Detection, classification and prediction of internal defects from surface morphology data of metal parts fabricated by powder bed fusion type additive manufacturing using an electron beam

Abstract

In powder bed fusion type additive manufacturing using an electron beam (PBF-EB), various process parameters have a significant influence on the performance of manufactured parts. To expand the use of PBF-EB technology in the material industry, one of the problems is the generation of internal defects (pores, unmelted powder, among others) during the process. In this study, we determined a quantitative criterion (Sdr < 0.015 for an even surface; Sa ≥ 80 µm for an uneven surface; Sdr ≥ 0.015 and Sa < 80 µm for a porous surface) for classifying surface quality based on surface flatness, and we revealed that different surface qualities (even, uneven, and porous) include different types of internal defects. The parts with even surfaces are free of internal defects and have the highest density(7.7962 g/cm3). Parts with uneven surfaces have a large number of spherical pores owing to their excessive energy input, while parts with porous surfaces have a considerable number of irregularly shaped defects and unmelted powders owing to their insufficient energy input. When the energy input is excessively high, the combination of the Marangoni effect, vapor recoil pressure, and electron beam agitation leads to a high velocity flow of liquid, which tends to form bumps, resulting in an uneven surface. Conversely, if the energy input is too low, the depth of the melt pool is too small to penetrate the thickness of the powder layers, resulting in incomplete melting of the powder at the bottom of the layers and the formation of defects due to lack of fusion between the layers. In addition, five types of machine learning technologies (logistic regression, support vector machine, decision tree, XGBoost, and naive Bayes) were applied to the PBF-EB process parameters optimization of the S30C alloy. A support vector machine has the highest model performance, and we use it to construct a processing map corresponding to the internal defects and determine the PBF-EB process window for the S30C alloy. The optimal PBF-EB process parameter ranges for S30C alloy were predicted as follows: current = 2.5–10 mA, scan speed = 200–1000 mm/s, line offset = 0.11–0.25 mm, or current = 2.5–10 mA, scan speed = 200–750 mm/s, line offset = 0.27–0.33 mm. Moreover, a new framework for constructing a process map of PBF-EB fabricated parts was proposed, which is an effective method to accelerate PBF-EB for manufacturing parts without internal defects.