A deep learning-based digital twin system for creep behaviour of metallic alloys

Creep deformation is a dominant failure mode in metallic materials during service, governing structural stability, reliability and design life. However, conventional creep testing is time-consuming and resource-intensive, limiting the availability of high-fidelity creep data for efficient prediction. To address this issue, this study aims to develop a material-level digital twin system capable of generating high-fidelity creep strain data under limited experimental conditions. The proposed system is demonstrated using a 7-series aluminium alloy and employs a physics-guided gated recurrent unit model that integrates real-time sensor data with dynamic updating for creep strain prediction. Results show that incorporating physical constraints improves prediction accuracy by approximately 12% compared with the GRU model. Comparison with continuum damage mechanics-based creep models further demonstrated the competitive predictive capability of the proposed approach. Furthermore, a standalone application was developed to enable real-time monitoring and prediction of the creep process, facilitating efficient experimental implementation.