| About this Abstract |
| Meeting |
2024 ASC Technical Conference, US-Japan Joint Symposium, D30 Meeting
|
| Symposium
|
2024 ASC Technical Conference, US-Japan Joint Symposium, D30 Meeting
|
| Presentation Title |
Discerning Microstructure of CMC Minicomposites from Micro-CT Imaging Data Using Machine Learning |
| Author(s) |
Adrian Brelay, Ashley Hilmas, Olesya Zhupanska |
| On-Site Speaker (Planned) |
Olesya Zhupanska |
| Abstract Scope |
In this work, machine learning (ML) algorithms for automatic image segmentation of micro-CT data of ceramic matrix composites (CMCs) are developed. CMCs under consideration are micro-composites consisting of continuous silicon carbide (SiC) fibers embedded in the SiC matrix. Fibers are coated with boron nitride (BN). The microstructure contains numerous voids and cracks developed due to tensile loading. The objective of this study was to obtain microstructural topology of CMCs with and without damage by accurate segmentation of material phases (fibers, matrix, interphase, and voids) and cracks. This is a nontrivial task due to a low contrast ratio of the images, because SiC fibers and SiC matrix are of similar density. A 2.5D U-Net Deep Learning Architecture was used to reconstruct microstructure by sequential segmentation of pores, fibers, interphases between fiber and matrix, and matrix. A separate model was developed to segment cracks. The analysis of segmentation results has been performed. The analysis was focused on investigation of the effects of the training datasets on the segmentation results. |
| Proceedings Inclusion? |
Definite: Post-meeting proceedings |