Digital Twin (DT) systems are shifting from pure state synchronization to task‑oriented, knowledge‑driven Cognitive Digital Twins (CDTs). Existing works often isolate techniques such as learning modules, knowledge graphs, or large language models, offering little guidance on systematic cognitive integration.
This paper introduces a four‑layer CDT architecture: the physical layer supplies real‑world states, the digital‑twin layer synchronizes these into manipulable models, the cognitive layer builds task‑specific cognitive models using knowledge, memory and attention, and the task layer generates concrete decisions under practical constraints. The layers form a closed operational loop: physical state → digital representation → cognitive modeling → task decision → operational feedback, which in turn refines the digital model and cognitive experience, enabling self‑evolution.
Within the cognitive layer, knowledge queries and attention mechanisms retrieve relevant information, while a memory module stores historical task experience to accelerate decision making in the task layer. Execution feedback updates the knowledge graph, annotations, and memory, supporting subsequent task interpretation, initiation and reasoning.
Two operation modes are distinguished: user‑request‑driven cognition, where external commands trigger task modeling, and self‑driven cognition, where the system autonomously detects opportunities and initiates tasks. Both share the same closed loop, differing only in the trigger.
Key enablers include semantic communication, cross‑layer knowledge querying, task orchestration and reliable closed‑loop synchronization. Deployment must address semantic consistency, latency and resource constraints.
A lightweight simulation demonstrates that closed‑loop tasks remain feasible even with limited semantic information, and that accumulated task experience markedly improves operational efficiency.
The proposed framework offers a structured foundation for designing future CDT systems, guiding researchers toward embedding adaptive, self‑evolving cognition into digital twins.
Review: The four‑layer closed loop tightly weaves cognition into the twin lifecycle, presenting a clear roadmap for self‑evolving DTs that merits real‑world testing.