基于机器学习的大锻件拔长变形预测  

Prediction of drawing deformation for heavy forgings based on machine learning

作　　者：张梓煜 曾攀[1] 雷丽萍[1] Zhang Ziyu;Zeng Pan;Lei Liping(Department of Mechanical Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学机械工程系,北京100084

出　　处：《锻压技术》2020年第10期209-216,共8页Forging & Stamping Technology

基　　金：国家重点研发计划(2017YFB0701803)。

摘　　要：大锻件成形通常借助有限元模拟来进行研究,由于大锻件的尺寸大、工序长而导致有限元计算耗费了大量时间。因此,首先,采用有限元软件DEFORM对大锻件拔长过程进行模拟,获得成形数据,构建了19维的输入特征量和以应力、应变为输出特征量的数据集。然后,应用机器学习中的随机森林和神经网络方法对数据集进行学习,训练对应模型。最后,利用机器学习模型对一个新的拔长过程进行应力和应变分布预测,与有限元模拟结果对比后发现,这些预测结果与有限元模拟结果相近。研究表明,通过机器学习可以快速预测拔长成形结果,进而进一步分析成形质量,节省计算时间。The forming of heary forgings is usually studied by means of finite element simulation,and because of large size and long working procedure for heavy forgings,the finite element calculation takes a lot of time. Therefore,firstly the drawing process of heavy forgings was simulated by finite element software DEFORM,and the forming data were obtained to construct a data set which consisted of nineteen dimensional input characteristic variables and output characteristic variables of stress and strain. Then,the data set was learned by random forest and neural network methods in machine learning,and the corresponding model was trained. Finally,the stress and strain distributions of a new drawing process was predicted by the machine learning model. Compared with the results of finite element simulation,the predicted results were similar to those of finite element simulation. The results show that the machine learning quickly predicts the result of drawing and analyzes the forming quality to save a lot of calculation time.

关 键 词：大锻件 拔长 机器学习 随机森林 神经网络 

分 类 号：TG302[金属学及工艺—金属压力加工]