Session: 15-05-01: Models and Methods for Probabilistic Risk Assessment

Paper Number: 167161

Integrated Bayesian-Based Multi-Hazard Risk Assessment Framework for Nuclear Power Plants 

The Fukushima Daiichi accident underscored the critical need for advanced multi-hazard risk assessment frameworks in nuclear power plants (NPPs), as conventional probabilistic risk assessment (PRA) methods—reliant on fault/event trees—fail to address complex interdependencies between hazards. These methods often treat hazards in isolation, failing to account for scenarios where one hazard can amplify or trigger another, such as earthquakes leading to fires or tsunamis

To address these limitations, this paper introduces an Integrated Bayesian-Based Multi-Hazard Risk Assessment Framework designed to systematically evaluate the risks posed by interacting hazards in NPPs. The framework is applied to a seismic-induced fire scenario in the auxiliary building of a pressurized water reactor (PWR), demonstrating its capability to model hazard interactions, mitigation systems, component damage, and recovery processes.

The framework adopts a layered approach of Bayesian Networks (BNs) to multi-hazard risk assessment, structured into four interconnected layers: hazard interaction, mitigation, hazard exposure, and recovery. The hazard interaction layer  models the dependencies and interactions between hazards. For example, a seismic event can trigger secondary hazards such as fires in electrical cabinets or impair fire suppression and detection systems. Aftershocks, which are modeled using the Epidemic-Type Aftershock Sequence (ETAS) model, further complicate the scenario by delaying recovery efforts and potentially causing additional damage. The ETAS model predicts the number and magnitude of aftershocks based on the mainshock magnitude and the sequence of aftershocks occurring between the mainshock and the mission time. This layer captures the cascading effects of hazards, providing a comprehensive understanding of how one hazard can influence the likelihood and severity of others.

The mitigation layer focuses on the systems and actions designed to suppress or mitigate the impacts of hazards. In the case of a seismic-induced fire, this layer is divided into two sub-layers: automatic suppression (e.g., sprinkler systems) and manual suppression (e.g., fire brigade). The functionality of these systems is influenced by the seismic demand at their locations, as seismic shaking can damage suppression equipment or hinder access for manual intervention.

The hazard exposure layer evaluates the impact of combined hazards on critical components within the NPP. This layer determines whether a component is damaged by assessing whether damage state thresholds are exceeded due to the combined effects of multiple hazards. For instance, seismic shaking may cause structural damage to a component and makes it more venerable to fire damage. The fire scenario is simulated using the Consolidated Model of Fire and Smoke Transport (CFAST) software, which calculates local fire intensity metrics such as heat flux and inner temperature. Uncertainties in fire parameters and the availability of mitigation features are accounted for using Monte Carlo sampling (MCS), ensuring a probabilistic representation of fire progression and suppression effectiveness.

Finally, the recovery layer addresses the post-hazard recovery process, which is critical for restoring the functionality of damaged components and systems. Recovery times are influenced by the damage state caused by each hazard and the occurrence of aftershocks. If an aftershock occurs during the recovery period, it can delay recovery actions, prolonging the time required to restore normal operations.

BNs provide a flexible and intuitive way to represent the dependencies and uncertainties inherent in multi-hazard scenarios, making them well-suited for this application. The framework can also be integrated with traditional PRA methods through a hybrid causal logic approach, where the outputs of the BN are linked to the basic events of fault trees. This allows for a transition to system-level risk quantification, ensuring compatibility with existing PRA practices.

Presenting Author: Akram Batikh North Carolina State University

Presenting Author Biography: Akram Batikh is a PhD candidate in the nuclear engineering department at NC State University. He works in the Probabilistic Risk Assessment group, focusing on multi-hazard risk assessment and external hazards. He received his master’s degree in nuclear engineering from NC State in 2022, working on a transient benchmark for the transient reactor test facility (TREAT). He is currently working on his PhD thesis.

Authors:

Akram Batikh North Carolina State University
Yahya Alzahrani North Carolina State University
Mihai Diaconeasa North Carolina State University