作者：李少波¹, 姚勇¹, 桂桂², 李想¹, 胡建军³

作者单位：1. 贵州大学机械工程学院, 贵州 贵阳 550025;
2. 中国测试技术研究院, 四川 成都 610021;
3. 南卡罗莱纳大学, 哥伦比亚 29208

关键词：齿轮故障诊断;声学信号;信息融合;卷积神经网络

摘要：

针对齿轮故障诊断任务中，振动信号受设备或工矿环境的影响难以获取，传统的单通道声学诊断法只能采集部分信息用于局部诊断，多通道声学诊断法权值确定过程复杂、实时性差等问题，结合深度学习理论，提出一种基于卷积神经网络与多通道声学信号齿轮故障诊断法。通过将传感器布置在不同测量点位以获取不同敏感度的故障信息，再以卷积神经网络作为融合技术，对4路齿轮声学信号进行特征级融合，实现对多级传动齿轮的故障诊断。实验结果表明：相比于单个传感器多个特征量信息的传统声学诊断方法，该文所提出的方法在齿轮故障识别率上有显著提升，可达99.8%。

Gear fault diagnosis based on CNN and multi-channel acoustic signals
LI Shaobo¹, YAO Yong¹, GUI Gui², LI Xiang¹, HU Jianjun³
1. School of Mechanical Engineering, Guizhou University, Guiyang 550025, China;
2. National Institue of Measurement and Testing Technology, Chengdu 610021,China;
3. University of South Carolina, Columbia 29208, USA
Abstract: For the fault diagnosis of gears, the vibration and acoustical signal analysis are the two commonly methods. However, in some conditions, the vibration is difficult to obtain because of its contact measuring. For the traditional acoustic method, acoustic signal analysis based on single channel measurement can be used only for local diagnosis and the procedure of determining the weight value of multi-channel acoustic diagnosis is complex. To solve these problems, a multi-channel acoustic diagnosis based on convolutional neural network for gears fault diagnosis is proposed. The fault information of different sensitivity is obtained by placing microphone at different measuring points. Then, the convolutional neural network is used as the fusion technology to fuse the four channel acoustic signals of gears at feature level to realize the fault diagnosis of multi-stage transmission gears. The experiment result show that the method that we propose for gears fault diagnosis based on multi-microphone information has achieve 99.8% accuracy in fault type recognition of gears which is higher than traditional method that based on a single microphone with multiple manual features.
Keywords: gear fault diagnosis;acoustical signal;information fusion;convolutional neural network
2019, 45(10):1-5  收稿日期: 2018-08-22;收到修改稿日期: 2018-11-08
基金项目: 国家自然科学基金（51475097，91746116）
作者简介: 李少波(1973-),男,湖南岳阳市人,教授,博士,研究方向为智能制造、多源数据融合等
参考文献
[1] LI S, LIU G, TANG X, et al. An ensemble deep convolutional neural network model with improved D-S evidence fusion for bearing fault diagnosis[J]. Sensors, 2017, 17(8):1729
[2] LI Y, CHENG G, PANG Y, et al. Planetary gear fault diagnosis via feature image extraction based on multi central frequencies and vibration signal frequency spectrum[J]. Sensors, 2018, 18(6):1735
[3] FIGLUS T, STAŃCZYK M. A method for detecting damage to rolling bearings in toothed gears of processing lines[J]. Metalurgija, 2016, 55(1):75-78
[4] 鲁文波, 蒋伟康, 潘思伟, 等. 基于近场声全息声像图纹理特征的机械故障诊断方法[J]. 振动工程学报, 2013, 26(4):617-623
[5] 鲁文波, 蒋伟康, 张一麟, 等. 基于声场空间分布特征的齿轮箱故障诊断研究[J]. 振动与冲击, 2013, 32(18):183-188
[6] 蒋玲莉, 刘义伦, 李学军, 等. 基于SVM与多振动信息融合的齿轮故障诊断[J]. 中南大学学报(自然科学版), 2010, 41(6):2184-2188
[7] 雷亚国, 林京, 何正嘉. 基于多传感器信息融合的行星齿轮箱故障诊断[C]//全国设备故障诊断学术会议, 2010.
[8] 雷亚国, 贾峰, 周昕, 等. 基于深度学习理论的机械装备大数据健康监测方法[J]. 机械工程学报, 2015, 51(21):49-56
[9] 罗鹏, 杨宇, 程军圣. AFDCNN方法及其在齿轮故障定量诊断中的应用[J]. 电子测量与仪器学报, 2018(1):33-39
[10] LIU C, CHENG G, CHEN X H, et al. Planetary gears feature extraction and fault diagnosis method based on VMD and CNN[J]. Sensors, 2018, 18(5):1523
[11] PICZAK K J. Environmental sound classification with convolutional neural networks[C]//IEEE, International Workshop on Machine Learning for Signal Processing, 2015.
[12] TOKOZUME Y, HARADA T. Learning environmental sounds with end-to-end convolutional neural network[C]//IEEE International Conference on Acoustics, Speech and Signal Processing, 2017.