Description
Creep rupture is a slow and silent failure in metallic materials such as 9Cr-1Mo steels, stainless steels, and titanium alloys, presenting significant safety risks in aerospace and power generation industries. Traditional parametric models—Larson-Miller and Manson-Haferd parameters—rely on a limited set of input features and demonstrate poor generalisation beyond training conditions. With the increase in data and evolution of Artificial Intelligence (AI), machine learning (ML) —a branch of AI approaches has achieved high interpolative accuracy but remains limited due to cost and scarcity of datasets, poor extrapolation capability, and a lack of physical interpretability. This paper reviews these limitations systematically and proposes a framework that combines classical parametric models with advanced ML architectures. By employing Affinity Propagation clustering with Generative Adversarial Networks (AP-GAN) for synthetic data augmentation, which solves the problem of data scarcity and integrates explainable AI through SHAP values and physics-informed constraints, it solves the explainability problem. The objective is to produce creep rupture life predictions that are not only accurate within known regimes but generalizable, interpretable, and practically deployable by engineers.| Period | 18 Jun 2026 |
|---|---|
| Event title | Yorkshire Innovation in Science and Engineering Conference 2026: A two-day PhD student-led conference of presentations, poster sessions, and discussions |
| Event type | Conference |
| Location | Huddersfield, United KingdomShow on map |
| Degree of Recognition | Local |