# Accurate segmentation of localized fuel cladding chemical interaction layers in SEM micrographs with deep learning method

## TL;DR

This paper introduces a deep learning method to accurately and rapidly segment FCCI layers in SEM micrographs of nuclear fuel, enabling efficient post-irradiation analysis.

## Contribution

The novel contribution is an end-to-end deep learning approach for FCCI layer segmentation in SEM images with high accuracy and real-time applicability.

## Key findings

- A deep learning model achieved 90.8% average mAP for FCCI layer segmentation in SEM micrographs.
- The method is suitable for real-time applications and can be extended to other materials with similar imaging characteristics.

## Abstract

U-Zr metallic fuels are promising fuel candidates for fast reactor applications. Fuel/cladding chemical interaction (FCCI) is a random, localized, complex interdiffusion phenomenon occurring at the fuel cladding interface under irradiation, thinning the cladding wall. This interaction has been recognized as a limiting factor in deploying metal fuels to achieve higher burnup under steady state operations. The post irradiation examination of Experimental Breeder Reactor II and Fast Flux Test Facility fuel pins with different irradiation conditions have been the primary method for investigating FCCI in metal fuels, utilizing various characterization techniques, including scanning electron microscopy (SEM). This study compared several computer vision and deep learning approaches for the automated segmentation of FCCI layers in SEM micrographs. We deployed and compared state-of-the-art deep learning models for the task of FCCI layer segmentation in SEM micrographs. A deep learning based end-to-end method proved its capability to enable rapid and accurate segmentation of FCCI in as-collected SEM micrographs, making it highly suitable for real-time applications to automate data analysis. The average segmentation mAP achieves a high performance of 90.8% for dozens of FCCI layers. Furthermore, the method reported in this study is extendable to segmentation tasks for other materials with similar resolution, texture, and contrast characteristics, paving the way for accelerated and automated analysis in characterization analysis and beyond.