北京航空航天大学学报 ›› 2011, Vol. 37 ›› Issue (3): 364-367,373.

• 论文 • 上一篇    下一篇

基于主元分析和BP神经网络对刀具VB值预测

聂鹏, 谌鑫   

  1. 沈阳航空航天大学 机电工程学院, 沈阳 110136
  • 收稿日期:2010-04-20 出版日期:2011-03-31 发布日期:2011-04-01
  • 作者简介:聂 鹏(1972-),男,吉林省吉林人,副教授,niehit@163.com.
  • 基金资助:

    沈阳市人才引进专项基金资助项目(07SYRC04)

Prediction of tool VB value based on PCA and BP neural network

Nie Peng, Chen Xin   

  1. School of Mechanical Engineering, Shenyang University of Aeronautics and Aviation, Shenyang 110136, China
  • Received:2010-04-20 Online:2011-03-31 Published:2011-04-01

摘要: 对声发射信号进行5层小波分解提取6个频段的能量值,把它与切削速度、切削深度、进给量和切削时间一起作为刀具状态的特征向量.通过主元分析进行降维、消除特征向量间的相关性后,把得到的主元作为BP(Back Propagation)神经网络的输入向量.BP神经网络应用改进的LM(Levenberg-Marquart)算法进行学习,利用输入向量对网络进行训练后,实现对刀具后刀面磨损量VB的预测.实验结果显示:基于主元分析和LM算法改进的BP神经网络建立的预测系统,网络输出与实测VB值的误差0.03以内;根据预测VB值的范围可判别出刀具的不同状态.

Abstract: Five layers of wavelet decomposition was applied on acoustic emission signals for extracting the acoustic emission(AE) signals energy value of six bands. Energy value and cutting speed, cutting depth, feed rate, cutting time were turned into state feature vectors of tool wear. The principal component analysis was used to reduce dimension and eliminate the correlation between the feature vectors. The principal components were seen as back propagation(BP) neural network input vector. Improved Levenberg-Marquart (LM) algorithm was used to BP neural network for learning, input vectors were trained for BP neural network. Then, the BP neural network would realize the forecast of tool flank wear VB value. The results indicate that the VB value forecast system based on principal component analysis (PCA) and the improved BP neural network with LM algorithm can accurately predict the tool flank wear VB value within the error range 0.03. The different states of tool wear can be judged according to the VB value.

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发