About this Abstract |
Meeting |
MS&T23: Materials Science & Technology
|
Symposium
|
Materials Processing and Fundamental Understanding Based on Machine Learning and Data Informatics
|
Presentation Title |
Representation, Regeneration and Prediction of Microstructure in Additive Friction Stirring via Deep Regeneration Neural Network |
Author(s) |
Yunhui Zhu, Xiaofeng Wu, Hang Yu |
On-Site Speaker (Planned) |
Yunhui Zhu |
Abstract Scope |
Our study presents a novel deep learning framework that employs visual data to quantitatively analyze microstructural variations in metal fabrication through Additive Manufacturing (AM) under different processing conditions. By directly learning the latent microstructural descriptors from electron backscatter diffraction patterns (EBSD), we obtain a quantitative measure of microstructure differences in a reduced representation domain. The prediction of new microstructures within the domain is enabled by a regeneration neural network. This approach allows us to explore the physical insights into implicitly expressed microstructure descriptors. To validate the effectiveness of the framework, we analyzed samples fabricated through a solid-state AM technology, additive friction stir deposition, and efficiently obtained a reduced representation with as few as 5 principal components. Our research is a significant step towards establishing quantitative processing-structure linkages in metal AM and has the potential to be utilized in general materials science problems, such as heterogeneous material design and optimization. |