Research on predicting the thermocompression deformation behavior of Mg–Li matrix composite using machine learning and traditional techniques

Artificial intelligence and machine learning (ML) technologies have emerged as powerful tools for analyzing the thermal compression deformation behavior of metal matrix composites, offering significant potential to optimize their plastic deformation processing techniques. In this study, the Al3La/LAZ532 composite based on in-situ self-reaction technology was successfully prepared by adding La2O3 particles. Then, the thermal compression flow behavior of the as-cast composite was comparatively researched using a traditional Arrhenius model and advanced machine learning methods (Linear Regression, AdaBoost, Random Forest, and XGBoost). The flow stresses were predicted under various thermal operating conditions, and the performance of all models was assessed using root mean square error (RMSE), coefficient of determination (R2) and mean absolute error (MAE). Analysis shows that the Random Forest model outperforms traditional models and other ML methods in predictive accuracy, achieving an R2 of 0.97, an MAE of 5.8, and an RMSE of 7.07. Additionally, the stable hot working zone for the composite at 300 °C and a strain rate of 0.001–0.01 s⁻1 can be identified by combining the microscopic structure with the macroscopic hot working map. This zone is characterized by a uniformly distributed, fine-grained structure with high levels of dynamic recrystallization (DRX). The texture with the axis of the grains aligned parallel to the compression direction (CD) is formed in the thermally deformed microstructure, and the DRX mechanism is discontinuous dynamic recrystallization (DDRX).