Bridging the Physical and Virtual: Automated Thematic Classification of IoT Sensors for Urban Digital Twin Catalogs

Urban Digital Twins (UDTs) rely on IoT sensors to bridge physical and virtual domains, but deployments with thousands of sensors face significant challenges in device information discoverability. Metadata catalogs address this by registering and describing sensors with standardized metadata, enabling stakeholders to efficiently discover and access available sensing services. However, current sensor registration to metadata catalogs relies on time-consuming, error-prone manual processes, creating stale entries when services change over time. Overcoming these manual registration challenges requires automated approaches that can maintain high-quality metadata and organize sensors into effective groups for enhanced discoverability. This paper introduces the WRENCH framework, a modular end-to-end framework for sensor device registration. The framework enables automated IoT sensor registration into metadata catalogs using Harvesters, Groupers, MetadataEnrichers, and Catalogers as key components. It provides scheduling and state management for continuous data harvesting and utilizes large language models to generate descriptive titles, improving the discoverability of catalog entries. Our framework features a novel clustering algorithm for IoT devices. KINETIC is implemented as a Grouper within the framework and utilizes various natural language processing techniques, including keyword extraction, co-occurrence networks, and community detection, to cluster sensor information documents. Unlike traditional methods, such as LDA (Latent Dirichlet Allocation), that fail with heterogeneous sensor data, KINETIC identifies underrepresented clusters while maintaining semantic coherence, demonstrating superior performance. The framework addresses central challenges in urban IoT metadata management and facilitates the growing utilization of digital twinning technologies through continuous and automated data exploration and registration.