Digital twin modeling of a large-scale forming system with high consistency: Methodology, validation, and application to condition monitoring

Large-scale forming equipment is widely used in high-end manufacturing, but continuous reciprocating high-load operation causes service-condition changes and component degradation. The strong coupling among electromechanical and hydraulic subsystems makes it difficult to construct a consistent digital twin (DT) model for condition monitoring. This paper proposes a high-consistency DT modeling method for a large-scale forming system. First, we establish integrated physical models of the hydraulic forming system. Then, Sobol parameter sensitivity analysis (SPSA) is used to reduce parameter dimensionality and identify parameters that strongly affect displacement and pressure observations. Based on the selected high-sensitivity parameters, a damped least-squares algorithm (DLSA) is applied to iteratively identify and optimize the dynamic consistency between the virtual model and the physical equipment. A Kalman filter (KF) is further embedded to fuse DT-generated virtual data with sensor data for anomaly recognition. The method is validated on a 20-t forming equipment platform involving five major components and forty parameters. SPSA and DLSA reduce the identification task to eight highly sensitive parameters, improving parameter identification by 80%. Experimental results under four working conditions show that the symmetric mean absolute percentage error is within 1.42%, and the pressure error is within 9.90%. The fused twin data achieved an anomaly recognition accuracy of 98.75%, outperforming sensor-only data by 1.95%.