基于多层域适应的无标签数据故障诊断方法

王进花; 刘瑞; 曹洁

doi:10.13700/j.bh.1001-5965.2023.0166

基于多层域适应的无标签数据故障诊断方法

doi: 10.13700/j.bh.1001-5965.2023.0166

王进花^1, ,,
刘瑞¹,
曹洁^{1, 2}

1.
兰州理工大学电气工程与信息工程学院，兰州 730050
2.
甘肃省制造信息工程研究中心，兰州 730050

基金项目: 国家重点研发计划(2020YFB1713600)；国家自然科学基金(62063020)；甘肃省自然科学基金(20JR5RA463)

详细信息

通讯作者:
E-mail：wjh0615@lut.edu.cn

中图分类号: TP277；TD453
计量
- 文章访问数: 38
- HTML全文浏览量: 7
- PDF下载量: 3
- 被引次数: 0
出版历程
- 收稿日期: 2023-04-06
- 录用日期: 2023-06-09
- 网络出版日期: 2023-06-29
- 整期出版日期: 2025-04-30

Unlabeled data fault diagnosis method based on multi-domain adaptation

WANG Jinhua^{1
, ,},
LIU Rui¹,
CAO Jie^{1, 2}

1.
College of Electrical and Information Engineering，Lanzhou University of Technology，Lanzhou 730050，China
2.
Gansu Manufacturing Information Engineering Research Center，Lanzhou 730050，China

Funds: National Key Research and Development Program of China (2020YFB1713600); National Natural Science Foundation of China (62063020)； Natural Science Foundation of Gansu Province (20JR5RA463)

More Information

Corresponding author: E-mail: wjh0615@lut.edu.cn

摘要

摘要:
在工业生产中，由于源域数据和目标域数据分布有差异且有标签的故障数据量较少，以至于现有的域适应轴承故障诊断方法大多精度不高。基于此，提出多层域适应神经网络(MDANN)故障诊断方法，用于无标签数据的滚动轴承故障诊断。使用小波包分解与重构(WPT)对原始振动信号进行处理，以降低信号冗余并避免关键信号特征遗失；利用多核最大均值差异(MK-MMD)算法对输入特征值进行差异计算，并通过反向传播更新多层域适应神经网络的参数，使其能够提取域不变特征；为保证无标签目标域数据可以正常参与网络训练，使用最大概率标签作为真实标签的伪标签策略，解决目标域无标签数据无法训练问题，增强模型可靠诊断知识的获取。采用2个公开数据集CWRU和PU进行验证。实验结果表明：所提方法与常见的域适应方法对比具有更高的诊断精度，说明该方法能够有效地学习可迁移特征，拟合2个数据集之间的数据分布差异。
- 滚动轴承 /
- 故障诊断 /
- 迁移学习 /
- 多层域适应 /
- 伪标签策略
Abstract:
in industrial production, due to the difference in the distribution of source domain data and target domain data and the small amount of labeled fault data, the accuracy of domain adaptation-based bearing fault diagnosis algorithms proposed in the past is generally not high. In view of this, the multi-domain adaptation neural network (MDANN) fault diagnosis method was proposed in this paper, which was used for rolling bearing fault diagnosis without labeled data. Firstly, the original vibration signal was processed by using wavelet packet transformation (WPT) to reduce signal redundancy and avoid the loss of key signal features. Secondly, the multi-kernel maximum mean discrepancy (MK-MMD) algorithm was used to calculate the difference of input eigenvalues, and the network parameters of MDANN were updated by backpropagation so that the network can extract domain invariant features. Finally, in order to ensure that unlabeled target domain data can participate in network training normally, the maximum probability label was used as a pseudo-label strategy of the real label to solve the problem that unlabeled target domain data cannot be trained and enhance the acquisition of reliable diagnosis knowledge of the model. Two publicly available datasets, CWRU and PU, were used for validation. The experimental results show that the proposed method has higher diagnosis accuracy compared with common domain adaptation methods, which further shows that the method can effectively learn the transferable features and fit the discrepancy in data distribution between the two datasets.
- rolling bearing /
- fault diagnosis /
- transfer learning /
- multi-domain adaptation /
- pseudo-label strategy

HTML全文

图 1 域适应

Figure 1. Domain adaptation

下载: 全尺寸图片幻灯片

图 2 本文方法网络结构

Figure 2. Network structure of the proposed method

下载: 全尺寸图片幻灯片

图 3 网络训练流程

Figure 3. Network training flow

下载: 全尺寸图片幻灯片

图 4 CWRU滚动轴承实验平台

Figure 4. CWRU rolling bearing test platform

下载: 全尺寸图片幻灯片

图 5 PU滚动轴承实验平台

1.电机；2.扭矩测量单元；3.轴承测量单元；4.飞轮；5.负载电机。

Figure 5. PU rolling bearing test platform

下载: 全尺寸图片幻灯片

图 6 不同实验参数的诊断结果

Figure 6. Diagnosis results of different experimental parameters

下载: 全尺寸图片幻灯片

图 7 轴承分类的混淆矩阵

Figure 7. Confusion matrix for bearing classification

下载: 全尺寸图片幻灯片

图 8 不同处理策略对于诊断精度影响的t-SNE可视化

Figure 8. t-SNE visualization of effects of different processing strategies on diagnosis accuracy

