Applied AI / 3D imaging workflow

Automated 3D XCT analysis

Applied AI and Python workflows for large volumetric XCT datasets at CNL, including segmentation, object measurements, material features, mesh export, and QA reporting.

Domain XCT, TRISO fuel, industrial inspection, advanced fuel manufacturing

Stack Python, PyTorch, CUDA, MONAI, OpenCV, C++ integration, automated reports

Systems Segmentation, instance detection, metrology, synthetic data, reports

Context CNL research and industrial inspection work

Different XCT scans of TRISO compacts showing particles and fuel kernels — XCT image from Canadian Nuclear Laboratories.

Project notes

Internal research and inspection work is summarized here at a practical level. The details below are the useful parts to understand without exposing internal code or project materials.

Large XCT datasets needed repeatable analysis and traceable outputs.
The workflows support scan review, segmentation, object measurements, material-feature extraction, mesh export, and QA reporting for nuclear materials research.
Python pipelines connect models to usable technical reports.
Work includes cascaded 3D U-Net segmentation, 3D instance detection, PCA-based orientation estimation, geometry extraction, local packing-density analysis, synthetic XCT data generation, and automated report outputs.
Designed for research and inspection decisions.
Outputs need to be traceable and readable by scientists, engineers, and collaborators, so the workflow produces measurements, figures, 3D views, and written conclusions rather than just masks.
Part of an active XCT lab workflow.
The lab handles more than 130 analyses per year, so automation is valuable only when it reduces manual work while keeping measurements interpretable.
Code and internal materials are not public.
This page describes the work at a non-proprietary level. The public CNL article linked above gives outside context for the XCT and TRISO fuel work.