徐志超,高松涛,曹增旭,等.基于机器学习的稀土镁合金本构模型修正与热变形行为研究[J].河南理工大学学报(自然科学版)，doi:10.16186/j.cnki.1673-9787. （ 2025030017）..

XU Z C, GAO S T, CAO Z X, et al． Constitutive model modification and hot deformation behavior of rare earth magnesium alloy based on machine learning[J]. Journal of Henan Polytechnic University (Natural Science)， doi: 10.16186/j.cnki.1673-9787. （ 2025030017）..

基于机器学习的稀土镁合金本构模型修正与热变形行为研究

徐志超1，高松涛1，曹增旭1，熊峰 1，王迪2，杨文举3

1.河南理工大学 材料科学与工程学院，河南 焦作 454003； 2. 河南理工大学 工商管理学院能源经济研究中心，河南 焦作 454003； 3.郑州大学 材料科学与工程学院，河南 郑州 450000

摘要: 目的 为克服Johnson-Cook模型（JC）本构方程在预测复杂非线性流动行为方面的不足，研究晶粒尺寸及织构对含长周期有序相（LPSO）稀土镁合金热加工性能的影响及变形机制。 方法本文对Mg-Zn-Y合金进行了热压缩，系统研究了Mg-Zn-Y合金的高温动态流变行为，结合机器学习技术与传统物理模型，提出了一种包含比例系数的人工神经网络辅助Johnson-Cook模型（ANN-JC），并深入探讨了Mg-Zn-Y合金在热变形过程中的显微组织特征及演化机制。结果 模型精度评估结果表明，与传统JC模型相比，ANN模型的RMSE和MAPE分别降低了59.36%和48.93%，提出的ANN-JC模型有效融合了传统本构模型的理论基础与机器学习非线性拟合能力，能够较好的补偿热变形过程中不同软化机制产生的误差，显著提升了对材料流变行为的预测精度和泛化能力。随着变形温度的升高，Mg-Zn-Y合金的动态再结晶程度显著增强。未再结晶晶粒内部通过低角度晶界的吸收与演化逐步形成高角度晶界，促进了动态再结晶晶粒的形核。随着变形温度的升高，动态再结晶的发生将导致晶粒细化和织构强度的降低，合金表现出显著的各向异性削弱效应。当变形温度上升到450°C时，织构强度出现回升，这个现象主要由于高温条件下动态再结晶对细化的作用趋于饱和，同时晶粒长大效应使晶粒间取向差减小，晶粒取向趋于一致，进而增强了合金的织构强度。

关键词:稀土镁合金；热变形；本构模型；机器学习；织构

doi: 10.16186/j.cnki.1673-9787. （ 2025030017）..

基金项目:国家自然科学基金资助项目(52103290)；河南省重点研发与推广专项(242102230050)；河南省高等学校青年骨干教师培养计划(2023GGJS056)；河南省自然科学基金面上项目(252300420037)；河南省哲学社会科学规划项目(2023CJJ145)；河南理工大学基本科研业务费专项项目(SKJYB2023-13)

收稿日期：2025-03-08

修回日期：2025-05-21

网络首发日期：2026-06-04

Constitutive model modification and hot deformation behavior of rare earth magnesium alloy based on machine learning

Xu Zhichao1, Gao Songtao1, Cao Zengxu1, Xiong Feng1, Wang Di2, Yang Wenju3

1. Department of Material Science & Engineering Henan Polytechnic University, Jiaozuo 454003, Henan, China; 2. Center of Energy Economy Study, School of Business Administration, Henan Polytechnic University, Jiaozuo 454003, Henan, China;3. Department of Material Science & Engineering, Zhengzhou university, Zhengzhou 450000, Henan, China

Abstract: Objectives To address the limitations of the Johnson-Cook (JC) constitutive equation in predicting complex nonlinear flow behavior, this study investigates the effects of grain size and texture on the hot workability and deformation mechanisms of LPSO-containing rare earth magnesium alloys. Methods The high-temperature dynamic flow behavior of LPSO-containing Mg-Zn-Y alloys was systematically studied through hot compression experiments. By integrating machine learning techniques with traditional physical models, an artificial neural network-assisted Johnson-Cook model (ANN-JC) incorporating proportional coefficients was proposed. Furthermore, the microstructural characteristics and evolution mechanisms of Mg-Zn-Y alloys during hot deformation were examined in depth. Results The evaluation results of model accuracy indicate that, compared to the traditional JC model, the RMSE and MAPE of the ANN model are reduced by 59.36% and 48.93%, respectively. The proposed ANN-JC model successfully integrates the theoretical foundation of the classical constitutive model with the nonlinear fitting capability of machine learning. This hybrid approach effectively compensates for errors arising from varying softening mechanisms during thermal deformation, thereby significantly enhancing the prediction accuracy and generalization capability for the rheological behavior of materials. As the deformation temperature increases, the degree of dynamic recrystallization (DRX) in Mg-Zn-Y alloys is significantly enhanced. Within unrecrystallized grains, the absorption and evolution of low-angle grain boundaries (LAGBs) gradually form high-angle grain boundaries (HAGBs), promoting the nucleation of DRX grains. With increasing deformation temperature, the occurrence of DRX results in grain refinement and a reduction in texture intensity, thereby weakening the alloy's anisotropic behavior. However, when the deformation temperature rises to 450°C, texture intensity increases again. This phenomenon is primarily attributed to the saturation of the grain refinement effect induced by DRX at high temperatures, coupled with grain growth, which reduces misorientation between grains and aligns their orientations, thereby reinforcing the alloy's texture intensity.

Key words: Rare earth magnesium alloy; Thermal deformation; Constitutive model; Machine Learning; Texture