Machine Learning-based Investigation of Wear and Frictional Behavior in Graphite-reinforced Aluminum Nanocomposites

The researcher used a cell segmentation technique in conjunction with other image analysis methods to quantitatively retrieve and compute the cellular microstructural structures in a sub-grain size of silicon carbide (SiC)-reinforced AA2219 made by powder fusion bed (size 0.5 - 1µm). Over 83 geometric features were retrieved and statistically analyzed using ML (Machine learning) techniques to examine the structure-property relationships in SiC-reinforced AlSi20Mg nanocomposites. These sub-grain cellular microstructure properties were utilized to develop hardness and relative mass density analytical models. Using principal component analysis (PCA), authors could narrow down the three variables. While all of the AlSi20Mg nanocomposite samples had identical Al-Si eutectic microstructures, the mechanical properties, such as hardness and relative mass density, varied widely depending on the laser parameters used to create them. Extra Tress regression models that attempted to predict hardness had a close error rate of 2.47%. Using a regression model based on Decision Trees, authors could predict relative mass density to within 0.42 standard deviations. The established models are shown to be capable of predicting the relative hardness and relative mass density of AlSi20Mg nanocomposites. The structure identified in this study has applications for controlling the mechanical properties of PFB (powder fusion beds) and could be applied to other additively manufactured alloys and composites.