Physical metallurgy-guided machine learning for strength-plasticity optimization in aluminum alloys

Yingqi Wang, Xiaolu Wei, Chenchong Wang, Wei Xu,
Physical metallurgy-guided machine learning for strength-plasticity optimization in aluminum alloys,
Materials Today Communications,
Volume 49,
2025,
113970,
ISSN 2352-4928,
https://doi.org/10.1016/j.mtcomm.2025.113970.
(https://www.sciencedirect.com/science/article/pii/S2352492825024821)
Abstract: Aluminum alloys are widely used in aerospace and transportation for their high specific strength and good machinability. However, achieving both high strength and ductility remains challenging. Traditional trial-and-error alloy design is inefficient and costly. While machine learning provides an alternative, the limited interpretability of purely data-driven models restricts their reliability. Thus, interpretable and accurate design frameworks with solid evaluation methods are needed. This study compiled a dataset of 7xxx series aluminum alloys, containing 322 data points for ultimate tensile strength and 275 for elongation. Physical metallurgy parameters such as precipitate volume fraction and precipitation driving force were incorporated to improve prediction accuracy and physical plausibility. Multiple ML and deep learning models were developed and evaluated. Results show that including physical parameters enhances model robustness and prediction reliability. A Gradient Boosting Regressor, combined with a Genetic Algorithm, was used for inverse design of composition and processing parameters. A ductility classification model served as a constraint. Experimental tests confirmed that the designed alloy exhibits better strength-ductility balance than existing benchmarks. This work offers a physics-informed data-driven framework for accelerated design of high-performance aluminum alloys and introduces a comprehensive model evaluation approach.
Keywords: Alloy design; Strength–plasticity optimization; Physical metallurgy parameter; Genetic algorithm; Machine learning