The operation of cyber-physical systems, such as research vehicles in a field, generates extensive volumes of heterogeneous data from various sensors. Field environments demonstrate variability due to real-world factors and weather conditions, requiring the integration of both raw sensor data and environmental data for comprehensive analysis. To enhance the usability of collected data, derived data generations, such as brightness-adjusted images, object detection results, and vibration analysis outputs, are essential. These datasets provide valuable insights and extend the applicability of raw data for other researchers. Researchers are not only interested in raw data but also in derived datasets, such as images where object detection has already been applied. Access to both raw and derived datasets prevents redundant efforts and increases the reusability of existing data.

However, identifying datasets relevant to specific research objectives remains challenging due to the volume, heterogeneity, and varying quality of field datasets. The presence of diverse data formats and conditions necessitates systematic selection methods. Ensuring that relevant datasets can be discovered and reused requires adherence to FAIR data principles (Findable, Accessible, Interoperable, and Reusable). Digital Twins, through systematic integration and organization of heterogeneous operational data from the physical entity, enhance research field data management and support data-driven decision-making by enabling dataset identification, filtering, and ranking.

This paper introduces the Dataset Selection Assistant (DSA), a Decision Support System (DSS) within a Digital Twin framework for managing data collected by cyber-physical systems operating in a field. The DSA leverages Natural Language Processing (NLP) to interpret user queries, applying filtering mechanisms based on predefined conditions (e.g., environmental parameters) and employing Multi-Criteria Decision Analysis (MCDA) to retrieve and rank datasets based on user-defined preferences (e.g., recency or dataset size). The functionality and potential of the proposed DSA are demonstrated through its application in an automotive use case.