China Mechanical Engineering ›› 2022, Vol. 33 ›› Issue (05): 583-588.DOI: 10.3969/j.issn.1004-132X.2022.05.007

Previous Articles     Next Articles

Prediction of Surface Roughness for Cryogenic Milling TC17 Titanium Alloys

LEI Yong;ZHAO Wei;HE Ning;LI Liang   

  1. College of Mechanical and Electrical Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing,210016
  • Online:2022-03-10 Published:2022-03-24

TC17钛合金低温铣削表面粗糙度预测

雷勇;赵威;何宁;李亮   

  1. 南京航空航天大学机电学院,南京,210016
  • 通讯作者: 赵威(通信作者),男,1977年生,教授、博士研究生导师。研究方向为高性能加工技术与装备、可持续制造技术。E-mail:nuaazw@nuaa.edu.cn。
  • 作者简介:雷勇,男,1998年生,硕士研究生。研究方向为先进切削技术。E-mail:3152326794@qq.com。
  • 基金资助:
    国家重点研发计划(2018YFB2002202)

Abstract: The cryogenic milling tests of TC17 titanium alloy were carried out to evaluate the machined surface roughness under varying cutting conditions. Firstly, the regression analysis method was used to establish an empirical model of the surface roughness, the effects of jet temperature, feed per tooth, milling speed, and radial cutting width on the surface roughness were investigated. And then a prediction model based on BP neural network of surface roughness was established and compared with the empirical model. The results show that the empirical model depicts a strong correlation(R2=0.92) between surface roughness and parameters. The most influential factor on surface roughness is feed per tooth, followed by the jet temperature, the radial cutting width and the milling speed, respectively. The mean square error between the predicted and the experimental values is as 1.73×10-4 μm2, and the maximum relative error is as 8.81%, and the error variation changes more significantly. For the neural network model, the mean square error between the predicted and the test value is as 3.53×10-5 μm2, the maximum relative error is as 3.64%, and the error changes uniformly. Compared with the empirical model, the neural network model has higher prediction accuracy and generalization ability, and may predict the effects of various parameters on surface roughness better.

Key words:  , TC17 titanium alloy, cryogenic milling, surface roughness, empirical model, BP neural network model

摘要: 进行了TC17钛合金低温铣削试验,研究了不同切削条件下的已加工表面粗糙度。采用回归分析方法建立了表面粗糙度经验模型,研究了射流温度、每齿进给量、铣削速度和径向切削深度对表面粗糙度的影响规律。基于BP神经网络建立了表面粗糙度预测模型,并与经验模型进行了对比分析。研究结果表明,基于经验模型表面粗糙度值与参数间存在强相关性(R2=0.92),对表面粗糙度影响最大的因素为每齿进给量,然后依次是射流温度、径向切削深度、铣削速度,预测值与试验值均方误差为1.73×10-4 μm2,最大相对误差为8.81%,误差变化幅度较大;而基于神经网络模型的预测值与试验值均方误差为3.53×10-5 μm2,最大相对误差为3.64%,误差变化幅度较小,与经验模型相比,神经网络模型的预测精度和泛化能力更高,可更好地实现各参数对表面粗糙度影响的预测。

关键词: TC17钛合金, 低温铣削, 表面粗糙度, 经验模型, BP神经网络模型

CLC Number: