| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Algorithms Development in Materials Science and Engineering
|
| Presentation Title |
Deep Learning for Quantitative Dynamic Fragmentation Analysis |
| Author(s) |
Erwin Cazares, Brian Elias Schuster |
| On-Site Speaker (Planned) |
Erwin Cazares |
| Abstract Scope |
We developed an image-based convolutional neural network (CNN) for quantitative time-resolved measurements of fragmentation in opaque brittle materials using ultra-high-speed optical imaging. This model enhances the U-net model and was trained on binary, 3-class, and 5-class datasets using supervised learning from dynamic fracture experiments on various opaque ceramics adhered to transparent polymer backings. The experiments involved applying spatially and time-varying mechanical loads to induce inelastic deformation and fracture, recorded at frequencies up to 5 MHz. The dataset includes common static and dynamic fracture modes like cone cracking, median cracking, and comminution. While our training data came from dynamic fragmentation experiments, the methodology is applicable to static loading due to similar crack speeds. This study aims to quantify failure processes in structural materials for protection applications and validate engineering models used in design. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Characterization, Ceramics |