Machine-Learning Assisted Laser Powder Bed Fusion Process Optimization for AlSi10mg: New Microstructure Description Indices and Fracture Mechanisms

Abstract In this study, a machine-learning approach based on Gaussian process regression was developed to identify the optimized processing window for laser powder bed fusion (LPBF). Using this method, we found a new and much larger optimized LPBF processing window than was known before for manufacturing fully dense AlSi10Mg samples (i.e., relative density ≥ 99%). The newly determined optimized processing parameters (e.g., laser power and scan speed) made it possible to achieve previously unattainable combinations of high strength and ductility. The results showed that although the AlSi10Mg specimens exhibited similar Al-Si eutectic microstructures (e.g., cell structures in fine and coarse grains), they displayed large difference in their mechanical properties including hardness (118 - 137 HV 10), ultimate tensile strength (297 - 389 MPa), elongation to failure (6.3 - 10.3%), and fracture toughness (9.9 - 12.7 kJ/m2). The underlying reason was attributed to the subtle microstructural differences that were further revealed using two newly defined morphology indices (i.e., dimensional-scale index Id and shape index Is) based on several key microstructural features obtained from scanning electron microscopy images. It was found that in addition to grain structure, the sub-grain cell size and cell boundary morphology of the LPBF fabricated AlSi10Mg also strongly affected the mechanical properties of the material. The method established in this study can be readily applied to the LPBF process optimization and mechanical properties manipulation of other widely used metals and alloys or newly designed materials.