卷积神经网络与有限元分析在棒材轧制孔型参数仿真中的应用  

Application of convolutional neural network and finite element analysis to simulation of bar rolling reduction

作　　者：张芳萍[1] 张宏政 贾怀博 张帆[1] 高毅 Zhang Fangping;Zhang Hongzheng;Jia Huaibo;Zhang Fan;Gao Yi(Engineering Research Center of Heavy Machinery Ministry of Education,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学重型机械教育部工程研究中心,太原030024

出　　处：《材料与冶金学报》2024年第4期379-385,共7页Journal of Materials and Metallurgy

基　　金：山西省先进钢铁材料重点科技创新平台项目(201805D115061-2);太原科技大学研究生教育创新项目(XCX212108)。

摘　　要：为了实现智能化轧制仿真过程,提出了将卷积神经网络(CNN)与有限元分析相结合的方法,利用机器代替人工选择出最佳的轧制仿真结果,并确定最优的孔型参数.首先,将CNN用于轧件道次分类,可发现“卷积神经网络+图像预处理”测试模型精度最高,为91.84%~96%;其次,使用5种评估方法对比轧制仿真结果与理想仿真结果后可得出,均方根误差和感知哈希对仿真结果更为敏感;最后,将孔型参数朝轧制仿真结果变优方向进行修正,对比仿真结果与修正前结果后发现,第2次修正更为理想.本研究中实现了CNN对棒材轧制道次分类,利用机器识别出更优的轧制仿真结果,并确定了更适合的孔型参数,在降低人力、物力消耗,以及实现仿真智能化等方面具有重要的理论意义和实际应用价值.In order to realize the intelligent rolling simulation process,the convolutional neural networks(CNN)and finite element analysis are combined to select the best rolling simulation results and determine the optimal pass parameters by machine instead of manual work.First of all,CNN is used for the classification of rolling pass,it is found that the test model“Convolutional neural network+image preprocessing”has the highest accuracy,which ranges from 91.84%to 96%.Secondly,five evaluation methods are used to compare the rolling simulation results with the ideal simulation results,and the RMS error and perceptual hash are more sensitive to the simulation results.Finally,the pass parameters are adjusted in the direction of the optimization of the rolling simulation results,and the simulation results are compared with the pre⁃adjustment results,and the second adjustment is more ideal.In this study,CNN classification of bar rolling passes has been realized,better rolling simulation results have been identified by the machine,and more suitable pass parameters have been determined.This method has important theoretical significance and practical application value in reducing manpower and material consumption and realizing intelligent simulation.

关 键 词：卷积神经网络 有限元分析 轧制 孔型参数 智能化 

分 类 号：TG142.71[一般工业技术—材料科学与工程]