A digital-twin framework (DDD-GenDT) that models a physical machine using generative AI rather than large historical datasets. Built on the Dynamic Data-Driven Application Systems paradigm, it captures machine states in an observation graph, extracts recent sensor windows, and feeds them to an ensemble of large language models that predict upcoming sensor values zero-shot, without retraining. A feedback loop adapts the twin to machine wear and aging. Demonstrated on the NASA CNC milling dataset with spindle current, it enables data-scarce, privacy-preserving industrial digital twins.
This paper proposes DDD-GenDT, a framework for building a digital twin (a continuously updated virtual model of a physical machine) that uses generative AI instead of large historical training sets. …