CARE data model

Project data 

Short description

While the creation of holistic digital twins continues to be the focus of research, their integration in practical workflows although potentially highly beneficial is often hindered due to heterogeneous and multi-modal data representations, volumes and formats. A new, holistic approach to data management and provision (CARE-DS, CARE-DIVE) is therefore to be developed, which will form the backbone of the CARE data model and thus the basis for data analysis and synthesis within and beyond CARE. To achieve this goal, this project proposes combining a symbolic spatial knowledge graph (a GeoSPARQL-enabled RDF store) with a sub-symbolic vector database resulting in a hybrid query layer as a methodological core.

Here, both parts compensate for the weakness of the other: a symbolic query allows for precise and transparent information retrieval that leverages both semantic and spatial information. Sub-symbolic approaches, on the other hand, rely on recognizing patterns (or deviations from usual patterns), allowing them to query ill-structured or incomplete data. This makes them particularly effective for unstructured sources such as site photos, sensor readings, design drawings, or point cloud segments. Additional synergy potential between the two data indexing methods is the management of vocabulary mismatches between domain ontologies, where semantic similarity and transformer models can be developed to infer synonyms at a schema level, lowering the threshold for the integration of heterogenous data within a holistic ontology.