一维残差卷积神经网络的刀具磨损识别方法研究

杨斌; 樊志刚; 王建国; 王民; 李志星

doi:10.13433/j.cnki.1003-8728.20200525

一维残差卷积神经网络的刀具磨损识别方法研究

doi: 10.13433/j.cnki.1003-8728.20200525

杨斌^{1, 2,},
樊志刚^{1, 2},
王建国^{1, 2},
王民³,
李志星^{1, 2}

1.
内蒙古科技大学机械工程学院，内蒙古包头　014010
2.
内蒙古自治区机电系统智能诊断与控制重点实验室，内蒙古包头　014010
3.
北京工业大学机械工程与应用电子技术学院先进制造技术北京重点实验室，北京　100124

基金项目: 国家自然科学基金项目（51865045，51805275）、内蒙古自然科学基金重大项目（2018ZD06）及内蒙古自治区高等学校科学研究项目（NJZY21380）

详细信息

作者简介:
杨斌（1983−），讲师，硕士，研究方向为设备状态监测与故障诊断，hbyjb@126.com

中图分类号: TG71
计量
- 文章访问数: 97
- HTML全文浏览量: 182
- PDF下载量: 26
- 被引次数: 0
出版历程
- 收稿日期: 2020-12-04
- 刊出日期: 2023-02-04

Research on Tool Wear Recognition Method based on One-dimensional Residual Convolutional Neural Network

YANG Bin^{1, 2
,},
FAN Zhigang^{1, 2},
WANG Jianguo^{1, 2},
WANG Min³,
LI Zhixing^{1, 2}

1.
School of Mechanical Engineering, Inner Mongolia University of Science and Technology, Baotou 014010, Inner Mongolia, China
2.
Inner Mongolia Key Laboratory of Intelligent Diagnosis and Control of Mechatronic Systems, Baotou 014010, Inner Mongolia, China
3.
Beijing Key Laboratory of Advanced Manufacturing Technology, College of Mechanical Engineering and Applied Electronics Technology, Beijing University of Technology, Beijing 100124, China

摘要

摘要: 传统的机器学习方法对于刀具磨损进行监测时需要人为提取特征，并且在刀具磨损监测过程出现所需时间较长、精度低等问题。本文提出基于一维残差卷积神经网络的刀具磨损状态识别方法。对原始振动信号进行小波包阈值降噪、快速傅里叶变换处理后，将生成的频谱数据作为残差卷积神经网络模型的输入，通过卷积连接、残差连接和融合等操作自动进行特征提取，最后与刀具磨损状态进行匹配。结果表明：与目前常用的其它神经网络相比较，本文所提出的方法在多次测试中后平均准确率提高了0.6%，训练耗时对于频谱图输入降低30%，具有流程简单、准确率更高的特点，相比于其他方法更有优势。
- 振动信号 /
- 残差连接 /
- 刀具磨损 /
- 卷积神经网络
Abstract: The traditional machine learning methods used to tool wear monitoring often need to manually extract features, and require a long time and have low accuracy in the process of tool wear monitoring. In this paper, a tool wear condition recognition method is proposed based on one-dimensional residual convolution neural network. Firstly, the original vibration signal was processed by wavelet packet threshold denoising and fast Fourier transform. Then, the generated spectrum data was taken as a input of the residual convolution neural network model, and the feature was extracted automatically through convolution connection, residual connection and fusion, and finally matched with the tool wear state. The results show that compared with other commonly used neural networks, the proposed method in this paper improves the average accuracy by 0.6% after multiple tests and reduces the training time by 30% for spectrogram input. Thus the proposed method has the characteristics of simple process and higher accuracy, and has more advantages than other methods.
- vibration signal /
- residual connection /
- tool wear /
- convolutional neural network

HTML全文

图 1 典型卷积神经网络模型

下载: 全尺寸图片幻灯片

图 2 残差连接结构

下载: 全尺寸图片幻灯片

图 3 小波包阈值降噪效果

下载: 全尺寸图片幻灯片

图 4 不同磨损阶段的幅频图

下载: 全尺寸图片幻灯片

图 5 一维残差卷积神经网络模型

下载: 全尺寸图片幻灯片

图 6 刀具磨损监测系统结构图

下载: 全尺寸图片幻灯片

图 7 一维残差卷积神经网络训练过程

下载: 全尺寸图片幻灯片

图 8 45#钢测试过程曲线

下载: 全尺寸图片幻灯片

图 9 测试混淆矩阵结果

下载: 全尺寸图片幻灯片

表 1 Cr12频谱数据集样本数量

磨损状态	训练集	测试集
初期磨损	273	117
正常磨损	392	168
急剧磨损	301	129

下载: 导出CSV

表 2 不同模型测试结果分析

序号	网络模型	迭代次数	特征集数	耗时/s
1	一维卷积网络	3500	512	900
2	本文方法	3500	512	1260
3	Alexnet	3500	128*128	1800
4	Lstm	3500	128	1238

下载: 导出CSV

表 3 对比试验准确率 %

类别	1	2	3	4	5	平均准确率
一维卷积网络	99.03	99.28	99.28	99.03	99.28	99.18
Alexnet	99.28	99.28	99.03	99.28	99.28	99.23
Lstm	95.65	96.38	97.10	95.77	96.38	96.256
本文方法	99.76	100	99.52	99.76	100	99.808

下载: 导出CSV

表 4 45#钢频谱数据集样本数量

磨损状态	训练集	测试集
初期磨损	414	138
正常磨损	1314	438
急剧磨损	990	330

下载: 导出CSV

参考文献(17)

