| About this Abstract |
| Meeting |
2022 TMS Annual Meeting & Exhibition
|
| Symposium
|
Phase Transformations and Microstructural Evolution
|
| Presentation Title |
Leveraging EBSD Data for Phase Transformation Product Quantification in a Low Carbon Steel by Deep Learning |
| Author(s) |
Simon Breumier, Tomas Martinez Ostormujof, Nathalie Gey, Audrey Couturier, Pierre-Emmanuel Aba-perea, Bianca Frincu, Natalia Loukachenko, Lionel Germain |
| On-Site Speaker (Planned) |
Tomas Martinez Ostormujof |
| Abstract Scope |
Quantification of the different phases in steels is challenging. It usually relies on micrographs, which features are sensitive to the imaging conditions (etchant, detector used…) and does not always provide enough contrast to differentiate some of the transformation products.
In contrast, EBSD analyses provides a wealth of various features specific to the material’s transformation history: Orientation Relationship, low angle misorientation densities, habit planes and diffraction contrast. This work investigates the application of deep learning approaches to quantify different phase transformation products in an industrial steel using EBSD data.
A U-NET convolutional neural network was developed for automatic segmentation of bainite, martensite and ferrite in a low carbon steel, using either the orientations, the misorientations and different pattern quality indicators provided by EBSD. Alternatively, semi-supervised approaches such as generative adversarial networks were explored to achieve similar results without any manual labelling before training. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Phase Transformations, Iron and Steel |