An executable machine-tool-level digital twin for closed-loop CNC milling

Digital twins in machining are increasingly investigated for monitoring and prediction; however, their deployment at the machine-tool level for executable closed-loop regulation remains limited. This paper proposes a machine-tool-level integrated digital twin framework for CNC milling that combines virtual commissioning, online surface roughness prediction, and constrained adaptive regulation within a unified workflow. An offline mechanism twin is first developed for scenario-based virtual commissioning under representative command/load conditions, generating a deployable baseline consisting of transferable servo parameters and calibrated closed-loop dynamic response signatures. An online process twin is then constructed for roughness-oriented prediction and regulation through multi-source fusion of static process parameters and dynamic signals using a lightweight CNN–Transformer architecture. To support sensor-limited operation, a cutting-force simulation model with structured residual compensation is introduced, providing surrogate force-related information and achieving prediction performance comparable to that obtained with measured-force inputs (mean error 5.23% vs. 4.56%). Based on the predicted roughness state, a digital-twin-enabled model predictive control strategy computes constrained spindle-speed and feed-rate updates, which are executed through LinuxCNC HAL-based writeback. Milling experiments under representative controlled conditions demonstrate the feasibility of the proposed framework for roughness-oriented closed-loop regulation, establishing an executable path from machine-tool-level digital twins to controller-oriented online machining regulation.