基于深度学习的螺旋叶片螺距预测  

Pitch prediction of spiral blade based on deep learning

作　　者：陈津南 王匀[1] 郝浩然 郑南 尹锐 CHEN Jin-nan;WANG Yun;HAO Hao-ran;ZHENG Nan;YIN Rui(School of Mechanical Engineering,Jiangsu University,Zhenjiang 212013,China;Zhenjiang Banghe Screw Manufacting Co.,Ltd.,Zhenjiang 212009,China)

机构地区：[1]江苏大学机械工程学院,江苏镇江212013 [2]镇江市邦禾螺旋制造有限公司,江苏镇江212009

出　　处：《塑性工程学报》2023年第1期70-76,共7页Journal of Plasticity Engineering

基　　金：国家自然科学基金面上资助项目(51575245);镇江市重点研发计划(KZ2020001)。

摘　　要：为研究工艺参数与螺旋叶片几何尺寸的映射关系和预测模型,以螺旋叶片螺距为预测目标,根据生产适应性及权重法确定喂料高度、相对平动量、辊子间距、带宽和带厚为关键的轧制工艺影响因素,结合正交试验和数值模拟分析,建立了工艺参数与螺距的关系数据库,在此基础上构建并比对了SVR、DNN和SNN这3种预测模型。结果表明,DNN模型在训练集与预测集上的预测精度和泛化性能更佳,在预测集上MAR、MAE、RMSE、R^(2)以及P_(earson)的值分别为3.86%、4.91、6.26、0.998以及0.95;与实测螺距比对,相对误差均小于5%,达到了预测效果,为材料成形质量预测提供了方法。To study the mapping relationship and prediction model between the process parameters and the geometric dimensions of spiral blade,with the pitch of spiral blade as the prediction target,according to the production adaptability and weight method,the feeding height,the relative translation,the roller spacing,the belt width and the belt thickness were determined as the key factors affecting the rolling process.Combined with the orthogonal test and numerical simulation analysis,the relationship database between the process pa-rameters and the pitch was established.Three prediction models of SVR,DNN and SNN were constructed and compared based on the da-tabase.The results show that the DNN model has better prediction accuracy and generalization performance in the training set and predic-tion set.The values of MAR,MAE,RMSE,R^(2) and P_(earson) in the prediction set are 3.86%,4.91,6.26,0.998 and 0.95,respectively.Compared with the measured pitch,the relative error is less than 5%,which achieves the prediction effect and provides a method for the prediction of material forming quality.

关 键 词：螺旋叶片 螺距 锥辊异面轧制 深度学习 预测 

分 类 号：TG335.56[金属学及工艺—金属压力加工]