下载: 全尺寸图片幻灯片

图 9 不同模型的t-SNE图

Figure 9. t-SNE diagrams of different models

下载: 全尺寸图片幻灯片

表 1 网络参数详解

Table 1. Detailed explanation of network parameters

网络层	网络参数	激活函数	输出尺寸
输入层			2048×1
卷积层C1	3×1×20	非线性映射Relu	2046×20
卷积层C2	3×20×20	非线性映射Relu	1020×20
卷积层C3	3×20×20	非线性映射Relu	508×20
卷积层C4	3×20×20	非线性映射Relu	252×20
池化层P1～P6	2×1
全连接层F1			5040×1
全连接层F2	5040×256		256×1
全连接层F3	256×5	分类函数SoftMax	5×1

下载: 导出CSV

表 2 数据集的基本情况

Table 2. Basic information of datasets

数据集	转速/（r·min⁻¹）	负载/N	训练集/测试集（故障样本数）
数据集A(CWRU)	1700	1470	1500/500
数据集B (PU)	1500	1000	1500/500

下载: 导出CSV

表 3 不同数据处理方法的诊断精度

Table 3. Diagnosis accuracy of different data processing methods

数据处理方法	输入特征	迁移任务	诊断精度/%
不做处理	2048×1	A->B	62.23
FFT	2048×1	A->B	71.48
CWT	2014×1	A->B	73.65
WPT	2048×1	A->B	82.14
不做处理	2048×1	B->A	54.21
FFT	2048×1	B->A	75.47
CWT	2014×1	B->A	71.56
WPT	2048×1	B->A	80.37

下载: 导出CSV

表 4 多层域适应对诊断精度的影响

Table 4. Effect of multi-domain adaptation on diagnosis accuracy

处理策略	训练时间/s	训练任务	诊断精度/%
不添加MK-MMD	93	A->B	37.21
F1处添加MK-MMD	108	A->B	71.32
F1,F2处添加MK-MMD	124	A->B	77.56
F1,F2,F3处添加MK-MMD	147	A->B	82.94
不添加MK-MMD	89	B->A	41.83
F1处添加MK-MMD	97	B->A	75.89
F1,F2处添加MK-MMD	123	B->A	79.87
F1,F2,F3处添加MK-MMD	148	B->A	83.21

下载: 导出CSV

表 5 不同模型的精度

Table 5. Accuracy of different models

网络模型	是否域适应	迁移任务	输入特征	诊断精度/%
TCA	否	A->B	2048×1	47.12
DDC	是	A->B	224×224×3	74.32
DAN	是	A->B	2048×1	76.86
DANN	是	A->B	224×224×3	77.58
MANN	否	A->B	2048×1	32.15
MDANN	是	A->B	2048×1	84.17
TCA	否	B->A	2048×1	28.21
DDC	是	B->A	224×224×3	65.36
DAN	是	B->A	2048×1	74.21
DANN	是	B->A	224×224×3	75.36
MANN	否	B->A	2048×1	36.58
MDANN	是	B->A	2048×1	82.15

下载: 导出CSV

参考文献(24)

