The effects of 0.4wt.% Sc addition on a typical Al-Si10-Mg alloy were systematically investigated in the present research. Samples with and without Sc produced refined dendritic arranged microstructures with sensitivity to the aging treatment after solidification, particularly in the case of the alloy without Sc. After being exposed to 300°C for 90 minutes, the dendritic spacing nearly doubled in the Al-10wt.%Si-0.45wt.%Mg samples. The rapidly solidified microstructures were constituted by the -Al dendritic phase surrounded by eutectic phases/intermetallics such as Si, Mg2Si and Al3Sc (in the case of the alloy containing Sc). 255°C and 300°C were deemed most appropriate temperatures for aging treatments, with four exposure times of up to 120 minutes tested for each alloy. The heat treatments allowed the Vickers hardness profiles to be plotted and compared. Moreover, in order to detect both Sc- and Mg- precipitates after aging, specific samples have been prepared for either SEM or TEM analyses. At first, the results pointed to a strong precipitate-related hardening effect formed as a result of the Sc addition to the alloy. All samples containing Sc showed a higher hardness value when compared to their respective treated samples without Sc. Secondly, when comparing the Al-10Si-Mg-Sc alloy samples among themselves after being treated at different conditions, high temperatures and excessive treatment times can become detrimental to the hardness. This was due to the growth of larger Sc-bearing precipitates of approximately 1 µm in size under such conditions, having less pronounced hardening effect. The best condition (255°C for 60 min) for as centrifuged samples in Cu-mold produced very fine dispersion of Mg- and Sc- intermetallics (~200 nm in size) with a peak hardness of 110 HV.

In this study, an extensive data set was based on existing literature records in order to enable the suitability of several predictive models, from Multiple Linear Regression (MLR) to Neural Networks (NN). The main objective was to, through regression analyses, generate model computations to correlate tensile properties (UTS- Ultimate Tensile Strength, YTS – Yield Tensile Strength and EF – Elongation-to-Fracture) to a given alloy composition and microstructural spacing. This investigation led to positive results, as the highest accuracies of the trained modules (in 80% of the database) were found to be above ~82% (UTS and EF) and a maximum of ~98% (YTS), when analyzing the results to a test data set. Later, these models were used to define trends for possible next solder alloy commercial compositions. Overall, using the standard model’s setup, the Random Forest and Decision Tree models showed the highest accuracy results, with 0.958 for YTS as opposed to 0.907 for MLR. Moreover, Multilayer Perceptron (MLP)-optimized models yielded the best results for each variable, with the highest increases in accuracy associated with the YTS and EF. The present contribution might imply an important milestone towards alloy design research based on data science guidelines to unlock the full potential of former experiments and their extensive set of results.