[1]	贾秀杰, 李剑峰, 孙杰. 刀具钝化对切削力及表面粗糙度的影响[J]. 计算机集成制造系统, 2011, 17(7): 1430-1434 JIA X J, LI J F, SUN J. Influence of cutting tool blade passivation on cutting force and surface roughness[J]. Computer Integrated Manufacturing Systems, 2011, 17(7): 1430-1434 (in Chinese)
[2]	REHORN A G, JIANG J, ORBAN P E. State-of-the-art methods and results in tool condition monitoring: a review[J]. The International Journal of Advanced Manufacturing Technology, 2005, 26(7-8): 693-710 doi: 10.1007/s00170-004-2038-2
[3]	MOHANRAJ T, SHANKAR S, RAJASEKAR R, et al. Tool condition monitoring techniques in milling process –a review[J]. Journal of Materials Research and Technology, 2020, 9(1): 1032-1042 doi: 10.1016/j.jmrt.2019.10.031
[4]	桑宏强, 张新建, 刘丽冰, 等. 基于工件纹理和卷积神经网络的刀具磨损检测[J]. 组合机床与自动化加工技术, 2019(7): 60-63,68 SANG H Q, ZHANG X J, LIU L B, et al. Tool wear detection based on workpiece texture and convolutional neural network[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2019(7): 60-63,68 (in Chinese)
[5]	鄢仁武, 林穿, 高硕勋, 等. 基于小波时频图和卷积神经网络的断路器故障诊断分析[J]. 振动与冲击, 2020, 39(10): 198-205 YAN R W, LIN C, GAO S X, et al. Fault diagnosis and analysis of circuit breaker based on wavelet time-frequency representations and convolution neural network[J]. Journal of Vibration and Shock, 2020, 39(10): 198-205 (in Chinese)
[6]	LIANG Y C, LI W D, LU X, et al. Fog computing and convolutional neural network enabled prognosis for machining process optimization[J]. Journal of Manufacturing Systems, 2019, 52: 32-42 doi: 10.1016/j.jmsy.2019.05.003
[7]	GOUARIR A, MARTÍNEZ-ARELLANO G, TERRAZAS G, et al. In-process tool wear prediction system based on machine learning techniques and force analysis[J]. Procedia CIRP, 2018, 77: 501-504 doi: 10.1016/j.procir.2018.08.253
[8]	AGHAZADEH F, TAHAN A, THOMAS M. Tool condition monitoring using spectral subtraction and convolutional neural networks in milling process[J]. The International Journal of Advanced Manufacturing Technology, 2018, 98(9): 3217-3227
[9]	张存吉, 姚锡凡, 张剑铭, 等. 基于深度学习的刀具磨损监测方法[J]. 计算机集成制造系统, 2017, 23(10): 2146-2155 ZHANG C J, YAO X F, ZHANG J M, et al. Tool wear monitoring based on deep learning[J]. Computer Integrated Manufacturing Systems, 2017, 23(10): 2146-2155 (in Chinese)
[10]	王丽华, 杨家巍, 张永宏, 等. 基于堆叠降噪自编码的刀具磨损状态识别[J]. 中国机械工程, 2018, 29(17): 2038-2045 doi: 10.3969/j.issn.1004-132X.2018.17.004 WANG L H, YANG J W, ZHANG Y H, et al. Tool wear condition recognition based on SDAE[J]. China Mechanical Engineering, 2018, 29(17): 2038-2045 (in Chinese) doi: 10.3969/j.issn.1004-132X.2018.17.004
[11]	王震, 黄如意, 李霁蒲, 等. 基于多尺度卷积神经网络的刀具磨损预测方法[C]//第十三届全国振动理论及应用学术会议论文集. 西安: 中国振动工程学会, 2019: 100-103 WANG Z, HUANG R Y, LI J P, et al. A multi-scale convolutional neural network for machine tool wear prediction[C]//Proceedings of the 13th National Conference on Vibration Theory and Application. Xi’an: Chinese Society of Vibration Engineering, 2019: 100-103 （in Chinese）
[12]	王民, 刘利明, 宋铠钰, 等. 基于主轴驱动电流杂波的立铣刀复杂工况下磨损状态辨识[J]. 计算机集成制造系统, 2021, 27(12): 3429-3438 WANG M, LIU L M, SONG K Y, et al. Wear status identification of end milling cutter under complex cutting conditions based on clutter signal of spindle current[J]. Computer Integrated Manufacturing Systems, 2021, 27(12): 3429-3438 (in Chinese)
[13]	FUKUSHIMA K. Neocognitron: a self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position[J]. Biological Cybernetics, 1980, 36(4): 193-202 doi: 10.1007/BF00344251
[14]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778
[15]	WANG S F, CHEN Z H, WANG J G, et al. Continuous leak detection and location through the optimal mother wavelet transform to AE signal[J]. Journal of Pipeline Systems Engineering and Practice, 2020, 11(3): 04020024 doi: 10.1061/(ASCE)PS.1949-1204.0000467
[16]	朱少民, 夏虹, 彭彬森, 等. 基于PCA的主泵传感器状态监测模型[J]. 核动力工程, 2020, 41(3): 170-176 ZHU S M, XIA H, PENG B S, et al. Condition monitoring model for sensors of reactor coolant pump based on PCA[J]. Nuclear Power Engineering, 2020, 41(3): 170-176 (in Chinese)
[17]	朱会杰, 王新晴, 芮挺, 等. 基于频域信号的稀疏编码在机械故障诊断中的应用[J]. 振动与冲击, 2015, 34(21): 59-64 ZHU H J, WANG X Q, RUI T, et al. Application of sparse coding based on frequency domain signals in machinery fault diagnosis[J]. Journal of Vibration and Shock, 2015, 34(21): 59-64 (in Chinese)