Numerous applications are being developed to enhance the energy efficiency of building systems, including fault detection and diagnosis, performance assessment, and intelligent control. For these applications to be effectively utilised, a data connection between the real and virtual worlds must be established. One potential solution to establish this connection and enable semantic enrichment of data with metadata is the semantic digital twin. Semantic digital twins use semantic technologies, such as ontologies, as metadata schemas. However, creating these twins requires substantial manual effort due to the need to examine diverse sources of information about the building systems and normalise this information into a metadata schema. This review investigates whether metadata inference based on time series data from building systems can assist in the automated creation of semantic digital twins. To this end, 53 artificial intelligence-based publications on metadata inference are analyzed for their applicability and efficiency. Three key tasks of metadata inference are examined to create a semantic digital twin: type classification, relation inference, and extraction of operational information. Based on these findings, future research directions are proposed.
