Thermographic Reference Dataset: Defect Detection in Nuclear Waste Barrel Cutouts Using Long Pulse Thermography

We introduce a thermographic dataset for subsurface defect detection in radioactive waste storage drums, comprising thermal sequences from 7 barrel specimens with artificially manufactured internal defects. The dataset was acquired using cost-effective halogen-lamp excitation (2kW per lamp) as an alternative to laser-based systems, with dual-camera thermal imaging (CMOS and bolometric) to enable performance comparison across imaging modalities. The specimens include both new and aged barrel types with controlled defects — FBHs, lines, crosses, triangles, and rectangles — simulating internal corrosion at varying scales (4mm to 60mm). Three heating regimes (both lamps, left only, right only) were systematically applied across multiple measurement regions per sample, yielding normalized thermal sequences. To lower the barrier for machine learning practitioners without thermography expertise, the dataset provides pre-computed features derived from principal component analysis, pulse phase thermography, and independent component analysis extracted using experimentally optimized time windows. Ground-truth binary masks mapping defect locations are included to enable supervised learning. This resource is designed to support the development and benchmarking of automated defect detection algorithms for non-destructive testing of curved, thin-walled metallic structures under realistic surface conditions (paint inhomogeneity, dirt, geometric artifacts), while validating low-cost thermographic inspection alternatives for industrial deployment.