Development of Unknown Risk Scenario Identification System with the Reduced Order Model

Kim, Hyeonmin; Park, Jinkyun

Permalink

https://hdl.handle.net/2142/121817 Copy

Description

Title

Development of Unknown Risk Scenario Identification System with the Reduced Order Model

Author(s)

• Kim, Hyeonmin
• Park, Jinkyun

Issue Date

2023

Keyword(s)

• Deep learning
• Reduced order model
• Dynamic PSA

Abstract

Since the introduction of deep learning techniques, an AI tsunami has emerged, leading many fields to explore their application. In addition, safety assessments of nuclear power plants have also been exploring the use of deep learning. These assessments include two types: deterministic and probabilistic. The probabilistic safety assessment (PSA) encompasses diverse accident scenarios, taking into account safety features and their likelihood of success or failure. The PSA uses the event tree and fault tree method, with the accident scenario defined by an event tree composed of headings and corresponding outcomes of success or failure. To determine the extent of core damage, Thermal-hydraulic (TH) analysis is performed on predefined accident scenarios. However, this process can be a major source of conservatism in conventional PSA. Deep learning techniques offer the potential to reduce conservatism in these assessments by considering a variety of accident scenarios beyond just success or failure, including factors such as timing and action. Although the PSA already takes into account a range of timing and action variables, determining the extent of core damage using TH analysis can be computationally intensive. To mitigate this conservatism, this paper proposes the use of a reduced order model (ROM) developed using deep learning techniques. Reducing the computation time of TH analysis via the ROM model can help identify unknown risk scenarios. The developed systems have been named URiSIS (Unknown Risk Scenario Identification System) and DeBATE (Deep learning Based Accident Trend Estimator), which is a deep learning-based ROM model.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121817

Owning Collections