Generative Artificial Intelligence for Underground Utility Digital Twins: A Review and Task-Oriented Framework

Résumé fourni par la source

Underground Utility Digital Twins (UUDTs) are increasingly recognized as critical components of Urban Digital Twins (UDTs), yet their development remains constrained by fragmented records, incomplete geometry, semantic inconsistencies, and governance barriers. Unlike above ground assets, underground infrastructures are characterized by sparse observations, missing depth information, heterogeneous data ownership, and high safety implications, making reliable digital representation particularly challenging. Existing standards and sensing technologies improve data quality and detection, yet they do not provide mechanisms for reconstructing or inferring undocumented assets and missing geometric attributes. This paper presents a comprehensive review and conceptual synthesis of how Generative Artificial Intelligence (GenAI) can address these structural limitations and provide a promising foundation for advancing the development and maintenance of UUDTs. We first review the core data, geometric, semantic, and governance challenges that undermine current Urban Digital Twin (UDT) implementations. We then systematically map major generative paradigms, including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), diffusion models, transformer-based architectures, and multimodal generative frameworks, to specific underground modeling tasks such as sensor data synthesis, geometry and topology completion, semantic enrichment, multi-source data fusion, and uncertainty quantification. Building on this synthesis, we propose a task-oriented integration framework for embedding generative modules within urban digital twin workflows, with the aim of supporting probabilistic reconstruction, continuous updating, and risk-aware decision support. By reframing underground infrastructure modeling as a probabilistic inference problem rather than a deterministic data integration exercise, this review establishes a conceptual foundation and research roadmap for next-generation UUDTs that are adaptive, uncertainty-aware, and aligned with the safety and resilience requirements of smart cities.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.