中文摘要：

采用Gleeb-1500D热模拟试验机对TiB2颗粒增强铝基复合材料进行等温压缩试验,获得了该复合材料在不同变形条件下的流变应力数据。结合试验数据,采用改进的BP网络算法要Levenberg-Marquardt算法建立了复合材料3×12×1三层网络结构模型的本构关系,并与Kumar模型计算的本构关系进行了对比分析。结果表明,神经网络模型和Kumar模型的总拟合度分别为2.1%和6.5%,两种模型建立的本构关系具有较高的精度,均能够描述该复合材料的高温变形力学行为,适用于复合材料热加工过程的数值模拟。由于Kumar模型把热变形激活能Q看作与应变变化无关的常数处理,而神经网络模型建模训练时包含了热变形激活能Q随应变改变的动态变化,因此神经网络模型的精度高于Kumar模型。

英文摘要：

The isothermal compression test of aluminum matrix composite reinforced by TiB2 particles was carried out on Gleeble-1500 D machine. The flow stress data were obtained under different deformation conditions. Based on experiment data, the constitutive relationship was separately established in the Kumar model and the model of 3×12×1 neural network by nonlinear regression method and improving BP network arithmetic-Levenberg-Marqurdt algorithm. The results show that the fitting degree of the Kumar model and artificial neural network model is 6.5% and 2.1%, respectively. The constitutive relationship of two models has higher predicted precision and can describe the hot deformation mechanical behavior for the composite and is the same with the numerical simulation in hot deformation process. Since the latter model has considered the hot deformation activation energy Q changing dynamically with true stain changing during the compression process which is regarded as a constant in the former one,and the latter model has higher precision than that of the former one.