This paper introduces a framework to automate the creation of Dynamic Master Logic (DML) models by converting technical documentation into Knowledge Graphs (KG-DML). By using Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs), the authors aim to overcome the scalability limitations inherent in manual DML construction, which traditionally requires experts to interpret complex system documentation.
Automating Model Construction
The framework automates the DML hierarchy construction by using targeted retrieval to extract information from system descriptions. This process preserves the functional dependencies and logical relationships between structural elements and functional objectives. By representing these relationships as a Knowledge Graph, the system creates an executable model that can be used for diagnostic reasoning, safety assessments, and tracing dependencies or failure propagation within a system.
Validation and Performance
To ensure the accuracy of the automated models, the researchers implemented a multi-level validation methodology. This approach evaluates the structural integrity of the graph, the consistency of logical gates, and the precision and recall of specific layers within the DML hierarchy. The authors tested this framework on the Low-Pressure Coolant Injection system of a decommissioned Boiling Water Reactor. The results showed that the automated process produced consistent reconstructions across repeated experimental runs.
Scaling Complex Diagnostics
The primary limitation addressed by this research is the reliance on manual expert interpretation, which restricts the ability to model large-scale, complex systems. By extending previous work on small-scale systems to larger, more intricate architectures, the authors provide a method to transform static technical documentation into functional, diagnostic models. This transition allows for more efficient reliability analysis and system diagnostics by leveraging the automated processing capabilities of LLMs.
Franklin Analysis
The evidence suggests that the integration of RAG and LLMs is effective for standardizing the translation of unstructured technical text into structured logic models. Because the framework maintains logical gate consistency and structural integrity during the conversion process, it offers a viable path for automating safety-critical documentation analysis. The consistency observed in the Boiling Water Reactor case study supports the conclusion that this method is reliable for complex system applications.
Comments (0)
to join the discussion
No comments yet
Be the first to share your thoughts!