Machine learning-driven analysis of erosion resistance in ZrSiO4/Al2O3 reinforced A356 hybrid composites: Integration of CART algorithm with Taguchi optimization

Abstract

Aluminum alloy A356 has been increasingly important in the production of aircraft wings and fuselage skins in recent years, primarily because of its lightweight nature and specialized strength characteristics. This study aims to produce and investigate the erosion resistance of A356 hybrid composites by incorporating reinforcements such as zirconium silicate (ZrSiO₄) and aluminium oxide (Al₂O₃) using the Stir Casting Process (SCP) technology. Microhardness tests were performed on different combinations of dual reinforcement to evaluate the erosion resistance of the specimen with the highest microhardness value. The erosion performance of this rigid sample is quantified under parameters such as jet angle, jet pressure, jet distance, mesh number of erodent and impinging duration. These parameters were optimized using the Taguchi technique, and their contributions were analyzed using ANOVA. Additionally, CART regression analysis was employed to predict the erosion rates and evaluate the relative importance of erosion parameters. Confirmation tests conducted on optimized parameters validated both models, with the regression model showing superior accuracy (average error 0.34 %) compared to the CART model (average error 1.18 %). CART results highlight the jet angle and erodent mesh number as the most significant predictors of erosion rate, with jet angle contributing 100 % and mesh number contributing 24.2 % to model performance. Surface morphology analysis revealed less deterioration in terms of grain refinement and uniform reinforcement particle distribution compared to the base A356 matrix.