Volumetric Pore Measurements for Learning in Metal Additive Manufacturing

Abstract

Learning for pore inspection should be tied to a defined three-dimensional engineering quantity. This article develops a volumetric measurement framework using the published extraction example in NIST AMS 100-73 from 2026. Deterministic constructed volumes isolate connected components, voxelization, boundary displacement, and morphological merging. Exact geometric calculations show why small pores and surface-derived descriptors can be sensitive to processing choices. The proposed study design holds out complete builds, preserves voxel calibration and partial-object selection, and evaluates descriptor and endpoint errors alongside segmentation. Generated volumes retain synthetic provenance. No new XCT specimen, material-property estimate, or trained inspection score is claimed.