Ontology-Based Knowledge Graph Architecture for AAS-Driven Reasoning in Digital Twins

Abstract

Asset Administration Shells (AASs) provide standardized and semantically rich representations of industrial assets; however, their integration into machine-interpretable knowledge systems that support automated reasoning remains limited. This paper presents an architecture that derives Knowledge Graphs (KGs) from AAS models and applies multi-layer ontology-based reasoning for operational decision support within Digital Twins (DTs). A template-driven mapping transforms incoming AAS submodels into Resource Description Framework (RDF) triples, which are queried using SPARQL Protocol and RDF Query Language (SPARQL) against an ontology stack covering assets, capabilities, and decision rules. A Large Language Model (LLM) facilitates natural language interaction by generating ontology-constrained SPARQL queries and explanations, thereby improving accessibility for non-expert users. The architecture is demonstrated in a manufacturing use case, where it identifies spare tools after tool breakage in a manufacturing cell. Capability constraints, business rules, and availability data are combined to derive feasible alternatives, with the LLM assisting in interpreting candidate trade-offs. Although the case study employs simplified scoring parameters, the results demonstrate that the architecture reduces downtime and facilitates structured, explainable decision-making. The approach highlights the potential of integrating AAS-derived knowledge graphs with layered reasoning and controlled LLM assistance for operational DTs.