留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

改进MoblieNet网络在轴承轻量化诊断中的应用

朱富 刘畅 王贵勇 杨永灿

朱富, 刘畅, 王贵勇, 杨永灿. 改进MoblieNet网络在轴承轻量化诊断中的应用[J]. 机械科学与技术, 2024, 43(1): 31-38. doi: 10.13433/j.cnki.1003-8728.20220208
引用本文: 朱富, 刘畅, 王贵勇, 杨永灿. 改进MoblieNet网络在轴承轻量化诊断中的应用[J]. 机械科学与技术, 2024, 43(1): 31-38. doi: 10.13433/j.cnki.1003-8728.20220208
ZHU Fu, LIU Chang, WANG Guiyong, YANG Yongcan. Application of Improved MobileNet Network in Bearing Lightweight Diagnosis[J]. Mechanical Science and Technology for Aerospace Engineering, 2024, 43(1): 31-38. doi: 10.13433/j.cnki.1003-8728.20220208
Citation: ZHU Fu, LIU Chang, WANG Guiyong, YANG Yongcan. Application of Improved MobileNet Network in Bearing Lightweight Diagnosis[J]. Mechanical Science and Technology for Aerospace Engineering, 2024, 43(1): 31-38. doi: 10.13433/j.cnki.1003-8728.20220208

改进MoblieNet网络在轴承轻量化诊断中的应用

doi: 10.13433/j.cnki.1003-8728.20220208
基金项目: 

云南省重大科技专项 202102AC080002

详细信息
    作者简介:

    朱富, 硕士研究生, 2229872301@qq.com

    通讯作者:

    刘畅, 高级工程师, 硕士生导师, 博士, 37615085@qq.com

  • 中图分类号: TN98;TN06;TH165.3

Application of Improved MobileNet Network in Bearing Lightweight Diagnosis

  • 摘要: 近年来,基于神经网络的故障诊断方法在诊断的准确性、效率等方面展现出巨大的优势,然而呈指数增长的模型参数量限制了神经网络在工程实际中的应用。针对这一问题,本文提出了一种基于一维卷积神经网络改进的MobileNet网络用于实现滚动轴承的故障诊断;改进的网络能够直接应用于一维振动信号,有效降低系统硬件资源的要求,实现网络的轻量化部署;使用西储大学轴承数据集和QPZZ-Ⅱ型故障模拟试验台数据集对所提方法进行验证,本文提出的模型准确率均达99.8%以上,参数量为标准卷积神经网络的1/2。本文所提方法为在轻资源嵌入式系统中实现智能诊断提供了一种新的方法和思路。
  • 图  1  深度可分离卷积结构图

    Figure  1.  Depth-separable convolution structure

    图  2  1D-CNN网络结构图

    Figure  2.  1D-CNN network structure diagram

    图  3  数据增强示意图

    Figure  3.  1D-CNN network structure diagram

    图  4  西储大学轴承实验台

    Figure  4.  Western Reserve University bearing test bench

    图  5  模型训练过程

    Figure  5.  Model training process

    图  6  混淆矩阵

    Figure  6.  Confusion matrix

    图  7  全连接层T-SNE可视化结果

    Figure  7.  Fully connected layer T-SNE visualization results

    图  8  模型抗噪性能对比

    Figure  8.  Comparison of model anti-noise performance

    图  9  QPZZ-Ⅱ型机械振动分析及故障模拟试验台

    Figure  9.  QPZZ-Ⅱ mechanical vibration analysis and fault simulation test bench

    图  10  信号时域波形

    Figure  10.  Signal time domain waveform

    图  11  模型训练情况

    Figure  11.  Model training

    图  12  T-SNE可视化结果

    Figure  12.  T-SNE visualization results

    表  1  1D-CNN网络参数设置

    Table  1.   1D-CNN network parameters

    No. Layers Kernel size/pool size Activation Input size Output size
    1 Conv1 64×1×16 ReLu 6.0 2 048×1 128×16
    2 Maxpooling1 2×1 128×16 64×16
    3 DSC1 6×1×32 ReLu 6.0 64×16 64×32
    4 Maxpooling2 2×1 64×32 32×32
    5 DSC2 6×1×64 ReLu 6.0 32×32 32×64
    6 Maxpooling3 2×1 32×64 16×64
    7 DSC3 6×1×64 ReLu 6.0 16×64 16×64
    8 Maxpooling4 2×1 16×64 8×64
    9 DSC4 6×1×64 ReLu 6.0 8×64 8×64
    10 Maxpooling5 2×1 8×64 4×64
    11 Flatten 4×64 256×1
    12 FC1 100×1 ReLu 256×1 100×1
    13 FC2 N×1 Softmax 100×1 N×1
    下载: 导出CSV

    表  2  实验1数据集

    Table  2.   Experiment 1 data set

    实验对象 负载 样本数 样本长度 故障类型 故障直径/mm 标签
    6205-2RS JEM SKF 1.5 kW 1 000 2 048 正常 0 0
    1 000 2 048 滚动体 0.007 1
    1 000 2 048 滚动体 0.014 2
    1 000 2 048 滚动体 0.021 3
    1 000 2 048 内圈 0.007 4
    1 000 2 048 内圈 0.014 5
    1 000 2 048 内圈 0.021 6
    1 000 2 048 外圈 0.007 7
    1 000 2 048 外圈 0.014 8
    1 000 2 048 外圈 0.021 9
    下载: 导出CSV

