SwarmFTM: An Intelligent Fault Tree Modeling System for Nuclear Power Plants Based on Multi-Agent Technology
Fault tree modeling is a crucial component of fault tree theory. However, traditional fault tree modeling methods have limitations in nuclear power applications, such as low modeling efficiency and a high rate of human error. This paper proposes an intelligent fault tree modeling system, SwarmFTM, which integrates multiple types of large language models (LLMs) and multi-agent technology. The system features four intelligent agents with distinct functions—observation, retrieval, reasoning, and construction—combining the visual parsing and logical reasoning capabilities of LLMs with the collaborative division of labor among agents to successfully achieve fully automated modeling, from the parsing of industrial system equipment diagrams to fault tree generation.Based on three common systems in nuclear power plants, the SwarmFTM fault tree modeling results are comprehensively evaluated using a four-dimensional EWM (entropy weight method)–TOPSIS (technique for order preference by similarity to an ideal solution) quantitative evaluation framework. The evaluation results show that SwarmFTM reliably extracts the underlying physical equipment and reproduces the fault tree structure, achieving a quality closeness to the manual reference of up to 0.886 on systems of low to moderate complexity, while reducing the modeling time from the order of days to the order of minutes.
Authors
- Ming Ding (ORCID: https://orcid.org/0000-0002-6376-3174)
- Xiaowei Chen
Institutions
- Harbin Engineering University (CN)
Publication Details
- Journal
- Nuclear Science and Engineering
- Published
- 2026-11-20
- DOI
- https://doi.org/10.1080/00295639.2026.2750780
- Primary Topic
- Risk and Safety Analysis
- Type
- article
- Field-Weighted Citation Impact
- 0.00