Coupon-level features and code for build-held-out evaluation of in-situ tensile-property prediction in laser powder bed fusion (Peregrine v2023-11)
Description
Coupon-level dataset and Python code accompanying the article “Can in-situ layer-wise imaging qualify parts on a build it has never seen?”. The CSV contains 11,776 SS-J3 coupons (6,379 with tensile results) extracted from the five builds of the ORNL Peregrine v2023-11 dataset: tensile properties (UTS, YS, TE, UE), process parameters, coupon location, 39 DSCNN segmentation features, 20 visible-light camera features and 18 machine-log features. The code reproduces the remote HDF5 extraction, coupon-centred image stacks, all models (LightGBM, 3D-CNN), validation protocols (random, within-build, part-grouped, leave-one-build-out), conformal calibration, decision analysis, tables and figures. Derived from: Scime et al., Peregrine v2023-11, https://doi.org/10.13139/ORNLNCCS/2001425.