About this Abstract |
Meeting |
2020 TMS Annual Meeting & Exhibition
|
Symposium
|
Fatigue in Materials: Fundamentals, Multiscale Characterizations and Computational Modeling
|
Presentation Title |
A Machine Learning Approach to Predict Fatigue Damage and Crack Initiation Sites in a BCC Steel Microstructure |
Author(s) |
Ali Riza Durmaz, Thomas Straub, Chris Eberl |
On-Site Speaker (Planned) |
Thomas Straub |
Abstract Scope |
A material fatigue lifetime is determined by the crack formation process: damage accumulation in individual grains, micro crack initiation, and finally short crack formation. In the past years, a testing methodology for fatigue damage evolution investigation was developed. This methodology uses sensitive measurements of the resonant frequency for correlation with damage initiation. Building on this work, a multi modal approach has been developed employing in-situ optical sample surface images.
The processed image data allows the creation of labels for machine learning (ML) methods. As ML features, various microstructure characteristics are extracted using ex-situ EBSD measurements processed with MTEX and complemented with crystal plasticity FEM (CPFEM). Feature importance and relationships are extracted from a random forest model. Semantical segmentation with UNET of post-mortem SEM images allows the distinction between extrusion and crack regions.
The first comparison of experimentally obtained, CPFEM and ML predicted damage initiation sites will be presented. |
Proceedings Inclusion? |
Planned: Supplemental Proceedings volume |