4-15. Prediction of the microstructural evolution and mechanical properties of mg alloys based on machine learning

Leyun Wang, Zhouruo Tong, Gaoming Zhu, Yanwei Liu, Xuenan Xu, Xiaoqin Zeng

School of materials science and engineering, Shanghai Jiaotong University

Abstract: Deformation microstructure of Mg alloys is mainly determined by twinning activity. Twin nucleation is affected by many factors, such as Schmid factor for twinning, grain size, and grain boundary strain transfer. However, there is still no effective criterion for the prediction of twin nucleation. Here, we use a machine learning method to establish a predictive model of twin nucleation. From a Mg-Ca alloy extruded bar, tensile specimens were taken in three different directions, named E-0, E-45, E-90. Tensile experiments were performed on these specimens in a scanning electron microscope, and EBSD was used to identify those grains that nucleated twins. We used 572 grains in the E-45 specimen as the training set, and established five models based on algorithms such as decision tree, artificial neural network, and support vector machine. The twins in E-0 and E-90 were then predicted using those models. In the end, the best model achieves predictive accuracy of 87%.

The macroscopic mechanical properties of Mg alloys mainly depend on the alloy composition and processing. We summarized the effects of alloying elements thermal processing parameters on the tensile properties of AZ31. The artificial neural network and support vector machine algorithms were used to establish models to predict the yield strength, tensile strength and elongation of the material.

基于机器学习的镁合金组织演变与力学性能预测

王乐耘，童周诺，朱高明，刘言伟，徐薛楠，曾小勤

上海交通大学材料科学与工程学院，上海 200240

摘要: 镁合金变形微观组织的演变主要由孪晶决定。孪晶形核受多个因素的影响，包括孪晶系的Schmid因子、晶粒尺寸、晶界应变传递等。然而，目前对于孪晶形核仍然缺乏有效的判定依据。在此，我们采用了机器学习方法来建立孪晶形核的预测模型。对Mg-Ca合金挤压棒材，沿三个不同方向取拉伸样品，分别命名为E-0,E-45,E-90。通过在扫描电镜中对三个样品分别进行拉伸实验，再利用EBSD来标定产生孪晶的晶粒。我们以E-45样品中的572个晶粒作为训练集，通过决策树、人工神经网络、支持向量机等算法建立了五个模型。然后以这些模型对E-0和E-90样品中的孪晶进行预测。最终，表现最好的模型预测准确率达到了87%。

镁合金的宏观力学性能主要取决于合金成分与加工工艺。我们总结了微量合金元素种类、含量以及热加工工艺参数对AZ31合金拉伸性能的影响，通过人工神经网络和支持向量机建立模型，实现了对材料屈服强度、拉伸强度以及延伸率的预测。