Abstract: Robotic reconfiguration of cosmetic production lines faces significant challenges, including high risk, long implementation cycles, performance fluctuations, and complex decision-making. To address these issues, this study proposes and validates a digital twin-driven systematic progressive reconfiguration framework incorporating a comprehensive risk control strategy. By constructing comprehensive system representation and dynamic process models using physical-logical dual-dimension representation and time-attribute Petri nets, the proposed framework achieves progressive reconfiguration through iterative optimization, virtual verification, and dynamic adjustment. It also systematically integrates a multi-level risk control mechanism based on failure mode and effects analysis (FMEA) for quantitative risk assessment and proactive intervention. Semi-physical simulation validation (N=20 independent trials) on a cosmetic production line demonstrates that, compared with traditional one-time reconfiguration schemes, the proposed digital twin-driven progressive strategy (DTPS) significantly reduces the reconfiguration risk index (from 0.70±0.05 to 0.45±0.03), system fluctuation level (from 38.3%±3.7% to 25.1%±2.2%), and total reconfiguration time (from 72.5±2.8 hours shortened to 50.3±1.9 hours). In addition, the reconfigured production line achieves a balance rate of 87.00% (from 72.00%), reduces labor demand to 39 operators (from 47 operators), increases production capacity to 64 bottles/min (from 57 bottles/min), and yields an investment payback period of approximately 30~32 months. The results indicate that the proposed framework provides a more comprehensive and risk-controllable solution for robotic reconfiguration while effectively balancing multiple objectives.