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

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

朱富^{1, 2,},
刘畅^{1, 2, ,},
王贵勇³,
杨永灿^{1, 2}

1.
昆明理工大学机电工程学院, 昆明 650500
2.
云南省先进装备智能制造技术重点实验室, 昆明 650500
3.
内蒙古第一机械集团有限公司, 内蒙古包头 014000

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
计量
- 文章访问数: 81
- HTML全文浏览量: 33
- PDF下载量: 15
- 被引次数: 0
出版历程
- 收稿日期: 2021-11-30
- 刊出日期: 2024-01-25

Application of Improved MobileNet Network in Bearing Lightweight Diagnosis

ZHU Fu^{1, 2
,},
LIU Chang^{1, 2
, ,},
WANG Guiyong³,
YANG Yongcan^{1, 2}

1.
Faculty of Mechanical and Electrical Engineering, Kunming University of Science and Technology, Kunming 650500, China
2.
Key Laboratory of Advanced Equipment Intelligent Manufacturing Technology of Yunnan Provincial, Kunming 650500, China
3.
Inner Mongolia First Machinery Group Co., Ltd., Baotou 014000, Inner Mongolia, China

摘要

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。
- 滚动轴承 /
- 故障诊断 /
- 神经网络 /
- MobileNet
Abstract: In recent years, fault diagnosis methods based on neural networks have shown great advantages in the accuracy and efficiency of diagnosis. However, the exponentially increasing number of model parameters limits the application of neural networks in engineering practice. Aiming at this problem, this paper proposes an improved MobileNet network based on one-dimensional convolutional neural network for fault diagnosis of rolling bearings. The improved network can be directly applied to one-dimensional vibration signals, effectively reducing the requirements of system hardware resources and realizing lightweight deployment of the network; The proposed method is validated using the Western Reserve University bearing dataset and the QPZZ-Ⅱ fault simulation test bench dataset. The accuracy of the model proposed in this paper is more than 99.8%, and the number of parameters is 1/2 of the standard convolutional neural network. The method proposed provides a new way for realizing intelligent diagnosis in light-resource embedded systems.
- rolling bearing /
- fault diagnosis /
- neural network /
- MobileNet

HTML全文

图 1 深度可分离卷积结构图

Figure 1. Depth-separable convolution structure

下载: 全尺寸图片幻灯片

图 2 1D-CNN网络结构图

Figure 2. 1D-CNN network structure diagram

下载: 全尺寸图片幻灯片

图 3 数据增强示意图

Figure 3. 1D-CNN network structure diagram

下载: 全尺寸图片幻灯片

图 4 西储大学轴承实验台

Figure 4. Western Reserve University bearing test bench

下载: 全尺寸图片幻灯片

图 5 模型训练过程

Figure 5. Model training process

下载: 全尺寸图片幻灯片

图 6 混淆矩阵

Figure 6. Confusion matrix

下载: 全尺寸图片幻灯片

图 7 全连接层T-SNE可视化结果

Figure 7. Fully connected layer T-SNE visualization results

下载: 全尺寸图片幻灯片

图 8 模型抗噪性能对比

Figure 8. Comparison of model anti-noise performance

下载: 全尺寸图片幻灯片

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

下载: 全尺寸图片幻灯片

图 10 信号时域波形

Figure 10. Signal time domain waveform

下载: 全尺寸图片幻灯片

图 11 模型训练情况

Figure 11. Model training

下载: 全尺寸图片幻灯片

图 12 T-SNE可视化结果

Figure 12. T-SNE visualization results

下载: 全尺寸图片幻灯片

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

参考文献(15)

[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)