Influence of Reinforcement Characteristics and Ultrasonic-Assisted Stir-Casting Conditions on AA6061/Al₂O₃/SiC Tensile Strength: Machine Learning Modeling

Abstract

The integration of machine learning (ML) with aluminium matrix composite manufacturing offers an effective approach for understanding the relationships between reinforcement characteristics, processing conditions, and mechanical performance. This study investigates the prediction of ultimate tensile strength (UTS) in AA6061 aluminium matrix nanocomposites containing Al₂O₃ and SiC nanoparticles using Gaussian Process Regression (GPR) and Least-Squares Boosting (LSBoost). A dataset comprising 240 observations was analysed using fourteen variables describing alloy composition, reinforcement concentration, particle dimensions, and ultrasonic-assisted stir-casting conditions. Eight reinforcement and processing variables were selected for predictive modelling, with 80% of the observations allocated to training and 20% to testing. Spearman correlation analysis identified SiC concentration (rₛ = 0.66) and Al₂O₃ concentration (rₛ = 0.47) as the variables most strongly associated with UTS. The predictive accuracy of both algorithms was assessed using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). GPR demonstrated superior predictive performance, achieving a testing R² of 0.9657, RMSE of approximately 2.85 MPa, and MAE of 2.28 MPa, compared with 0.9391, 3.80 MPa, and 3.12 MPa, respectively, for LSBoost. Residual analysis further indicated improved prediction consistency for GPR. LSBoost predictor-importance analysis highlighted the predominant influence of SiC and Al₂O₃ concentrations on the fitted tensile-strength response. These findings demonstrate the potential of GPR for modelling nonlinear composition–processing–property relationships and provide a computational foundation for further experimental investigation of ceramic-reinforced AA6061 composites.