About this Abstract |
Meeting |
2025 TMS Annual Meeting & Exhibition
|
Symposium
|
Algorithms Development in Materials Science and Engineering
|
Presentation Title |
Revealing Features in Kikuchi Patterns to Predict Plastic Deformation Localization Differences Between Wrought and Additively Manufactured Metallic Materials |
Author(s) |
Mathieu Calvat, Christopher Bean, Jean-Charles Stinville |
On-Site Speaker (Planned) |
Mathieu Calvat |
Abstract Scope |
Using high-resolution digital image correlation, significant differences were observed in how plastic deformation localizes at the sub-grain scale within wrought and additively Inconel 718 and stainless steel 316L. These differences are related to the complex microstructure produced during additive manufacturing. Instead of relying on EBSD maps (crystallographic and lattice curvature information) to explain these differences, a machine-learning approach was employed to identify relationships between Kikuchi patterns and plastic deformation localization. For this purpose, contrastive learning was used to encode the important features of Kikuchi patterns into a 64-feature vector, capturing all relevant information from EBSD measurements, including crystallographic details, curvature, and, more importantly, dislocation density and residual strain. Subsequently, a CNN model was trained to predict strain distribution within grains from spatially correlated sets of encoded Kikuchi patterns. This model revealed key microstructural features that control deformation localization in additively manufactured materials. |
Proceedings Inclusion? |
Planned: |
Keywords |
Additive Manufacturing, Machine Learning, Computational Materials Science & Engineering |