Surface quality represents a critical challenge in additive manufacturing (AM), with surface roughness serving as a key parameter that influences this aspect. In the aerospace industry, the surface roughness of the aviation components is a very important parameter. In this study, a typical Al alloy, AlSi10Mg, was selected to study its surface roughness when using Laser Powder Bed Fusion (LPBF). Two Random Forest (RF) models were established to predict the upper surface roughness of printed samples based on laser power, laser scanning speed, and hatch distance. Through the study, it is found that a two-dimensional (2D) RF model is successful in predicting surface roughness values based on experimental data. The best and minimum surface roughness is 2.98 μm, which is the minimum known without remelting. More than two-thirds of the samples had a surface roughness of less than 7.7 μm. The maximum surface roughness is 11.28 μm. And the coefficient of determination (R²) of the model was 0.9, also suggesting that the surface roughness of 3D-printed Al alloys can be predicted using ML approaches such as the RF model. This study helps to understand the relationship between printing parameters and surface roughness and helps print components with better surface quality.

This work investigates the effects of various processing parameters (laser power, scanning speed, hatch distance and beam offset) on the resultant inclined up-facing surface roughness of AlSi10Mg alloys produced by laser powder bed fusion (LPBF). A two-step approach, orthogonal test followed by the Doehlert matrix design (DMD) test is used to efficiently optimize the up-facing surface and contour parameters. The former method aims to determine the significance of variables while the latter one facilitates a rapid optimization. The results show that the interaction and interdependency among the parameters are of great significance to the obtainable surface roughness. Using a rational design of experiments, the optimized up-facing surface roughness of Ra of 5.4 μm is achieved. This is attributed to the elimination of the laser partition track and the reduction in irregularities at the edges of the parts. This work demonstrates an effective approach of experimental processing parameter optimization to improve the surface finish of LPBF parts.