• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2019, Vol. 55 ›› Issue (14): 1-10.doi: 10.3901/JME.2019.14.001

• 仪器科学与技术 •    下一篇


李宏坤, 郝佰田, 代月帮, 杨蕊   

  1. 大连理工大学机械工程学院 大连 116024
  • 收稿日期:2018-07-02 修回日期:2019-04-25 出版日期:2019-07-20 发布日期:2019-07-20
  • 作者简介:李宏坤,男,1974年出生,教授,博士研究生导师。主要研究方向为机械设备动态分析与故障诊断,信号处理。E-mail:lihk@dlut.edu.cn
  • 基金资助:

Wear Status Recognition for Milling Cutter Based on Compressed Sensing and Noise Stacking Sparse Auto-encoder

LI Hongkun, HAO Baitian, DAI Yuebang, YANG Rui   

  1. School of Mechanical Engineering, Dalian University of Technology, Dalian 116024
  • Received:2018-07-02 Revised:2019-04-25 Online:2019-07-20 Published:2019-07-20

摘要: 数控机床在加工过程中,刀具磨损会对被加工零件的表面质量、尺寸精度产生巨大影响,而传统依靠切削力系数来分析刀具磨损的方法,需要在工作台上安装额外测力装置,这将干扰机床正常加工,限制被加工零件尺寸,引起加工质量降低等问题,限制了其在实际工业环境中的应用。针对上述问题,提出利用主轴电流结合深度学习网络识别铣刀磨损状态的监测方法。首先,理论论证利用主轴电流代替切削力识别刀具磨损的可行性;然后,利用压缩感知对电流信号的频域数据进行数据压缩,其中为提高网络的鲁棒性,对观测信号添加高斯白噪音;最后,将压缩后的数据输入堆栈稀疏自编码网络,利用有监督学习与无监督学习相结合的方法,提取刀具磨损所引起的特征信息,用于表征刀具磨损程度。试验结果表明,该方法可以有效对铣刀磨损程度进行识别。

关键词: 堆栈稀疏自编码器, 铣刀磨损, 压缩感知, 主轴电流

Abstract: In the machining process of CNC machine tools, the state of tool wear has a great influence on the surface quality and dimensional accuracy of the parts being machined. However, the traditional method of relying on cutting forces to analyze tool wear requires the installation of additional force measuring devices on the workbench. This will interfere with the normal machining of the machine tool, limit the size of the part being machined, cause the processing quality to be reduced and other issues, and limit its application in the actual industrial environment. Aiming at the above problems, a monitoring method for identifying the wear state of the milling cutter by using the spindle current combined with the deep learning network is proposed. Firstly, the feasibility of using spindle current instead of cutting force to identify tool wear is demonstrated. Then, the frequency domain data of the current signal is compressed by compressed sensing, and Gaussian white noise is added to the observation signal to improve the robustness of the network. Finally, the compressed data is input into a stack sparse auto-encoder, which uses a combination of supervised learning and unsupervised learning to extract feature information caused by tool wear and to characterize tool wear. The experimental results show that this method can effectively monitor the wear state of the milling cutter.

Key words: compressed sensing, milling cutter wear, spindle current, stack sparse auto-encoder