【摘要】 目的 借助Gleeble-3500热模拟试验机进行热压缩实验，研究新型Ni-Cr-Fe基高温合金在变形温度为1 075～1 150℃、应变速率为0.001～1 s^-1条件下的流变行为。方法 采用金相显微镜和透射电子显微镜观察合金热变形显微组织。构建了基于应变补偿的Arrhenius模型和BP网络模型。结果 新型Ni-Cr-Fe基高温合金流变应力受热变形参数的影响较为显著，与变形温度呈负相关，并与应变速率呈正相关。由显微组织分析可知，在1 150℃/0.01 s^-1变形条件下，合金内部原始晶粒基本被细小的动态再结晶晶粒所取代。在0.1 s^-1/1 075℃变形条件下，可以明显观察到大量位错缠结堆积在一起；同时还能观察到由于位错堆积和迁移而形成的位错墙。当应变速率降低至0.01 s^-1时，晶粒内部位错密度显著降低且还能观察到动态再结晶晶核。利用2类模型预测了合金流变应力随应变的变化情况，其中BP神经网络模型的相关系数为0.998 5、平均相对误差为1.752 1%，预测精度较基于应变补偿的Arrhenius本构模型更高。结论 建立的BP神经网络模型更加适用于预测新型Ni-Cr-Fe基高温合金的流变应力。更多还原

【Abstract】 The work aims to study the hot deformation behavior of a novel Ni-Cr-Fe based superalloy in the temperature range of 1 075-1 150 ℃ and strain rate range of 0.001-1 s^-1 by carrying out hot compression tests on a Gleeble-3500 thermomechanical simulator. Metallographic microscope and transmission electron microscope were used to observe hot deformation microstructure of the studied alloy. At the same time, the Arrhenius constitutive model based on strain compensation and BP（back-propagation） neural network model were also established. The flow stress of a novel Ni-Cr-Fe based superalloy was significantly affected by hot deformation parameters, which was negatively correlated with deformation temperature and positively correlated with strain rate. The microstructure analysis showed that the original grains inside the studied alloy were basically replaced by fine dynamic recrystallization（DRX） grains under the deformation condition of 1 150 ℃/0.01 s^-1. Under the deformation condition of 0.1 s^-1/1 075 ℃, a large number of dislocation tangles were clearly observed. At the same time, the dislocation wall formed by dislocation accumulation and migration was also observed. As the strain rate decreased to 0.01 s^-1, it was found that the dislocation density inside the grains decreased significantly and the DRX nuclei was observed. Two kinds of models were used to predict the variation of flow stress with strain. The correlation coefficient of BP neural network model was 0.998 5 and the average relative error was 1.752 1%. The prediction accuracy of BP neural network model was higher than that of Arrhenius constitutive model based on strain compensation. The established BP neural network model is more suitable for accurately predicting the flow stress of a novel Ni-Cr-Fe based superalloy.更多还原