Abstract Understanding the hot deformation behaviors of materials has a great importance in hot working and shaping for promoting their industrial application. In this study, a Ti-based metallic glassy matrix composite was employed to study the deformation behaviors in supercooled liquid region at strain rates of 10−4/s - 10−2/s by compression tests. The stress-strain curves were composed of the overshoot behavior attributed by the glassy matrix and the work-hardening behavior caused by the dendrite phase. The nanocrystals generated within the glass matrix during hot compression deformation. The experimental data with variables of the strain rate, strain and temperature were employed to build a back-propagation artificial neural network with AdaGrad algorithm. The model’s accuracy and reliability were evaluated by examining the correlation coefficient, average absolute relative error, root mean square error and relative error. It was found that this model predicted result is consistent with the experimental data, providing a powerful approach for describing the hot deformation behaviors and working of MGMCs at high temperature.

中文翻译：

---

基于人工神经网络的钛基金属玻璃基复合材料过冷液区变形行为研究

摘要 了解材料的热变形行为对于热加工成形对于促进其工业应用具有重要意义。在这项研究中，采用钛基金属玻璃基复合材料通过压缩试验研究了应变速率为 10-4/s - 10-2/s 时过冷液体区域的变形行为。应力-应变曲线由玻璃基体引起的超调行为和枝晶相引起的加工硬化行为组成。在热压缩变形过程中在玻璃基体中产生的纳米晶体。以应变率、应变和温度为变量的实验数据，利用AdaGrad算法构建反向传播人工神经网络。通过检验相关系数、平均绝对相对误差、均方根误差和相对误差来评价模型的准确性和可靠性。发现该模型预测结果与实验数据一致，为描述MGMCs在高温下的热变形行为和工作提供了有力的方法。