A Multimodal Hybrid Retrieval Framework for Digital Twins in Civil and Structural Engineering

Abstract: Digital twin adoption is hindered by fragmented symbolic and subsymbolic construction data from heterogeneous sources. We propose a hybrid knowledge‑graph and vector‑database architecture that integrates a GeoSPARQL‑enabled RDF store with semantic embeddings to unify technical documentation and support both precise reasoning and fuzzy retrieval. A hybrid RAG workflow on PDF‑based data demonstrates improved ranking quality and contextual relevance compared to single‑paradigm retrieval. This lays the groundwork for future work on infrastructure-focused digital twins. Comprehensive support for multimodal data and case studies, such as minimally invasive bridge strengthening and scaffold‑free robotic assembly, is left as a planned continuation of the research.