Asset Administration Shell for Real Digital Twins

The Asset Administration Shell (AAS) is the Industry 4.0 standard that connects physical assets with their digital representations. Several open-source tools implement it, and choosing between them is a practical obstacle — particularly for small and medium-sized enterprises.

This study compares the four actively maintained tools — AASX Server, Eclipse BaSyx, FA³ST Service and NOVAAS — against criteria including ease of deployment, metamodel V3 compliance, real-time data monitoring, user interface and industrial suitability, together with Docker deployment benchmarks (build time, image size, start-up time, CPU and memory).

The tools were then applied to a real case: digital twins of the recovery absorption and stripping columns of an ethylene oxide / ethylene glycol (EO/EG) plant. Flow, pressure, temperature and pH measurements were streamed through MQTT and Telegraf into InfluxDB and served by a reactive (Type 2) AAS with Digital Nameplate, Technical Data and Time Series submodels.

Eclipse BaSyx proved the most suitable for this case thanks to its middleware architecture, plugin extensibility, time-series visualisation, mobile support and documentation; AASX Server suits low-resource assets, FA³ST edge synchronisation and NOVAAS event-driven configurations. The study argues that future AAS should evolve from passive data containers into proactive, learning "cognitive engines".

A follow-up paper at MobiWIS 2025 (LNCS) extends Eclipse BaSyx with a custom anomaly-detection plugin that reads its parameters from the AASX semantic metadata. It polls InfluxDB every three seconds and offers two unsupervised methods: Median Absolute Deviation as a metadata-free default, and Mahalanobis distance using a mean vector, inverse covariance matrix and threshold stored in the AASX file. On a labelled test stream (temperature, humidity, air quality), MAD reached F1-scores of 0.84–0.92 per sensor and Mahalanobis distance reached precision 0.968, recall 1.000 and F1 0.984.