China Mechanical Engineering

Previous Articles     Next Articles

In-process Tool Wear Monitoring Model Based on LSTM-CNN

HE Yan1;LING Junjie1;WANG Yulin2;LI Yufeng1;WU Pengcheng1;XIAO Zhen1   

  1. 1.State Key Laboratory of Mechanical Transmission,Chongqing University,Chongqing,400044
    2.School of Mechanical Engineering,Nanjing University of Science and Technology,Nanjing,210094
  • Online:2020-08-25 Published:2020-09-17

基于长短时记忆卷积神经网络的刀具磨损在线监测模型

何彦1;凌俊杰1;王禹林2;李育锋1;吴鹏程1;肖圳1   

  1. 1.重庆大学机械传动国家重点实验室,重庆,400044
    2.南京理工大学机械工程学院,南京,210094
  • 基金资助:
    国家科技重大专项(2018ZX04002001-008)

Abstract: To improve the accuracy of in-process tool wear monitoring in machining processes, an in-process tool wear monitoring model was proposed based on LSTM-CNN. In the monitoring model, the vibration, force and acoustic emission signals during the cutting processes of the tool was collected respectively by vibration, force and acoustic emission sensors, the collected datum were essentially time series datum. Considering the sequence and multidimensional characteristics of the collected datum, the LSTM-CNN performed sequence and multidimensional feature extraction on the collected datum, and used linear regression to map the features to the tool wear values. The validity and feasibility of the model were verified by experiments. Compared with other methods, the accuracy of the model is greatly improved.

Key words: tool wear monitoring, long short term memory(LSTM) neural network, convolution neural network(CNN), feature extraction

摘要: 为了提高机械加工过程中刀具磨损在线监测的准确性,提出了一种基于长短时记忆卷积神经网络(LSTM-CNN)的刀具磨损在线监测模型。在该监测模型中,通过振动、力、声发射传感器对刀具切削过程中的振动、力和声发射信号进行采集,采集的数据其本质为时间序列数据。考虑采集数据的序列和多维度特性,采用LSTM-CNN网络对采集的数据进行序列和多维度特征提取,利用线性回归实现特征到刀具磨损值的映射。通过实验验证了该模型的有效性和可行性,模型的精度较其他几种方法有了较大的提高。

关键词: 刀具磨损监测, 长短时记忆神经网络, 卷积神经网络, 特征提取

CLC Number: