Characterisation and prediction of mechanical properties in laser powder bed fusion-printed parts: a comparative analysis using machine learning

Abstract

ABSTRACT This study investigates the effects of process parameters including scanning strategy, build orientation, and hatching distance on the mechanical properties of AlSi10Mg parts produced by Laser Powder Bed Fusion (L-PBF). The experiment varied these parameters within defined ranges and used statistical analysis to evaluate their impact on tensile strength and ductility. Results showed that scanning strategy had the greatest influence, followed by hatching distance, while build orientation affected anisotropic properties. Microstructural analysis showed clear correlation between process conditions and mechanical strength, thereby showing the underlying mechanisms that govern material behavior. Moreover, Machine learning models, including Random Forest Regression (RFR), Support Vector Regression (SVR), and Artificial Neural Networks (ANNs), were applied to predict tensile strength and ductility characteristics. RFR and SVR outperformed ANNs, showing high predictive accuracy with limited datasets. These findings emphasize the importance of optimizing L-PBF process parameters to minimize anisotropy and achieve consistent mechanical properties in produced parts.