Data–Driven Stress Prediction and Microstructure Characterization During Hot Deformation of Al–Zn–Mg–Cu Alloys

Abstract

A data-driven model for stress prediction of hot-deformed Al–Zn–Mg–Cu alloys was developed. The model utilized 9397 datasets, encompassing 22 features including alloy composition, homogenization treatment parameters and hot deformation parameters. Machine learning methods, including Linear Regression, Random Forest Regression, Decision Tree (DT), Artificial Neural Network, Support Vector Machine, and Gaussian Process Regression, are used to develop models to predict flow stress. Through data preprocessing and feature selection, 19 key features were identified, and a data partition ratio of 8:2 for training-to-test sets was found to yield optimal model performance, with an adjusted coefficient of determination (R²) of 0.93329 and an outlier-bias-bias error ratio of 0.00851. The developed models were used to predict the flow stress of the Al–6.3Zn–2.5Mg–2.6 Cu–0.11Zr alloy upon hot compression with interpolation and extrapolation strategies. The results indicated that the RF and DT models demonstrated excellent stability and generalization capability in predicting the alloy’s flow behavior. Microstructure changes at various deformation temperatures and strain rates were characterized, and the correlation among flow stresses, deformation parameters and microstructure was analyzed, providing deeper insights into the hot deformation behavior of Al–Zn–Mg–Cu alloys and potentially guiding the process optimization for industrial applications.

Graphical Abstract