| 研究実績の概要 |
The multi-sized porous electrodes are crucial for the electrochemical performance of fuel cells and batteries, but their quantitative characterization is a challenging task. This project developed an automated framework for microstructure analysis using machine learning. The framework addresses various problems, including improving 3D microscopy measurements by super-resolving in-depth direction with the patch-VDSR network and semantic segmentation by patch-CNN. The GAN 2D->3D algorithm was proposed for reconstructing artificial isotropic 3D models from 2D images, while the weak GAN 2D->3D network was developed for anisotropic materials. Additionally, the C-GAN model was trained with a microstructural database, enabling the fabrication of realistic models with predefined properties and gradients.
Further, the GDC-based solid oxide fuel electrodes with the controlled properties were successfully fabricated and quantitatively characterized with proposed algorithms. Optimal fabrication conditions were discussed in respect to the porosity, composition, and particle size. Increased GDC composition and decreased particle size resulted in improved electrochemical performance. However, high GDC composition decreased porosity and increased gas diffusion resistance due to particle coarsening. Moreover, GDC share over 80vol% led to long-term operation stability issues due to the formation of nano-cracks. This study emphasizes the significance of microstructure characterization and the potential of machine learning in understanding of microstructure and degradation related phenomenon.
|
