Answers are grounded in this paper and cite their sources.
This study explores the use of artificial neural networks (ANN) and transfer learning (TL) to develop processing maps that guide defect-free manufacturing of as-built L-PBF aluminum (AlSi10Mg) and stainless steel (SS316L) specimens. The complex non-linear relationships between processing parameters and the thermal properties of the materials, which influence melt pool development, highlight the need for machine learning (ML) tools to achieve high-quality processability in a cost-effective manner. Commercial AlSi10Mg and SS316L powders were processed using L-PBF, resulting in various types of porosity, such as keyhole and lack-of-fusion defects, under different processing conditions. We first characterized the bulk density and melt pool features (width and depth) through optical microscopy and image analysis. Next, we trained ANN base models using data from existing literature to predict the bulk density and melt pool geometries of the as-built samples. Finally, we refined these models with our experimental data after transferring the base models. The results indicate that our proposed models and TL methodology effectively predict processing maps, identify optimal processing parameters for maximum density, and establish the threshold for lack-of-fusion porosity.
No takes yet. Share an insight, caveat, or question.
Bahari-Sambran et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: