基于卷积神经网络的Mg-9Gd-Y-Zn-Zr镁合金力学性能预测

Prediction of Mechanical Properties of Mg-9Gd-Y-Zn-Zr Alloy Based on Convolutional Neural Network

对不同工艺的变形稀土镁合金样品进行了系列拉伸试验,借助卷积神经网络构建模型训练,提出Mg-9GdY-Zn-Zr合金力学性能的预测模型。结果表明,测试集抗拉强度、屈服强度的相对平均误差分别为3.25%和4.77%。随机取3个Mg-9Gd-Y-Zn-Zr合金试样再次验证该模型预测力学性能,误差均小于5%,表明该卷积神经网络模型能够以较高的精度预测力学性能。细晶强化对3个试样强化的贡献分别为68.8、88.4、98.1MPa,位错强化的贡献分别为17.62、22.36、22.06MPa。

A series of tensile tests were carried out for deformed rare earth magnesium alloy samples with different processes,and a prediction model of mechanical properties of Mg-9Gd-Y-Zn-Zr alloy was put forward with the help of convolutional neural network model training.The results show that,the relative average errors of tensile strength and yield strength of the test set are 3.25%and 4.77%respectively.Three samples of Mg-9Gd-Y-Zn-Zr alloy are randomly selected to verify the predicted mechanical properties of the model again,and the errors are all within 5%,indicating that the convolutional neural network model can predict the mechanical properties with high accuracy.The contribution of fine grain strengthening to the strengthening of the three samples is 68.8 MPa,88.4 MPa and 98.1 MPa,respectively,and the contribution of dislocation strengthening is 17.62 MPa,22.36 MPa and 22.06 MPa respectively.