Hierarchical machine learning driven process optimization of laser directed energy deposition for AlSi10Mg alloy by simultaneously minimizing porosity and pore area

Laser directed energy deposition (LDED) is a promising additive manufacturing technique for aluminum alloys, but LDED‑formed parts often suffer from porosity due to rapid solidification and unstable melt‑pool dynamics. This study proposes a hierarchical machine‑learning framework to optimize the LDED process for AlSi10Mg by simultaneously minimizing porosity and maximum pore area (MPA). An ensemble model is first trained to predict porosity (R² = 0.98), revealing the coupled balance between laser power and powder feed rate as the primary control lever and defining a low‑porosity process window (<0.3%). A gradient‑boosting model is then used to predict MPA (R² = 0.997) and to link transient melt‑pool instabilities to the formation of large pores within this constrained domain. Based on this hierarchical strategy, an optimal process window with an energy density of 100 J/mm and a powder utilization rate of 1.12 mg/W is identified. Experimental validation demonstrates that the framework accurately identifies process parameters for simultaneously minimizing porosity and MPA in LDED‑formed AlSi10Mg components. The optimized sample exhibits a porosity of 0.11% and an MPA of 0.384 μm², both substantially lower than those obtained with non‑optimized parameters. These results show that the hierarchical framework provides a robust, data‑driven pathway for controlling porosity defects in additively manufactured aluminum alloys.