| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Materials Informatics for Images and Multi-dimensional Datasets
|
| Presentation Title |
Extraction of Local Scalar 3D Microstructural Properties of SOFC Electrodes from 2D Micrographs Using Convolutional Neural Networks |
| Author(s) |
William Flaherty Kent, Rochan Bajpai, Rachel Kurchin, William Epting, Harry Abernathy, Paul Salvador |
| On-Site Speaker (Planned) |
William Flaherty Kent |
| Abstract Scope |
Local microstructural properties control the electrochemical performance of SOFC electrodes. Common methods for obtaining fine-resolution, 3D reconstructions of microstructures are expensive, time-intensive, and limited in their ability to sample significant volumes, such as those relevant to heterogeneities in commercial SOFCs. Stereological methods can predict, albeit with significant uncertainties, average scalar properties from large field-of-view 2D images, but fail to capture heterogeneities relevant to local properties. These limitations are barriers to accurately studying microstructural changes that occur during cell operation, such as nickel coarsening. We discuss an approach to solve these problems using convolutional neural networks (CNNs) trained from 2D images from large volumes of reconstructed 3D microstructures. After training on 2D images labeled with their corresponding 3D scalar microstructural properties, the CNNs can extract from 2D images all relevant local scalar parameters with sufficient accuracy to study heterogeneity and degradation in performance. |