About this Abstract |
Meeting |
MS&T21: Materials Science & Technology
|
Symposium
|
Emergent Materials under Extremes and Decisive In Situ Characterizations
|
Presentation Title |
Investigation of Kirkendall Pore Formation and Evolution Using 4D Spatio-Temporal X-ray Tomography and Deep Learning |
Author(s) |
Arun J. Bhattacharjee, Pradyumna Elavarthi, Anca Ralescu, Ashley E. Paz y Puente |
On-Site Speaker (Planned) |
Arun J. Bhattacharjee |
Abstract Scope |
In-situ synchrotron x-ray tomography was performed during pack titanization of 75 and 100 µm diameter Ni wires at 925 °C. Due to the imbalance of Ni and Ti intrinsic diffusivities, during Ti deposition there is a net flux of vacancies toward the radial center of the sample that eventually coalesce leading to the formation of Kirkandall pores. In many cases, a multi-pore structure develops within a particular cross-section and some of these pores appear to sinter to the surface while others continue to grow further. To better understand the mechanisms governing this pore development and evolution, a fully convolutional neural network inspired from the U-net architecture was trained using masks created from previous in-situ homogenization tomography data to track individual pores during Ti deposition. The network was also used to predict the pore and phase evolution on intermediate sizes of the samples for which experimental data was not available. |