Machine Learning-ANN Driven Hardness Modeling in SPS-Processed AlCuCrFeMnWx High-Entropy Alloys

Abstract This study explores the design and synthesis of nanocrystalline AlCuCrFeMnWₓ high-entropy alloys through an integrated approach that combines mechanical alloying (MA) and spark plasma sintering (SPS), thereby enabling precise control over microstructural evolution and phase stability. Beyond experimental development, the work advances into a data-driven paradigm by establishing a robust artificial neural network (ANN) framework to capture the complex, nonlinear relationships between compositional variables, processing parameters, and resultant mechanical properties. A dataset comprising 20 samples, curated from experimental and literature sources, was used to develop and rigorously validate the model using exhaustive 20-fold leave-one-out cross-validation (LOOCV), incorporating six independent compositional inputs (Fe excluded due to near-perfect collinearity with Al, r = 0.9996) along with sintering temperature as a processing parameter. A reduced ANN architecture (7–5–1), employing Bayesian-regularized backpropagation, was evaluated alongside linear regression, random forest, support vector regression, and gradient boosting under an identical, leakage-free protocol with fold-wise feature standardization. Under LOOCV, the support vector regression model achieved the best generalization performance (R2 = 0.5301, RMSE = 144.52 HV, accuracy = 67.23%), followed by the ANN (R2 = 0.4084, RMSE = 162.16 HV, accuracy = 64.47%); all evaluated models’ RMSE exceeded the experimental Vickers indentation measurement uncertainty ( ± 11.75 HV), indicating that the residual error reflects genuine model limitation rather than measurement noise. This synergy between advanced processing techniques and a statistically rigorous, leakage-free machine-learning validation protocol establishes a transparent and reproducible pathway toward cost-effective alloy design, where experimental efforts are guided and refined through computational intelligence.