Development of Probabilistic Risk Assessment Methodology Using Artificial Intelligence: 2. Automatic Fault Detection Method for Building Reliability Database Technology

Ujita, Hiroshi; Morimoto, Tatsuya; Futagami, Satoshi; Yamano, Hidemasa; Kurisaka, Kenichi

Permalink

https://hdl.handle.net/2142/121826 Copy

Description

Title

Development of Probabilistic Risk Assessment Methodology Using Artificial Intelligence: 2. Automatic Fault Detection Method for Building Reliability Database Technology

Author(s)

• Ujita, Hiroshi
• Morimoto, Tatsuya
• Futagami, Satoshi
• Yamano, Hidemasa
• Kurisaka, Kenichi

Issue Date

2023

Keyword(s)

• Reliability database
• Text mining
• Data mining
• Rule-based approach
• Fault detection

Abstract

For the construction of reliability databases in Japan, the Central Research Institute of Electric Power Industry and the Japan Nuclear Safety Institute have been developing the NUCIA database for light water reactors, and the Japan Atomic Energy Agency has been developing the CORDS database for sodium-cooled fast reactors. These databases allow us to reference individual cases and support analysis through database searches and have also been used for creating failure rate databases. However, challenges exist due to the vast amount of data requiring significant human resources for analysis, and the potential for variability and bias in analysis due to the involvement of multiple personnel (primary analysis). Furthermore, it is practically impossible to extract characteristics of temporal and spatial events to gain an overall understanding of the massive amount of data (secondary analysis). Therefore, by utilizing artificial intelligence (AI) technologies, it is believed that the following solutions against the above challenges can be addressed and resolved, with the greatest expectation being the possibility of discovering new insights that may not be discernible by humans. Challenge 1: Improvement of efficiency in building reliability databases (primary analysis) • Acceleration and automation of processing, • Promotion of accuracy and uniformity (elimination of individual differences among analysts). Challenge 2: Improvement of analysis capability for extracting temporal and spatial characteristics (secondary analysis) • Enhancement of uniformity, • Acquisition of new insights (Extraction of common factors and changes over time through automatic extraction of temporal and spatial characteristics using big data processing). The methodology will be developed in a three-year plan with the following steps: Establish and prototype the methodology for extracting failure occurrence locations (system/equipment), failure modes, and causes from databases such as NUCIA and CORDS, and transforming them into a database using AI technologies. Prototype a method for analyzing differences in events and time using AI technologies, enabling differentiation based on factors such as power company, plants, operation/maintenance, and mechanical/electrical disciplines, as well as differences before and after the Fukushima Daiichi accident. Prototype a method for determining common factors such as common equipment, common operations, and common organizational characteristics through analysis using AI technologies. Apply these prototyped methods to create reliability data using existing maintenance information, verify their validity, and identify future development challenges. In the paper, the results of Step 1 are described, presenting the overview of the overall development as follows: Acquisition of each event (failure) report from database sites' information. Removal of unnecessary information based on a rule-based approach. Text cleaning using a rule-based approach and the functions of natural language processing framework. Summarization using embedding method and graph-based algorithm. Statistical analysis. Zero-shot text classification as an unsupervised multi-label classification problem.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121826

Sponsor(s)/Grant Number(s)

MEXT Inoovative Nuclear Research and Development Program Grant Number JPMXD0222682583

Owning Collections