About this Abstract |
Meeting |
MS&T25: Materials Science & Technology
|
Symposium
|
Integrated Computational Materials Engineering for Physics-Based Machine Learning Models
|
Presentation Title |
Physics-Based Machine Learning Framework for Fatigue-Life Estimation in Wrought Mg Alloys |
Author(s) |
Taekyung Lee, Jinyeong Yu, Sung Hyuk Park |
On-Site Speaker (Planned) |
Taekyung Lee |
Abstract Scope |
Mg alloys, promising candidates for lightweight structural applications, are commonly processed in wrought form to address the alloys' poor mechanical properties. The limited c-axis deformation mechanisms present in a hexagonal close-packed structure result in strong crystallographic textures during thermomechanical processes. This leads to asymmetric stress–strain responses under low-cycle fatigue, as the twinning-detwinning mechanism generates non-zero mean stress depending on the loading direction. Conventional models for fatigue-life estimation are not applicable in this case. We suggest an alternative predictive strategy that combines an energy-based model and machine learning (ML) to address this issue. The key disadvantage of the energy-based model is that it relies on prior knowledge of strain energy values, restricting its applicability under untested conditions. To overcome this limitation, we propose a physics-informed framework comprising three ML-assisted modules: a hysteresis curve generator, feature extractor, and fatigue-life estimator. The proposed model demonstrates strong generalization across various wrought Mg alloys. |