Time series-based metadata extraction for the creation of semantic digital twins of building systems

Applying fault detection in building systems requires mapping building data to a standardized data model to ensure interoperability. One approach to achieving this is the creation of semantic digital twins, which encapsulate and structure the data. However, generating these twins often involves significant manual effort for metadata extraction. This paper presents a methodology for automating metadata extraction from building system time series data to facilitate creating semantic digital twins. The approach consists of three key steps: data point classification, data point grouping, and topology detection. The methodology was evaluated using the Brick schema as the data model and the Mortar dataset as the ground truth data source. The performance of the algorithms indicates that further improvements are needed to enable practical application, as current accuracy levels remain below 65% in all key steps. However, this study establishes a comprehensive methodology for digital twin-focused metadata extraction and identifies challenges associated with time series-based approaches, providing a foundation for future research.