Machine learning approach toward near-homogeneous properties of eutectic Aluminum silicon alloy fabricated by a wire arc direct energy deposition

This study presents a parameter-based machine learning approach to predict optimal bead geometries in wire arc direct energy deposition (Wa-DED), aiming to reduce the time-consuming and costly trial-and-error procedures typically employed during process development. Decision trees, random forests, and K-nearest neighbors (KNN) models were trained and validated, with all three achieving comparable accuracy. Notably, the random forest model demonstrated superior performance in terms of accuracy, F1 score, and area under the curve (AUC). To further validate the optimized parameters, multilayer thin-walled Al-4047 structures were fabricated and comprehensively evaluated for both microstructural and mechanical properties. Microstructural analysis revealed α-Al dendrites with short, rounded Si in the interlayer and fine, fibrous Si in the melting zone, while EBSD confirmed predominantly equiaxed dendritic grains without a dominant crystallographic orientation, highlighting strong heterogeneity during solidification. Notably, sample-2 exhibited refined grains (92 ± 59 µm) and the highest average misorientation angle (10.14°) owing to its lower heat input. Meanwhile Sample 3 fabricated at equivalent heat input but with higher current amplitude, exhibited elevated KAM values and the highest dislocation density (~ 2.3 x 10¹⁴ m⁻²) due to intensified thermal gradients and cyclic thermal strains. These microstructural features strongly correlated with mechanical behavior, as Sample 2 with its finer grains, higher misorientation, and reduced porosity, achieved superior tensile strength (187.56 ± 11.40 MPa) and ductility (8.32 ± 0.66%), thereby demonstrating the efficacy of combining machine learning optimization with microstructural validation in tailoring Wa-DED components.