Original Research Article

Machine Learning–Based Flow Stress Prediction and Processing Map Optimization of Al/SiCp Metal Matrix Composites

---

Abstract

This study establishes a high-accuracy machine learning framework for predicting flow stress in 2124 Al/10% SiCp composites during hot deformation. Eight advanced algorithms—including XGBoost, Gradient Boosting, and an Enhanced Ensemble model—were thoroughly optimized and evaluated using experimental data spanning 300–500  and strain rates of 0.001–1.0  . Physics-informed feature engineering, centered on the Zener–Hollomon parameter, significantly enhanced prediction accuracy. The XGBoost model achieved exceptional performance (  = 0.9985, MAPE = 3.26%), outperforming traditional regression by 75% and fuzzy logic by 46%. Interpretability analysis confirmed the dominance of thermally activated deformation mechanisms, validating the physical consistency of the models. The predicted flow stresses enabled the development of reliable processing maps, identifying optimal hot working conditions at 400–450  and 0.01–0.1  . This work demonstrates a robust, interpretable, and industrially applicable data-driven approach for optimizing the thermomechanical processing of metal matrix composites.

Keywords:

• Metal Matrix Composites,
• Flow Stress Prediction,
• Machine Learning,
• Ensemble Methods,
• Processing Maps.,
• Optimization.,
• Processing Maps,
• Optimization