Methodologies to strategize the utilization of modeling and simulation for probabilistic risk assessment (PRA) of nuclear power plants: Applications in fire PRA

Alkhatib, Sari

Permalink

https://hdl.handle.net/2142/127443 Copy

Description

Title

Methodologies to strategize the utilization of modeling and simulation for probabilistic risk assessment (PRA) of nuclear power plants: Applications in fire PRA

Author(s)

Alkhatib, Sari

Issue Date

2024-10-08

Director of Research (if dissertation) or Advisor (if thesis)

Mohaghegh, Zahra

Doctoral Committee Chair(s)

Mohaghegh, Zahra

Committee Member(s)

• Aven, Terje
• Barkan, Christopher P.L.
• Sakurahara, Tatsuya
• Uddin, Rizwan
• Xi, Jianqi

Department of Study

Nuclear, Plasma, & Rad Engr

Discipline

Nuclear, Plasma, Radiolgc Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Degree of Realism
• Modeling and Simulation
• Nuclear Power Plants (NPPs)
• Probabilistic Risk Assessment (PRA)
• Screening Analysis

Abstract

Based on the literature, one way to conceptualize risk is: Risk={A,C,P_f^*,P(P_f^* ) | U^',K}, where A represents events including initiating events and scenarios; C denotes the consequences of these events (A); P_f^* stands for the estimated frequency of occurrences of events A; P(P_f^* ) is a probability distribution that characterizes the uncertainty in the estimated frequency P_f^*; U^' covers uncertainties about events A and consequences C that are not captured by probabilistic calculations of P_f^* and P(P_f^* ); and K includes background knowledge that underlies the estimation of P_f^* and its probability distribution P(P_f^* ). This conceptualization is adopted for this research to effectively communicate the scientific methodological contributions of this thesis. This research investigates constituent risk elements {A,C,P_f^*,P(P_f^* ) | U^',K} within the framework of probabilistic risk assessment (PRA), which plays a crucial role in ensuring the safety and operational efficiency of nuclear power plants (NPPs). PRA quantitatively assesses safety margins and identifies risk-contributing factors, supporting risk-informed decision-making of both the nuclear industry and the regulators. In PRA, the calculation of P_f^* and P(P_f^* ) for A and C involves leveraging background knowledge (K) that can be derived from various sources, including data-driven statistical analyses, modeling and simulation (M&S) of underlying phenomena, expert elicitation, or hybrid approaches that integrate these methods. Presently, there is a growing reliance on M&S in the PRA of NPPs for several reasons: (i) having limited empirical data for advanced reactors, (ii) questioning the relevance of historical data for aging plants due to design changes or programmatic shifts, (iii) needing detailed risk information at the level of underlying phenomena to support risk-informed design and enhance the economic viability of NPPs, and (iv) the importance of enhancing the realism of risk estimation to address emergent safety concerns while adhering to regulatory standards. Given the resource-intensive nature of M&S in PRA, analysts are constrained in their ability to conduct M&S for all PRA events. Therefore, it is crucial for them to carefully select PRA events and determine the appropriate degree of realism for M&S applications. These choices, currently made under screening analysis in PRA, significantly influence the background knowledge (K) supporting PRA and introduce associated uncertainties (U^'). This research has developed methodologies that enhance the screening analysis to strategize the utilization of M&S for PRA of NPPs, making three scientific contributions: Developing a systematic decision-making methodology to formalize the selection of degree of realism in screening analysis. The current screening analysis methods rely on prescriptive procedures and trial-and-error methods to gradually refine simulation models. This research develops a methodology that formalizes the selection of the degree of realism for simulation models as a multi-criteria decision problem. It considers two primary decision attributes: the anticipated impact on the plant risk estimates and the projected resources needed for the analysis. This decision-making process involves determining the appropriate level of background knowledge (K) that is pertinent to the system under analysis in the screening analysis, considering the influence of the resulting unquantified uncertainty (U^') on risk information and the resources required to achieve that level of K. This research is the first to propose a predictive approach for estimating these two decision attributes across multiple degrees of realism alternatives. In literature, the impacts of chosen degrees of realism on PRA outputs are usually analyzed after the baseline risk quantification, as demonstrated in methods like model uncertainty or the “strength of knowledge.” There is also literature that calculates generic estimates of the PRA resource requirement without explicitly assessing the impact of multiple alternatives for the degrees of realism on the required resources. Based on a literature review, no study has quantitatively analyzed the tradeoff between the impact on the safety risk estimates and the required resources for conducting the analysis associated with the PRA of NPPs. The applicability of the proposed methodology is shown using a case study of the ignition source screening in an NPP fire compartment, demonstrating the fire model selection from two alternatives with different degrees of realism: an engineering correlation and a two-zone model. Developing a new methodology based on phenomenological nondimensional parameter (PNP) decomposition to generate surrogate values for input parameters of M&S, supporting screening analysis. The data collection process for input parameters of M&S can be a significant resource burden. In current screening analyses, this challenge is mitigated by using conservative values for individual input parameters, potentially leading to an excessive risk overestimation. This research offers a novel way of using the limited background knowledge (K) that could reduce conservatism (i.e., unquantified uncertainty, U^') without significant investment in the refinement of K. The proposed new methodology groups multiple input quantities of simulation models using a PNP that represents the underlying physical phenomena. By considering a scientifically justifiable range for the PNP, this methodology decomposes the PNP and generates surrogate values to constitute the physical input parameters of the simulation models. This approach helps minimize the necessity to acquire precise values for input parameters, thus circumventing extensive data collection and extraction efforts. This thesis also develops a computational platform equipped with the PNP decomposition to facilitate the fire PRA of NPPs. The feasibility of the PNP decomposition methodology and the computational platform are demonstrated through a multi-compartment fire case study at an NPP. Developing a novel approach to guide and scientifically justify the formulation of modeling assumptions in screening analysis. Justifying the modeling assumptions in screening analysis can be challenging, especially when the industry-wide consensus method is not well established or previous PRA experiences are limited, for instance, when analyzing a new reactor design or introducing an advanced M&S tool. This research offers a new approach that helps select a set of proper modeling assumptions that minimize the potential negative outcome (namely, a false negative result in the screening analysis) of the unquantified uncertainty (U^') within the constraint of the current state of background knowledge (K). This thesis harnesses recent rapid advancements in computational capabilities to conduct sensitivity analysis aimed at informing the generation of modeling assumptions during screening analysis. This approach diverges from current PRA practices, where sensitivity analysis is conducted as part of the post-processing of PRA output and after decisions regarding modeling assumptions have already been made during the screening analysis. The proposed proactive approach provides scientific justifications for establishing appropriate modeling assumptions during the screening analysis phase. This research (i) develops a methodological procedure to guide the formulation of modeling assumptions by combining uncertainty quantification and sensitivity analysis methods, (ii) integrates this methodological procedure with the aforementioned PNP decomposition approach to strengthen the justification for deriving surrogate values for physical input parameters in M&S, thereby supporting the screening analysis of PRA, (iii) incorporates these advancements into the fire PRA computational platform introduced under Contribution ‘B’ of this thesis, and (iv) applies the advanced computational platform to a multi-compartment fire case study.

Graduation Semester

2024-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/127443

Copyright and License Information

Copyright 2024 Sari Alkhatib

Owning Collections