Digital Twin for Field Data Management: Text-Based Scenario Descriptions for Knowledge Extraction from Sensor Data

The reuse of data from field-deployed cyber-physical systems (CPS) in research and engineering requires structured methods for interpreting and analyzing field experiments. Raw sensor measurements are often stored without contextual information, limiting interpretability and reducing data reusability. Within a digital twin for field data management, text-based scenario descriptions can provide an additional semantic layer that enables the sharing of data following the FAIR (Findable, Accessible, Interoperable, and Reusable) principles beyond raw measurements alone. This paper presents a method for generating textual descriptions of field experiment scenarios. A modular data pipeline is implemented to extract knowledge from raw sensor inputs by identifying scenario-relevant features, including spatial, dynamic, and visual characteristics, which are then transformed into textual descriptions using large language models (LLMs). As a use case, the method is applied to research vehicle test drives using data from Global Positioning System (GPS), inertial measurement units (IMU), and camera sensors, enriched with external application programming interfaces (APIs). The generated texts are evaluated using reference-based and reference-free metrics to assess their consistency with the underlying data. By complementing raw sensor data with textual descriptions, the proposed approach supports more efficient dataset sharing and retrieval and facilitates decision-making in research activities supported by digital twin frameworks.