Operationalising cognitive digital twins through a cognitive-function framework and GenAI tools
Year: 2026
Type: article
Abstract: Cognitive Digital Twin (CDT) is emerging as an evolution of Digital Twin (DT) for smart manufacturing, but most CDT studies remain at the conceptual level and lack implementable cognitive-module architectures. This paper proposes an implementation-oriented CDT reference architecture that formalises four cognitive functions, Perception, Attention, Reasoning & Problem Solving, and Memory & Learning, through defined computational roles and interaction mechanisms. To examine how these architectural roles can be operationalised, we instantiate a bounded subset of their interactions as an L3 geometric-modelling workflow in a robotic cell. In this prototype, YOLOv11 provides coarse asset-detection and position-change cues, precise device poses are obtained from a pre-calibrated registry with human confirmation, and a GPT-5.1-driven agentic workflow performs spatial-instruction parsing, geometric constraint formulation, verification, bounded replanning, and FreeCAD-script synthesis. Controlled experiments with 2–32 devices characterise the pre-execution workflow time, token consumption, internal constraint acceptance, and replanning outcomes. The results provide initial evidence that selected architectural interactions can be translated into an executable CDT function, while also revealing limitations for large, structurally over-dense layouts.
Institutions: University of Auckland, China Academy of Engineering Physics
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Funder: China Scholarship Council
OpenAlex ID: W7212270436