Toward Federated Cognitive Digital Twins over the Edge–to-Cloud Continuum

Digital Twins (DTs) are increasingly adopted to monitor, analyze, and optimize Cyber-Physical Systems (CPSs) by enabling continuous interaction between physical assets and their digital representations. Despite their potential, current DT architectures face significant limitations when applied to distributed environments such as smart cities. In particular, most existing solutions rely on centralized and monolithic designs, which introduce latency, scalability, and resilience issues. Morever, they provide limited support for semantic integration and high-level reasoning, which hinders the effectiveness of DTs-based decision-making.Recent research has explored Federated Digital Twins (FDTs) to decompose complex systems into multiple interacting twins, improving scalability and modularity. However, existing FDT approaches often lack a clear architectural framework and still concentrate intelligence in cloud-based components, thus limiting local autonomy. In parallel, Cognitive Digital Twins (CDTs) have been proposed to enhance DTs with semantic reasoning and explainability, leveraging Artificial Intelligence and Large Language Models (LLMs). While promising, these approaches are typically centralized and difficult to integrate within distributed architectures.In this paper, we propose a unified approach that combines federation and cognition within a single architecture, referred to as the Federated Cognitive Digital Twin (FCDT). The proposed architecture distributes intelligence across the edge-to-cloud continuum by introducing local twins deployed close to physical systems and global twins operating at the cloud level. Local twins provide real-time monitoring, analysis, and first-level decision-making, enhanced by lightweight cognitive capabilities; global twins perform system-level reasoning, simulation, and coordination, and leverage more computationally intensive models for explanation and decision support.By combining distributed autonomy at the edge with global cognitive simulation and reasoning in the cloud, the FCDT architecture addresses both structural and semantic limitations of current DT solutions. It enables scalable and responsive DT systems, improves decision-making capabilities, and simplifies the engineering of complex CPSs in distributed environments.