Patch-Based convolutional neural networks for multiple microstructural features detection in FIB-SEM micrographs of...

Area of Science:

• Materials Science
• Nuclear Engineering
• Computer Science

Background:

• Focused ion beam scanning electron microscopy (FIB-SEM) tomography is crucial for 3D microstructure analysis in irradiated nuclear materials.
• Current limitations include time-intensive data collection, microstructure differentiation challenges, and image alignment issues, hindering broader application.

Purpose of the Study:

• To develop an automated framework for microstructure identification in irradiated nuclear materials using convolutional neural networks (CNNs).
• To overcome the limitations of traditional FIB-SEM tomography analysis, particularly with limited annotated data.

Main Methods:

• Creation of a reliable annotated dataset with paired SEM and ground truth EDS maps.
• Application of CNNs, including Segment Anything Model (SAM) for image alignment and Patch-based U-Net, Attention U-Net, and Residual U-Net for segmentation.
• Validation of model performance for microstructure identification in U-10Zr metallic fuel.

Main Results:

• Patch-based U-Net demonstrated superior segmentation performance and consistency compared to other evaluated CNN models.
• The proposed framework successfully integrates SAM and Patch-based CNNs for semantic segmentation of irradiated nuclear materials.
• The approach reduces reliance on energy dispersive spectroscopy (EDS) detectors, accelerating analysis.

Conclusions:

• The developed framework automates microstructure identification in FIB-SEM tomography data, significantly accelerating nuclear material analysis.
• This study presents the first framework integrating SAM and Patch-based CNNs for semantic segmentation of irradiated nuclear materials.
• The methodology holds potential for broader applicability to other tomography datasets in materials science.