基于BP神经网络预测的TC4热加工图  

Predicted Processing Map of TC4 Titanium Alloy Based on BP Neural Network

作　　者：岳远旺[1] 温彤[1] 刘澜涛[1] 刘磊[1] 陈世[1] 吴颖[1] 

机构地区：[1]重庆大学材料科学与工程学院,重庆400044

出　　处：《稀有金属》2014年第4期567-573,共7页Chinese Journal of Rare Metals

基　　金：国家工信部科技重大专项项目(2012ZX04010-081)资助

摘　　要：在Gleeble-1500热模拟实验机上对多组TC4钛合金试样进行热压缩实验,获得了变形温度在1053~1273 K、应变速率在0.01~10.00s-1情况下的真应力-应变曲线。通过BP神经网络对实验数据进行训练,建立了流变应力与应变、应变速率和温度相对应的预测模型,并对该模型的预测性能进行评估验证,采用预测数据构造了预测加工图,最后结合微观组织对预测加工图的可行性进行验证。结果表明,预测数据和实验数据的相关系数R为0.99886,平均相对误差为-0.21%,相对误差标准偏差为2.48 MPa,此模型具有良好的预测性能。预测加工图与实验加工图能够很好的吻合,通过预测加工图对材料的可加工性能进行预测,在一定程度上可以解决实验数据不足的缺陷。真应变为0.916的预测加工图大致分为A,B,C3个区域。失稳A区η值出现极小值（-0.16）,应变速率较高时,材料局部发生动态再结晶,出现局部变形失稳的现象;应变速率较低时,组织很不均匀,易失稳。稳定B区具有较大的η值,并出现极大值（0.45）,其α相球化效果显著、组织均匀,在相界处出现一定数量的细小等轴组织和较大比例的片状组织,确定此区为最优加工区。稳定C区α相球化效果比较明显,可作为加工区。The hot deformation behavior of TCA titanium alloy was investigated based on the compression test on a Gleeble-1500 simulator from 1053 to 1273 K with the strain rate from 0.01 to 10.00 s^-1 and the true stress-strain curve was obtained. The relevant prediction models among flow stresses and strain, strain rate, and temperature were established by samples training based on BP neural network. The prediction performance of this model was evaluated and the processing map of these forecast data was constructed, then the feasibility of the predicted map was verified by the mierostructure. The results showed that the correlation coefficient R between predicted and experimental values reached 0.99886, while the average relative error was -0.21% and the root mean square error was 2. 48 MPa, so the model had a strong prediction performance. The predicted processing map agreed well with the experimental processing map. The projections of materials＇ workability through processing map, to a certain extent, could solve the shortcoming of lacking experimental data. The predicted map at the strain of 0.916 was roughly divided into A, B and C three regions. In the instability Region A where η reached the minimum （ -0. 16）, local deformation instability might occur with the appearance of local dynamic reerystallization at high strain rate; while the instability of the alloy was apt to occur due to the uneven organization at low strain rate. In the stability Region B, 7/had the maximum value （0.45）, the spheroidizing effect of α phase was obvious and the mierostrueture was homogeneous. A certain number of fine equiaxed structure and high proportion of lamellar structure appeared at the phase boundary. Such a region could be considered as the optimal processing zone. The stability Region C had relatively obvious effect of α phase globalization and could be used as stable processing zone.

关 键 词：TC4钛合金 流变应力 神经网络 加工图 微观组织 

分 类 号：TG146.23[一般工业技术—材料科学与工程]