    表  3  不同模型性能对比

    Table  3.   Performance comparison of different models

    模型 准确率/% 模型参数量 预测时间/s
    1D-CNN 99.6 30 998 2.495 9
    标准CNN 99.9 61 190 3.961 8
    SVM 88
    下载: 导出CSV

    表  4  实验2数据表

    Table  4.   Experiment 2 data set

    实验对象 负载 转速/(r·min-1) 样本数量 样本长度 状态 标签
    圆柱滚子轴承N205EN 30 kg 1 200 1 000 2 048 正常 0
    1 200 1 000 2 048 内圈故障 1
    1 200 1 000 2 048 外圈故障 2
    1 200 1 000 2 048 滚动体故障 3
    下载: 导出CSV

    表  5  模型性能对比

    Table  5.   Model performance comparison

    模型 准确率/% 模型参数量 预测时间/s
    1D-CNN 100 30 192 2.403 5
    标准CNN 100 59 188 3.621 1
    SVM 84.375
    下载: 导出CSV
  • [1] 李道军, 李廷锋, 刘德平. 基于LMD与改进SVM的轴承故障诊断方法[J]. 机械制造, 2021, 59(6): 84-88. doi: 10.3969/j.issn.1000-4998.2021.06.021

    LI D J, LI T F, LIU D P. Bearing fault diagnosis method based on LMD and improved SVM[J]. Machinery, 2021, 59(6): 84-88. (in Chinese) doi: 10.3969/j.issn.1000-4998.2021.06.021
    [2] CHEN Q Q, DAI S W, DAI H D. A rolling bearing fault diagnosis method based on EMD and quantile permutation entropy[J]. Mathematical Problems in Engineering, 2019, 2019: 3089417.
    [3] YUAN X J, WU W B, YE F L, et al. Application of a wavelet transform after signal differentiation in fault diagnosis[J]. Journal of Coastal Research, 2020, 105(S1): 61-66.
    [4] KHORRAM A, KHALOOEI M, REZGHI M. End-to-end CNN+LSTM deep learning approach for bearing fault diagnosis[J]. Applied Intelligence, 2021, 51(2): 736-751. doi: 10.1007/s10489-020-01859-1
    [5] WANG H, XU J W, YAN R Q, et al. Intelligent bearing fault diagnosis using multi-head attention-based CNN[J]. Procedia Manufacturing, 2020, 49: 112-118. doi: 10.1016/j.promfg.2020.07.005
    [6] ZHANG W, PENG G L, LI C H, et al. A new deep learning model for fault diagnosis with good anti-noise and domain adaptation ability on raw vibration signals[J]. Sensors, 2017, 17(2): 425. doi: 10.3390/s17020425
    [7] YE Z, YU J B. Deep morphological convolutional network for feature learning of vibration signals and its applications to gearbox fault diagnosis[J]. Mechanical Systems and Signal Processing, 2021, 161: 107984. doi: 10.1016/j.ymssp.2021.107984
    [8] Google LLC. Efficient convolutional neural networks and techniques to reduce associated computational costs: US, 20190347537A1[P]. 2019-11-14.
    [9] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90. doi: 10.1145/3065386
    [10] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[C] //3rd International Conference on Learning Representations. San Diego: ICLR, 2015.
    [11] YU W B, LV P. An end-to-end intelligent fault diagnosis application for rolling bearing based on MobileNet[J]. IEEE Access, 2021, 9: 41925-41933. doi: 10.1109/ACCESS.2021.3065195
    [12] WANG X, MAO D X, LI X D. Bearing fault diagnosis based on vibro-acoustic data fusion and 1D-CNN network[J]. Measurement, 2021, 173: 108518. doi: 10.1016/j.measurement.2020.108518
    [13] 郑一珍, 牛蔺楷, 熊晓燕, 等. 基于一维卷积神经网络的圆柱滚子轴承保持架故障诊断[J]. 振动与冲击, 2021, 40(19): 230-238.

    ZHENG Y Z, NIU L K, XIONG X Y, et al. Fault diagnosis of cylindrical roller bearing cage based on 1D convolution neural network[J]. Journal of Vibration and Shock, 2021, 40(19): 230-238. (in Chinese)
    [14] MAULUD D H, AMEEN S Y, OMAR N, et al. Review on natural language processing based on different techniques[J]. Asian Journal of Research in Computer Science, 2021, 10(1): 1-17.
    [15] 刘恒畅, 姚德臣, 杨建伟, 等. 基于多分支深度可分离卷积神经网络的滚动轴承故障诊断研究[J]. 振动与冲击, 2021, 40(10): 95-102.

    LIU H C, YAO D C, YANG J W, et al. Fault diagnosis of rolling bearings based on a multi branch depth separable convolutional neural network[J]. Journal of Vibration and Shock, 2021, 40(10): 95-102. (in Chinese)
  • 加载中
图(12) / 表(5)
计量
  • 文章访问数:  118
  • HTML全文浏览量:  61
  • PDF下载量:  20
  • 被引次数: 0
出版历程
  • 收稿日期:  2021-11-30
  • 刊出日期:  2024-01-25

目录

    /

    返回文章
    返回