Automate extraction and updating of ontology towers from unstructured operational records

Develop an automated process that extracts ontology towers for multiple similar industrial equipment systems from equipment repositories and incorporates relevant operating-journal entries, thereby structuring context that exists only as unstructured records.

Background

The study evaluates one low-humidity air-handling system, but deployment on an industrial site would involve many similar systems with system-specific details and increasingly large operating journals. The proposed tower architecture allows common knowledge to be shared across systems while retaining instance-specific nodes.

Scaling the approach requires automated extraction of towers and automated incorporation of journal entries. The paper identifies these as unresolved steps because manually constructing a tower for every system would not improve on manually preparing the projected text, and because the relevant operational context is often present only in unstructured records.

References

Two steps decide whether this holds: the extraction of towers from equipment repositories must be automated, since a tower written by hand for each system is no better than the text it replaces, and the incorporation of journal entries must become part of the extraction. Both are instances of one problem that any form of supply faces, how context that exists only as unstructured records is to be structured.

— Narrow and Deep: An Ontology Tower as the Knowledge of an LLM Agent for an Industrial Equipment System  (2610.11768 - Joo et al., 8 Oct 2026) in Section 4.4, “Extension to many similar systems”; reiterated in Section 5, Conclusion

This is the case the review queue, drift proposals and vocabulary editor are designed for, but their effectiveness has not yet been measured.

— Plant Memory: Technical Language Processing and a Verified Maintenance Ontology for Agent-Accessible Failure Knowledge from SAP PM Work Orders  (2610.19131 - Raj, 13 Oct 2026) in Section 6, Discussion and limitations