Enhancing Nuclear PRA with Large Language Models: Integrating Artificial Intelligence into SAPHIRE for Advanced Analysis and Synthesis of Results

Ball, Erick

Permalink

https://hdl.handle.net/2142/121806 Copy

Description

Title

Enhancing Nuclear PRA with Large Language Models: Integrating Artificial Intelligence into SAPHIRE for Advanced Analysis and Synthesis of Results

Author(s)

Ball, Erick

Issue Date

2023

Keyword(s)

• Large language model
• Probalistic Risk Assessment (PRA)
• SAPHIRE

Abstract

The Nuclear Regulatory Commission's (NRC) probabilistic risk assessment (PRA) software tool, SAPHIRE, is a crucial instrument in evaluating the safety and reliability of nuclear power plants, but the usefulness of PRA can be limited by the time and expertise required to get meaningful results. This study proposes a new approach to integrating a large language model (LLM) into SAPHIRE, aimed at improving the efficiency and comprehensibility of PRA analyses. By incorporating natural language processing capabilities for the PRA model, users can describe their requirements in a human-readable format, while the LLM creates macros to perform the analysis, generates reports, and extracts high-level insights. The core of this project lies in identifying the optimal data inputs and prompts to enable seamless LLM- SAPHIRE interaction. We will examine the accuracy of the results by comparing the LLM-generated conclusions with expert evaluations, and evaluate a self-criticism mechanism for improving accuracy. Our innovative approach seeks to unlock the full potential of AI in PRA, streamlining the analysis process and providing comprehensive insights into high-frequency cut sets and sequences in natural language. By incorporating state-of-the-art AI systems into nuclear risk assessments, this research aims to establish a new paradigm for making risk analysis more accessible and efficient for stakeholders and regulators.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121806

Owning Collections