Functional requirements to enhance the traceability of a deep-learning based reduced order model in PSA applications

Park, Jinkyun; Kim, Hyeonmin

Permalink

https://hdl.handle.net/2142/121834 Copy

Description

Title

Functional requirements to enhance the traceability of a deep-learning based reduced order model in PSA applications

Author(s)

• Park, Jinkyun
• Kim, Hyeonmin

Issue Date

2023

Keyword(s)

• Probabilistic safety assessment
• Nuclear power plants
• Accident scenario identification
• Deep learning
• Reduced order model
• Functional requirements

Abstract

There is no objection to the fact that the technique of probabilistic safety/risk assessment (PSA/PRA) is critical for visualizing the safety of nuclear power plants (NPPs). One of the important contributions expected from the PRA technique is the identification of accident scenarios that can result in the occurrence of unexpected events such as the core damage or the release of radioactive materials to the environment. That is, it is strongly anticipated that the more the accident scenarios are identified the more the safety of NPPs will enhanced by providing relevant countermeasures preventing the occurrence of the unexpected events. To this end, it is indispensable to overcome the curse of thermal-hydraulic (TH) code runs because it is extremely rare to get necessary information from existing operation experience. In other words, the evaluation of end states with respect to promising accident scenarios should be carried by using the a precise TH code that requires a lot of computational resources. The number of TH code runs to explore all of the promising accident scenarios, however, is generally too large to be simulated with the precise TH code (e.g., the state explosion problem). For this reason, the use of a reduced order model (ROM) can be regarded as a resolution of the state explosion problem. At the same time, the use of the ROM creates a critical issue in terms of the traceability defined as: “If the sequences of events or system trajectories can be fully traced from an initial condition to end-states (i.e., accident events, undesirable consequences)” [1]. Accordingly, from the perspective of practicality, it is necessary to develop a traceable ROM that can resolve two crucial issues hampering the identification of accident scenarios: (1) the state explosion problem and (2) the traceability. In this regard, this paper elaborates (1) human responses that is one of the main factors causing the state explosion problem and (2) an idea addressing how to resolve the traceability issue with a ROM

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121834

Owning Collections