| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Novel Strategies for Rapid Acquisition and Processing of Large Datasets from Advanced Characterization Techniques
|
| Presentation Title |
Deep Learning-Driven Semantic Segmentation of large 4D Lab-Scale X-ray Tomography Data for Quantification of Microstructural Features |
| Author(s) |
Eshan Ganju, Nikhilesh Chawla |
| On-Site Speaker (Planned) |
Eshan Ganju |
| Abstract Scope |
Advancements in X-ray tomography have enabled the acquisition of high-resolution time-resolved 3D tomography (4D) datasets at the lab scale. However, the analysis and quantification of microstructural features in these large datasets remain challenging due to their complexity and size. In this study, we employed a novel deep-learning (DL) based approach for the semantic segmentation of large 4D lab-scale X-ray tomography data aimed at accelerating and accurately quantifying microstructural features within the dataset. Our approach used a sophisticated Generative Adversarial Network (GAN) based architecture to handle the spatial and temporal dimensions of data. A comprehensive dataset comprising diverse materials was utilized to train and validate the network. The performance of the DL approach, compared to traditional segmentation methods, was evaluated via quantitative metrics, including Intersection-Over-Union (IoU), segmentation accuracy, and computational efficiency. This study highlights the potential of DL in transforming lab-scale 4D X-ray tomography data in high-throughput and autonomous experimental approaches. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Characterization, Machine Learning, Other |