Adaptive powder-spreading strategy driven by powder characteristics for high-quality LPBF forming

Fluctuations in powder characteristics, including particle morphology and size distribution, have a profound impact on the powder bed uniformity and overall fusion quality in laser powder bed fusion (LPBF) processes, ultimately hindering process stability and performance. To address these challenges, this study developed an adaptive powder-spreading strategy that integrates discrete element method (DEM) simulations with in-situ monitoring and machine learning (ML) techniques. A non-spherical particle DEM model was employed to capture the complex dynamic interactions governing powder behavior. The results reveal the competition between the powder-driven inertial and contact effects that dominate the spreading behavior of both gas-atomized (GA) and plasma rotating electrode process (PREP) TC4 powders under varying layer thicknesses and spreading speeds. Building upon this physics-informed understanding, a high-accuracy (93 %) support vector machine model was trained on DEM-generated data and further augmented with experimental measurements to enable rapid and robust optimization of spreading parameters tailored to specific powder characteristics. Under these ML-optimized conditions, the forming quality of PREP and GA-TC4 powders, as quantified by surface roughness (Ra), was significantly improved by 26.5 %–37.9 % relative to the equipment default settings. This integrated DEM–ML framework provides a physics-based, efficient, and adaptive pathway for process optimization, significantly reducing the reliance on costly trial-and-error experiments and enhancing the stability, reproducibility, and reliability of LPBF for complex powder systems.