基于机器学习的GH738高温合金流动应力本构关系研究

Study on Flow Stress Constitutive Relationship of GH738 Superalloy Based on Machine Learning

摘要:

在变形温度980~1100 ℃、应变速率0.001~0.1 s^-1条件下，采用Gleeble 3500热模拟试验机对GH738高温合金进行等温恒应变速率压缩试验，研究了合金的流动应力行为，建立了随机森林、支持向量机和遗传算法优化反向传播网络（GA-BP）3种机器学习型本构关系模型，并对模型预测精度进行了对比分析。结果表明，GH738高温合金的流动应力随变形温度升高和应变速率降低而减小。随机森林、支持向量机和GA-BP本构关系模型的相关系数R分别为0.921、0.998和0.999，平均绝对相对误差分别为14.587%、2.112%和0.901%，说明支持向量机和GA-BP本构关系模型预测精度远高于随机森林模型，它们能较准确地预测GH738高温合金的流动应力行为，可为不同变形条件下的变形抗力和锻造吨位计算提供理论依据，也可为锻造过程数值模拟提供准确、可靠的流动应力数据。

Abstract:

Hot compression experiments were conducted on GH738 superalloy using Gleeble 3500 thermal simulation machine at deformation temperature of 980–1100 °C and strain rate of 0.001–0.1 s^-1 to study the flow stress behavior of the alloy. Three machine learning algorithms, namely random forest (RF), support vector machine (SVM), and genetic algorithm-back propagation (GA-BP) neural networks, were employed to establish constitutive relationship models for the flow stress behavior of GH738 superalloy. Subsequently, these models were compared and analyzed in terms of their predictive accuracy. The results indicate that the flow stress of GH738 superalloy decreases with the increase in deformation temperature and the decrease in strain rate. The correlation coefficients for the RF, SVM, and GA-BP constitutive relationship models are determined as 0.921, 0.998, and 0.999, while the average absolute relative errors as 14.587%, 2.112%, and 0.901%, respectively. The results demonstrate that SVM and GA-BP constitutive relationship models have better prediction accuracy than RF model in predicting the flow stress behavior of GH738 superalloy. It can provide a theoretical basis for the calculation of deformation resistance and forging tonnage under different deformation conditions, and it can also provide reliable flow stress data for numerical simulation of forging process.