Metadata-, Ontology- and LLM-based Data Provenance Collection in an Aircraft Digital Twin

Dies ist die aktuellste Version dieses Eintrags.

Kurzfassung

Recent years have seen much research and development in the area of Digital Twins (DT), in Aviation, for instance, Aircraft DTs (ADT). Most DT definitions include three main elements: a particular (physical) entity, its lifecycle, and software that allows to store and use data about the entity throughout its lifecycle. Given such scope, DTs are characterized by high-rate flows of heterogenous data, which raise the question of how to assess the trustworthiness of such data in order to ensure accuracy and reliability. One aspect of trustworthiness is data provenance, which may be subject to a variety of approaches. Ecosystems that require user and data heterogeneity (e.g. research or design data management systems) may profit from data provenance pipelines that include semantic services, because ontology-based data provenance provides a flexible way to track and enrich data by linking it with explicit relationships, which allow to analyze differences and resolve inconsistencies between and across datasets generated by the (physical) entity. In this context, the architecture of a Data Provenance and Metadata Demonstrator (DPMD) for an ADT is presented. DPDM is a python/SPARQL/owlready pipeline for metadata structuring coupled with an example ontology for provenance. The pipeline spans four phases from metadata to provenance (Data Collection, Metadata Extraction, Provenance Collection, Provenance Analytics). Most implementation activities have so far focused on the first three phases, with the interplay between the python script and the ontology occurring in Provenance Collection. This prototype has been further expanded to cope with robustness issues caused by the nesting of metadata and by mismatches between incoming metadata and the demonstrator's knowledge base. The first issue has been tackled through a recursive version of the knowledge construction algorithm; the second issue has been tackled by including in the algorithm Natural Language Processing (NLP) techniques, such as calls to Large Language Models (LLM).

Verfügbare Versionen dieses Eintrags

• Metadata-, Ontology- and LLM-based Data Provenance Collection in an Aircraft Digital Twin. (deposited 26 Feb 2026 16:01) [Gegenwärtig angezeigt]