| Project/Area Number |
21K14090
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 19020:Thermal engineering-related
|
|---|
| Research Institution | The University of Tokyo |
|---|
Principal Investigator |
SCIAZKO Anna 東京大学, 生産技術研究所, 特任助教 (30898945)
|
|---|
| Project Period (FY) |
2021-04-01 – 2023-03-31
|
| Project Status |
Completed (Fiscal Year 2022)
|
| Budget Amount *help |
¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000)
Fiscal Year 2022: ¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000)
Fiscal Year 2021: ¥2,990,000 (Direct Cost: ¥2,300,000、Indirect Cost: ¥690,000)
|
|---|
| Keywords | solid oxide fuel cell / machine learning / 3D microstructure / FIB-SEM / GAN network / semantic segmentation / super-resolution / artificial structure / porous material / porous electrode |
|---|
| Outline of Research at the Start |
The multiscale porous media found interest in many fields of engineering. In particular, the porous electrode microstructure determines the performances of fuel cells and batteries. The multi-sized pore design is beneficial as large pores enhance the gas transport and nano-pores increase active reaction area. Here, the characterization methods of multi-sized porous media based on deep neural networks will be proposed. The focus is to provide high resolution large volume 3D characterization, fabricate synthetic 3D structure from single 2D image and correlate microstructure with performance.
|
| Outline of Final Research Achievements |
An automated microstructure analysis framework using machine learning was developed. The framework was built to improve three-dimensional (3D) microscopy analysis by super resolution and semantic segmentation algorithms. The developed framework can contribute in shortening actual measurement time up to 8 times and data post-processing time by two orders of magnitude. Furthermore, a novel method with generative network was proposed to create an artificial 3D microstructure model from a single two-dimensional image. Additionally, the generative network trained with microstructure datasets, can fabricate realistic microstructure models with predefined properties and gradients. The automated segmentation framework was used to characterize gadolinium doped ceria ceria-based solid oxide fuel (SOFC) anodes with the controlled properties (porosity, material composition, and particle size) and it enables to reconstruct SOFC anode with carbon deposition.
|
| Academic Significance and Societal Importance of the Research Achievements |
The porous media found interest in many fields of engineering. Particularly, the porous electrode microstructure determines the performances of fuel cells and batteries. This study proposed comprehensive framework for analyzing porous media microstructures based on machine learning methods.
|
