs and FTs are investigated to understand the advantages and limitations of delegating portions of the model to either the BN or ETs/FTs.