孙岑花,施文鹏,王津,等.基于Arrhenius与机器学习的TC21钛合金本构模型研究[J].精密成形工程,2024,16(8):102-110.
SUN Cenhua,SHI Wenpeng,WANG Jin,et al.Constitutive Model of TC21 Titanium Alloy Based on Arrhenius Equation and Machine Learning[J].Journal of Netshape Forming Engineering,2024,16(8):102-110. |
| 基于Arrhenius与机器学习的TC21钛合金本构模型研究 |
| Constitutive Model of TC21 Titanium Alloy Based on Arrhenius Equation and Machine Learning |
| 投稿时间：2024-01-19 |
| DOI：10.3969/j.issn.1674-6457.2024.08.012 |
| 中文关键词: TC21钛合金 本构模型 Arrhenius方程 机器学习 支持向量机 |
| 英文关键词: TC21 titanium alloy constitutive model Arrhenius equation machine learning support vector machine |
| 基金项目:国家自然科学基金项目(12062016) |
|
| 摘要点击次数: 268 |
| 全文下载次数: 43 |
| 中文摘要: |
| 目的 比较Arrhenius方程和机器学习方法在TC21钛合金本构模型建立中的优劣，为TC21钛合金在实际工程应用中的性能预测、优化设计和安全评估提供理论指导。方法 通过使用Gleeble-3500热模拟机，获取了锻态TC21钛合金在不同温度和应变速率下的真实应力应变数据。基于实验结果，分别采用Arrhenius方法和支持向量机方法建立了相应的本构模型。相较于基于热力学原理的Arrhenius本构方程，采用支持向量机方法的本构模型更为先进。该模型能够从有限的数据中深入挖掘材料性能与温度、应变速率之间的复杂非线性关系，从而更准确地预测TC21钛合金在不同条件下的力学性能。为了全面评估这2种模型的预测准确性，计算了它们的模型相关系数和均方根误差。结果 研究结果表明，基于机器学习的本构模型在预测TC21钛合金的应力应变行为方面展现出显著的优势。其相关系数高达0.977 4，远高于Arrhenius模型的0.931 7。在评估预测精度的均方根误差上，机器学习方法也表现出色，仅为5.49，相较于Arrhenius模型的20.67显著降低。结论 利用机器学习方法建立的TC21钛合金本构模型具有更高的精度和可靠性。在实际工程应用中，这将为钛合金的性能预测、优化设计和安全评估提供更为准确的科学依据。 |
| 英文摘要: |
| The work aims to compare the advantages and disadvantages of Arrhenius equation and machine learning method in the constitutive model establishment of TC21 titanium alloy to provide theoretical guidance for the property prediction, optimization design and safety evaluation of TC21 titanium alloy in practical engineering applications. The true stress-strain data of wrought TC21 titanium alloy at different temperatures and strain rates were obtained by Gleeble-3500 thermal simulator. Based on the experimental results, the corresponding constitutive models were established by Arrhenius method and support vector machine method respectively. Compared with the Arrhenius constitutive equation based on thermodynamic principle, the constitutive model established by support vector machine method was more advanced. The model could dig the complex nonlinear relationship between material properties, temperature and strain rate from the limited data, so as to predict the mechanical properties of TC21 titanium alloy under different conditions more accurately. In order to comprehensively evaluate the prediction accuracy of these two models, their model correlation coefficients and root-mean-square errors were calculated. The results showed that the constitutive model based on machine learning had a significant advantage in predicting the stress-strain behavior of TC21 titanium alloy. The correlation coefficient was as high as 0.977 4, which was much higher than 0.931 7 in Arrhenius model. The machine learning method also performed well in evaluating the root-mean-square error of prediction accuracy, which was only 5.49, significantly lower than 20.67 in the Arrhenius model. The constitutive model of TC21 titanium alloy established by machine learning method has higher accuracy and reliability. In practical engineering applications, this will provide a more accurate scientific basis for the property prediction, optimal design and safety evaluation of titanium alloys. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
渝公网安备 50010702501719号