Conventional equipment reliability analysis in nuclear power plants relies heavily on the subjective experience of engineers. Furthermore, multi-source heterogeneous data—such as equipment diagrams, system functional descriptions, and reliability reports—are typically siloed across disparate departments in unstructured formats. This fragmentation leads to inefficient data processing, poor information integration, subjective assessments, and decision chains that lack traceability, ultimately failing to meet increasingly stringent regulatory demands for analytical transparency and evidence integrity. To address these deficiencies, this paper proposes an automated failure mode and effects analysis (FMEA) and fault tree analysis (FTA) system for nuclear power plant equipment, driven by large language models (LLMs). The core of this system consists of LLM-powered intelligent agents. Guided by a domain-specific knowledge graph and rule-based prompt constraints, these agents transform unstructured, multi-source data into standardized, structured tables suitable for safety modeling, while automatically generating equipment correlation matrices. Building on this foundation, the conventional seven-step FMEA procedure is streamlined into a cohesive, end-to-end workflow. This encompasses equipment information extraction, functional state definition, failure mode matching, risk assessment, mitigation strategy formulation, and result documentation, thereby enabling the automated construction of both FMEA and FTA models. The system architecture comprises three primary modules: automated FMEA modeling, knowledge base maintenance, and probabilistic safety assessment (PSA) integration. Seamlessly interfacing with existing PSA platforms and automated modeling tools, it establishes a comprehensive pipeline from raw data ingestion to intelligent analysis and model generation. Preliminary application results demonstrate that the system significantly reduces manual intervention and analysis cycles while enhancing the accuracy, completeness, and traceability of nuclear safety analyses. Consequently, it provides robust, data-driven decision support for full-lifecycle equipment management and proactive risk early warning.

nuclear power plant equipment management,large language model; intelligent agent,failure mode and effects analysis,fault tree analysis,knowledge graph