[1]	YU X, LIANG Z T, WANG Y J, et al. A wavelet packet transform-based deep feature transfer learning method for bearing fault diagnosis under different working conditions[J]. Measurement, 2022, 201: 111597. doi: 10.1016/j.measurement.2022.111597
[2]	HU Q, SI X S, QIN A S, et al. Balanced adaptation regularization based transfer learning for unsupervised cross-domain fault diagnosis[J]. IEEE Sensors Journal, 2022, 22(12): 12139-12151. doi: 10.1109/JSEN.2022.3174396
[3]	赵小强, 张青青, 陈鹏, 等. 基于PSO-BFA和改进Alexnet的滚动轴承故障诊断方法[J]. 振动与冲击, 2020, 39(7): 21-28. ZHAO X Q, ZHANG Q Q, CHEN P, et al. Rolling bearing fault diagnosis method based on improved Alexnet and PSO-BFA[J]. Journal of Vibration and Shock, 2020, 39(7): 21-28(in Chinese).
[4]	XU W, WAN Y, ZUO T Y, et al. Transfer learning based data feature transfer for fault diagnosis[J]. IEEE Access, 2020, 8: 76120-76129. doi: 10.1109/ACCESS.2020.2989510
[5]	MA P, ZHANG H L, FAN W H, et al. A novel bearing fault diagnosis method based on 2D image representation and transfer learning-convolutional neural network[J]. Measurement Science and Technology, 2019, 30(5): 055402. doi: 10.1088/1361-6501/ab0793
[6]	WU Z H, JIANG H K, ZHAO K, et al. An adaptive deep transfer learning method for bearing fault diagnosis[J]. Measurement, 2020, 151: 107227. doi: 10.1016/j.measurement.2019.107227
[7]	刘飞, 陈仁文, 邢凯玲, 等. 基于迁移学习与深度残差网络的滚动轴承快速故障诊断算法[J]. 振动与冲击, 2022, 41(3): 154-164. LIU F, CHEN R W, XING K L, et al. Fast fault diagnosis algorithm for rolling bearing based on transfer learning and deep residual network[J]. Journal of Vibration and Shock, 2022, 41(3): 154-164(in Chinese).
[8]	ZOU Y S, LIU Y Z, DENG J L, et al. A novel transfer learning method for bearing fault diagnosis under different working conditions[J]. Measurement, 2021, 171: 108767. doi: 10.1016/j.measurement.2020.108767
[9]	ZHAO M H, ZHONG S S, FU X Y, et al. Deep residual shrinkage networks for fault diagnosis[J]. IEEE Transactions on Industrial Informatics, 2020, 16(7): 4681-4690.
[10]	SHAO J J, HUANG Z W, ZHU J M. Transfer learning method based on adversarial domain adaption for bearing fault diagnosis[J]. IEEE Access, 2020, 8: 119421-119430. doi: 10.1109/ACCESS.2020.3005243
[11]	LI B, TANG B P, DENG L, et al. Joint attention feature transfer network for gearbox fault diagnosis with imbalanced data[J]. Mechanical Systems and Signal Processing, 2022, 176: 109146. doi: 10.1016/j.ymssp.2022.109146
[12]	张根保, 李浩, 冉琰, 等. 一种用于轴承故障诊断的迁移学习模型[J]. 吉林大学学报(工学版), 2020, 50(5): 1617-1626. ZHANG G B, LI H, RAN Y, et al. A transfer learning model for bearing fault diagnosis[J]. Journal of Jilin University (Engineering and Technology Edition), 2020, 50(5): 1617-1626(in Chinese).
[13]	LIU M K, TRAN M Q, WENG P Y. Fusion of vibration and current signatures for the fault diagnosis of induction machines[J]. Shock and Vibration, 2019, 2019(1): 7176482.
[14]	JIE H J, WANDA P. RunPool: a dynamic pooling layer for convolution neural network[J]. International Journal of Computational Intelligence Systems, 2020, 13(1): 66-76. doi: 10.2991/ijcis.d.200120.002
[15]	YANG Z N, WANG X Y, YANG R. Transfer learning based rolling bearing fault diagnosis[C]//Proceedings of the IEEE 10th Data Driven Control and Learning Systems Conference. Piscataway: IEEE Press, 2021: 354-359.
[16]	徐冰冰, 岑科廷, 黄俊杰, 等. 图卷积神经网络综述[J]. 计算机学报, 2020, 43(5): 755-780. doi: 10.11897/SP.J.1016.2020.00755 XU B B, CEN K T, HUANG J J, et al. A survey on graph convolutional neural network[J]. Chinese Journal of Computers, 2020, 43(5): 755-780(in Chinese). doi: 10.11897/SP.J.1016.2020.00755
[17]	SHABBEER BASHA S H, DUBEY S R, PULABAIGARI V, et al. Impact of fully connected layers on performance of convolutional neural networks for image classification[J]. Neurocomputing, 2020, 378: 112-119. doi: 10.1016/j.neucom.2019.10.008
[18]	史加荣, 王丹, 尚凡华, 等. 随机梯度下降算法研究进展[J]. 自动化学报, 2021, 47(9): 2103-2119. SHI J R, WANG D, SHANG F H, et al. Research advances on stochastic gradient descent algorithms[J]. Acta Automatica Sinica, 2021, 47(9): 2103-2119(in Chinese).
[19]	CAO X Y, LAI L F. Distributed gradient descent algorithm robust to an arbitrary number of Byzantine attackers[J]. IEEE Transactions on Signal Processing, 2019, 67(22): 5850-5864. doi: 10.1109/TSP.2019.2946020
[20]	ACHARYA U R, YANTI R, ZHENG J W, et al. Automated diagnosis of epilepsy using CWT, HOS and texture parameters[J]. International Journal of Neural Systems, 2013, 23(3): 1350009. doi: 10.1142/S0129065713500093
[21]	SHEN J X, SONG D, MA T C, et al. Blade crack detection based on domain adaptation and autoencoder of multidimensional vibro-acoustic feature fusion[J]. Structural Health Monitoring, 2023, 22(5): 3498-3513. doi: 10.1177/14759217221139134
[22]	ZHANG W P, ZHANG P, HE X H, et al. Convolutional neural network based two-layer transfer learning for bearing fault diagnosis[J]. IEEE Access, 2022, 10: 109779-109794. doi: 10.1109/ACCESS.2022.3213657
[23]	LONG M S, ZHU H, WANG J M, et al. Deep transfer learning with joint adaptation networks[C]//Proceedings of the 34th International Conference on Machine Learning. New York: ACM, 2017: 2208-2217.
[24]	WANG Q, MICHAU G, FINK O. Domain adaptive transfer learning for fault diagnosis[C]//Proceedings of the Prognostics and System Health Management Conference. Piscataway: IEEE Press, 2019: 279-285.