Feature-based volumetric defect classification in metal additive manufacturing

Abstract

AbstractVolumetric defect types commonly observed in the additively manufactured parts differ in their morphologies ascribed to their formation mechanisms. Using high-resolution X-ray computed tomography, this study analyzes the morphological features of volumetric defects, and their statistical distribution, in laser powder bed fused Ti-6Al-4V. The geometries of three common types of volumetric defects; i.e., lack of fusions, gas-entrapped pores, and keyholes, are quantified by nine parameters including maximum dimension, roundness, sparseness, aspect ratio, and more. It is shown that the three defect types share overlaps of different degrees in the ranges of their morphological parameters; thus, employing only one or two parameters cannot uniquely determine a defect’s type. To overcome this challenge, a defect classification methodology incorporating multiple morphological parameters has been proposed. In this work, by employing the most discriminating parameters, this methodology has been shown effective when implemented into decision tree (>98% accuracy) and artificial neural network (>99% accuracy).