Abstract: A machine-learning approach based on gaussian process regression (GPR) was proposed to optimize the processing window of laser power (P) and scanning speed (v) in the IN625 alloy fabricated by laser powder bed fusion (LPBF) using the experimental data of relative density, crystallography orientation and shape aspect ratio of columnar grains. The effect of laser power-scanning speed combinations and laser energy density (ED) on the relative density and microstructure of the LPBF specimens was investigated as well. The results show that the applied laser ED≥55 J/mm³ is prerequisite for fully dense LPBF specimens with relative density ≥99%. The optimized L-PBF processing window for manufacturing fully dense IN625 alloy is pear-shaped, and the selectable scanning speed range can be wide in the case of high laser power. The LPBF specimens possess the columnar grains with the preferred orientation <001> parallel to build direction (BD). The preferred orientation intensity and the shape aspect ratio of the columnar grains tend to increase with the laser ED but marginally decrease after a high ED value. The leave-one-out cross validation reveals that the GPR model predicting the optimized LPBF process window based on relative density and microstructure features is reliable and can be readily applied to multi-objective optimization of laser additive manufacturing process in other metals and alloys.

Key words: laser powder bed fusion, gaussian process regression, relative density, microstructure