Automated Construction of Knowledge Graph-Based Dynamic Master Logic Models Using LLMs and RAG

What changed

This study presents a novel framework for the automated construction of Dynamic Master Logic (DML) models, representing them as Knowledge Graphs (KG-DML). Traditional DML construction relies heavily on manual expert interpretation of technical documentation, which is a bottleneck for complex systems. The proposed framework utilizes Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs) to automate this process, enabling the construction and evaluation of KG-DML for substantially larger and more complex systems than previously possible.

Why it matters for builders

For AI builders and engineers, this research provides a pathway to automate the generation of functional system models from raw technical documentation. This automation can reduce the manual effort and expertise required, making diagnostic reasoning, safety assessments, and failure propagation analysis more scalable and efficient. The KG-DML output supports detailed tracing of dependencies and functional relationships.

Practical impact

The framework constructs KG-DML by preserving functional dependencies and logical relationships across the DML hierarchy through targeted retrieval. The resulting KG-DML can be used for diagnostic reasoning, safety assessment, and tracing failures upwards or dependencies downwards. A multi-level validation methodology is employed to assess precision, recall, logical consistency, and structural integrity. The framework was successfully applied to the Low-Pressure Coolant Injection system of a decommissioned Boiling Water Reactor, demonstrating consistent reconstruction across multiple runs.

Caveats and source limits

The source is a research paper detailing a proposed framework and its application. While it demonstrates successful reconstruction on a specific system, further validation and broader application across diverse complex systems would be beneficial. The paper does not provide specific benchmark numbers for performance improvements over manual methods or details on the computational resources required for the automated construction process.