Towards Federated, Autonomous and Cognitive Digital Twins with DARLING

Abstract

With the widespread adoption of Digital Twins (DTs) in research and industry, this technology is emerging as a key decision-making tool in Industry 4.0. Advancements have expanded its applications across manufacturing, healthcare, aerospace, and robotics. Open and enterprise platforms now simplify deployment, enabling asset management, communication, IoT integration, data aggregation, visualisation, simulation, and real-time updates to physical counterparts. These improvements enhance efficiency, reduce costs, and support sustainability. Despite these benefits, existing DT platforms provide limited support for AI-driven reasoning in distributed environments. Simulation analysis and decision-making typically occur after data consolidation within a single component, constraining scalability and adaptability. Federated Learning partially addresses this by enabling decentralised model training but lacks continuous reasoning and dynamic what-if scenario evaluation. This paper introduces DARLING (Digital Twins with Autonomous Reconfiguration and Learning), a flexible DT framework embedding AIdriven reasoning for distributed and edge ecosystems. It extends conventional DTs through an Imaginary Twin, a conceptual entity enabling predictive simulations, autonomous goal-setting, and scenario planning without modifying the physical counterpart. The architecture supports distributed inference at the edge, allowing AI models to run locally on DT nodes, reducing reliance on centralised processing while improving scalability and privacy. Additionally, we position this work within the broader DT research perspective, drawing from existing studies on AI-driven reasoning, hierarchical DT representations, and varying levels of autonomy. Our aim is to address their gaps by embedding real-time AI-driven reasoning whilst supporting adaptive, goaloriented coordination across distributed DTs.