Research and Application of Prognostics and Health Management for Forging Press Oriented to Cyber-Physical Symbiosis

Forging is established as a critical manufacturing process for producing key components in high-end equipment. However, it is currently confronted with a prominent contradiction between extreme quality requirements and limited process controllability. To overcome the limitations of traditional digital twin models in mismatch and predictive capability, a new Cyber-Physical Symbiosis paradigm is proposed. This paradigm establishes a tripartite integrated system consisting of the physical forging system, the digital forging model, and the cyber-physical interaction mechanism, enabling bidirectional driving and collaborative evolution between the physical entity and its virtual counterpart. Within this paradigm, Prognostics and Health Management (PHM) is positioned as the core value carrier and practical implementation pathway of cyber-physical symbiosis. Using a forging press as a case study, an integrated technical solution for anomaly detection and remaining useful life prediction driven by cyber-physical symbiosis is constructed. A complete Cyber-Physical Symbiosis -enabled intelligent health management system is developed, with a forging press as the research object. In the perception layer, key parameters—including forming force, temperature fields, die temperatures, ram displacement, and velocity—are captured in real time via multi-source sensor networks. These data streams are preprocessed through denoising, normalization, and cleansing to supply high-quality inputs for the virtual model. In the modeling layer, a high-fidelity surrogate model based on a physics-informed neural network (PINN) is designed. Learnable physical parameters—such as friction coefficient, leakage coefficient, and viscous damping coefficient—are embedded into the model, enabling physics-driven tracking of equipment degradation while maintaining prediction accuracy. Furthermore, a hybrid VAE–LSTM anomaly detection model incorporating physics-based loss is proposed. Combined with a dynamic threshold mechanism, this model enhances both sensitivity to early faults and operational robustness. For RUL prediction, a physically interpretable health indicator is constructed using degradation-related parameters identified via the PINN. A BiLSTM–VAE time-series prediction framework is employed to achieve high-accuracy RUL estimation along with uncertainty quantification. In the application layer, a maintenance decision-support mechanism based on a multi-objective Markov decision process is established. Diagnostic alerts and RUL predictions are translated into actionable commands, such as preventive maintenance, parameter adjustment, or emergency shutdown. A closed-loop feedback from the virtual space to the physical system is thereby formed, ensuring continuous optimization. The proposed methodology was systematically validated using industrial forging press data. Experimental results demonstrated that the developed PINN surrogate model achieved low root mean square errors under both low- and high-pressure working conditions, exhibiting excellent predictive consistency and physical plausibility. The VAE-LSTM anomaly detection module was proven effective in identifying early-stage faults induced by abnormal furnace temperatures, while feature importance analysis was further employed to provide interpretable evidence for identifying root causes of anomalies. The RUL prediction model showed high agreement with the actual degradation trajectory throughout the entire life cycle, confirming its reliability in practical operational scenarios. Furthermore, the MDP-based maintenance strategy was observed to reduce the total lifecycle cost by approximately 65.14% compared with conventional scheduled maintenance, demonstrating its significant economic advantage and engineering applicability. This finding not only verifies the feasibility of the proposed symbiotic paradigm but also provides a new pathway for the intelligent health management of forging equipment.