Effects of Scan Parameters in Machine Learning-based Defect Detection of Reconstructed 3D Tomographic Datasets

Inertial confinement fusion (ICF) experiments on high density carbon (HDC) capsules have achieved record fusion yields at the National Ignition Facility (NIF). X-ray computed tomography (CT) is a non-destructive technique used during the production of HDC capsules for such experiments. This technique is used to identify defects located within the capsule wall and inner surface, which degrade capsule performance. The long acquisition time (~16 hours) required for high-resolution CT scans is a major throughput obstacle, placing large constraints on equipment and personnel time. SHELLNET is a machine learning algorithm created by General Atomics that analyzes thousands of images generated from CT datasets and quantifies capsule defect data. This report studies the sensitivity of SHELLNET to changes in acquisition parameters such as number of projections, acquisition time and image binning. Enhanced understanding of SHELLNET's capabilities would lead to improved acquisition time per capsule without compromising output quality, ultimately allowing for more capsules to be examined and a larger selection pool of capsules for ICF shots at the NIF.