Enabling infrastructure subsystems integration via agentic ontology engineering

Semantic harmonisation of heterogeneous infrastructure information systems is fundamental to data and system integration for infrastructure asset management (IAM). However, manual knowledge engineering over complex organisational databases is labour-intensive. This paper presents an agentic ontology engineering method that uses large language models (LLMs) within a multi-agent framework to automatically extract knowledge and construct ontologies from heterogeneous IAM databases. The system progressively generates classes, hierarchies, relations and logical restrictions, while agents align cross-system concepts and detect contradictions to ensure consistency. Evaluated on four real-world IAM schema sets, the method achieves 92.1% entity recall against a ground truth ontology, 90.91% entity classification accuracy, and 83.18% relation classification accuracy. Compared with a single-LLM baseline, the multi-agent approach yields higher quality ontologies with fewer constraint violations. The resulting IAM ontology supports interoperability across subsystems, demonstrating the potential of an agentic, LLM-driven approach to reduce modelling effort while improving ontology fidelity for infrastructure information management.