About this Abstract |
| Meeting |
MS&T26: Materials Science & Technology
|
| Symposium
|
Uncertainty Quantification in Ultra-High Temperature Materials Manufacturing
|
| Presentation Title |
End-to-End Machine Learning for Creep Modeling: Data Processing, Parameter Learning, and Uncertainty Analysis |
| Author(s) |
Milly Yi, Yiqing Zhang, Xiang Chen, Mark Messner, Anthony Rollett |
| On-Site Speaker (Planned) |
Yiqing Zhang |
| Abstract Scope |
In high-temperature applications such as nuclear reactors and heat exchangers, creep is a critical degradation mechanism that can lead to premature failure. Traditional models capture isolated metrics but often fail to represent the full evolution of creep strain. This work presents a data-centric machine learning framework for full creep curve prediction, emphasizing robust data processing, feature extraction, and parameter learning from limited datasets. A structured preprocessing pipeline enables noise-tolerant strain-rate estimation and segmentation of creep regimes, from which physically meaningful parameters are extracted and learned as functions of stress and temperature. To address data scarcity, uncertainty quantification is incorporated to assess prediction reliability. In addition, an AI-assisted workflow is developed to automate data ingestion, parameter fitting, and model training. The approach is demonstrated on steels and nickel-based superalloys, with a focus on Haynes 282, providing a scalable framework for creep modeling in extreme environments. |