Automated KG-DML Construction for Complex System Diagnostics

What changed

This study introduces a novel framework for the automated construction of Dynamic Master Logic (DML) models, representing them as Knowledge Graphs (KG-DML). DML models typically provide a hierarchical structure to depict system behavior by connecting functional objectives with their underlying structural components. Historically, the creation of these models has been a labor-intensive process, heavily reliant on expert interpretation of technical documentation, which poses limitations for scalability when dealing with highly complex systems.

The presented framework extends prior research on small-scale systems to enable automated KG-DML construction and evaluation for substantially larger and more intricate systems. The model construction process operates across the DML hierarchy, employing targeted retrieval mechanisms to maintain functional dependencies and explicit logical relationships. The resulting KG-DML outputs are designed to support a range of analytical tasks, including diagnostic reasoning, safety assessments, upward failure propagation analysis, and downward dependency tracing.

To ensure the quality and integrity of the generated models, a multi-level validation methodology is employed. This methodology assesses layer-specific precision and recall, the consistency of logical gates, and the overall structural integrity of the knowledge graph. The framework's applicability was demonstrated through an application to the Low-Pressure Coolant Injection system of a decommissioned Boiling Water Reactor. This application showed consistent reconstruction of the DML model across multiple execution runs.

Why it matters for builders

For AI builders and system engineers, this research presents a significant advancement in automating the creation of complex system models. The ability to automatically transform unstructured technical documentation into structured, executable functional models using LLMs and RAG can drastically reduce the manual effort and expertise required. This opens up possibilities for more efficient diagnostic and reliability analysis, especially for systems where documentation is extensive or expert knowledge is scarce.

Builders can explore how to integrate such frameworks into their workflows for system design, maintenance, and safety analysis. The methodology suggests a pathway to create more robust and scalable diagnostic tools by leveraging modern AI techniques to interpret and structure complex technical information.

Practical impact

The framework's core contribution is the automation of KG-DML construction, enabling the conversion of technical documentation into executable functional models. This has direct implications for diagnostic and reliability analysis. Builders can investigate the potential of using similar RAG and LLM-based approaches to process their own system documentation. The validation methodology described, focusing on precision, recall, and logical consistency, provides a blueprint for evaluating the quality of AI-generated system models.

The successful application to a real-world system like the Low-Pressure Coolant Injection system of a Boiling Water Reactor suggests that this approach is viable for complex industrial applications. Builders involved in critical infrastructure or complex machinery might find this method valuable for enhancing their diagnostic capabilities.

Caveats and source limits

The source material is a research paper detailing a framework and its application. Specific implementation details, such as the exact LLMs and RAG techniques used, the performance benchmarks beyond the qualitative consistency observed in the reactor system example, and the computational resources required, are not fully elaborated. The paper focuses on the methodology and its potential, rather than providing a ready-to-deploy tool or library. Further details on the scalability limits and the types of technical documentation most amenable to this automated construction process would be beneficial for practical adoption.