Digital twin-based online forging stress prediction for fast forging hydraulic presses

In fast forging hydraulic presses, the equivalent internal stress of forgings directly determines the forming quality and the service life of the resulting components. However, such stress cannot be directly measured during forging operations, which limits real-time process monitoring and quality control. To address this challenge, a digital twin-driven method is proposed to rapidly predict the equivalent stress of forgings. A multi-domain digital twin model is constructed by integrating a hydraulic model of the fast forging press with a finite element (FE) model of the workpiece. Equivalent stress data corresponding to various operating conditions are systematically generated through FE simulation, thereby providing the training dataset for surrogate modeling. On this basis, an adaptively fused LSTM-XGBoost surrogate model is developed. MC Dropout and ensemble learning strategies are employed to quantify the uncertainty of the LSTM and XGBoost sub-models, respectively. The results indicate that the proposed adaptive LSTM-XGBoost surrogate model achieves a coefficient of determination ( R ² ) of 0.971 and a root-mean-square error (RMSE) of 12.3 MPa. To enable online equivalent stress prediction, the trained surrogate is coupled with a Simulink-based hydraulic model that reproduces the real-world operating behavior of the fast forging hydraulic press. This configuration enables online stress prediction driven by real-time process parameters generated by the hydraulic simulation model. The proposed method offers a simulation-based approach to online stress prediction for forging operations, providing a foundation for future physical deployment. It also provides guidance for improving the forming quality of large-scale forgings and optimizing relevant process parameters.