Automated detection of corrosion in used nuclear fuel dry storage canisters using residual neural networks

Nondestructive evaluation methods play an important role in ensuring component integrity and safety many industries. Operator fatigue can a critical the reliability of such methods. This is for inspecting high value assets or with consequence failure, as aerospace nuclear components. Recent advances convolution neural networks support automate these inspection efforts. paper proposes using residual (ResNets) real-time detection corrosion, including iron oxide discoloration, pitting stress corrosion cracking, dry storage stainless steel canisters housing used fuel. The proposed approach crops canister images into smaller tiles, trains ResNet on classifies corroded intact per-image count tiles predicted by ResNet. results demonstrate that deep learning allows to detect locus via at same time infer accuracy whether image comes from canister. Thereby, holds promise speed up fuel inspections, minimize costs, partially replace human-conducted onsite thus reducing radiation doses personnel.