NSUF 25-5555: Machine Learning-Based Analysis of Post-Irradiation Examination Data for Fuel-Cladding Chemical Interaction Characterization in Metallic Fuels
The proposed research will advance post-irradiation examination (PIE) analysis by applying automated machine learning (ML) to characterize fuel-cladding chemical interaction (FCCI) layers and bubble features in irradiated metallic fuels, directly supporting the mission of DOE’s Office of Nuclear Energy. The primary goal is to develop and validate deep learning models on Idaho National Laboratory (INL) high-performance computing (HPC) clusters for automated analysis of existing Advanced Fuels Campaign (AFC) datasets. Efforts will focus on FCCI layer characterization in scanning electron microscopy (SEM) images and bubble quantification in optical microscopy (OM) images.
The project employs advanced ML architectures, including YOLO for rapid detection and Mask R-CNN for precise segmentation, adapted for nuclear materials analysis. Building on prior work demonstrating human-level accuracy in bubble and void detection, custom algorithms will be designed to identify FCCI layers and analyze bubble morphology. Remote access to curated AFC datasets via INL HPC enables processing of large-scale image collections, otherwise infeasible with conventional methods. Quantitative outputs will include FCCI layer thickness evolution, composition gradients, bubble size distributions, spatial correlations, and morphological statistics.
If successful, this research will establish automated, standardized protocols for nuclear fuel characterization, unlocking the scientific value of existing AFC datasets. The resulting capabilities will accelerate materials qualification for advanced reactors by enabling systematic correlation of microstructural evolution with irradiation conditions and performance parameters. Automated analysis reduces subjectivity, expands dataset coverage, and supports data-driven materials design.
The 12-month project, aligned with Super RTE guidelines, includes dataset access and model adaptation (Months 1–6), automated pipeline development and validation (Months 7–9), and scientific analysis with dissemination (Months 10–12), using 240–400 hours on INL HPC clusters. Anticipated outcomes include validated ML models, quantitative insights into FCCI and bubble behavior, predictive correlations, open-source tools, publications, and conference presentations that advance nuclear fuel qualification.
Additional Info
| Field | Value |
|---|---|
| Awarded Institution | Missouri University of Science and Technology |
| DOI | 10.46936/NSUF/60015723 |
| Embargo End Date | 2028-01-22 |
| Facility Tech Lead | Noé Morales |
| NSUF Call | FY 2025 Super RTE Call |
| PI | Shradha Agarwal |
| Project Member | Dr. Tiankai Yao, Staff Scientist and Nuclear Facility Engineer - Idaho National Laboratory (https://orcid.org/0000-0001-8330-7638) |
| Project Member | Jason Schulthess, Nuclear Facility Research Engineer - Idaho National Laboratory (https://orcid.org/0000-0002-4289-7528) |
| Project Member | Dr. Yachun Wang, Nuclear Engineer - Idaho National Laboratory (https://orcid.org/0000-0002-4952-3633) |
| Project Member | Dr Shradha Agarwal - Missouri University of Science and Technology (https://orcid.org/0000-0002-6901-8087) |
| Project Type | RTE |