基于两阶段域泛化学习框架的轴承故障诊断方法

谢刚; 韩秦; 聂晓音; 石慧; 张晓红; 田娟

doi:10.16383/j.aas.c230716

基于两阶段域泛化学习框架的轴承故障诊断方法

doi: 10.16383/j.aas.c230716 cstr: 32138.14.j.aas.c230716

谢刚^{1, 2,},
韩秦^{1, 2,},
聂晓音^{1, 2,},
石慧^{1, 2,},
张晓红^{1, 2,},
田娟^{1, 2,}

1.
太原科技大学电子信息工程学院太原 030024
2.
先进控制与工业智能山西省重点实验室太原 030024

基金项目: 山西省科技重大专项计划“揭榜挂帅”项目(202201090301013), 山西省重点研发计划项目(202102020101002, 202202100401002, 202202150401005), 山西省基础研究计划(自由探索类)面上项目(20210302123206, 202203021211194, 202203021221142)资助

详细信息

作者简介:
谢刚：太原科技大学教授, 博士生导师. 主要研究方向为机器视觉与图像处理、大数据驱动的智能故障诊断. E-mail: xiegang@tyust.edu.cn

韩秦：太原科技大学电子信息工程学院硕士研究生. 2021年获得中原工学院测控技术与仪器专业学士学位. 主要研究方向为大数据驱动的智能故障诊断. E-mail: hanqin@stu.tyust.edu.cn

聂晓音：太原科技大学电子信息工程学院讲师. 主要研究方向为大数据驱动的智能故障诊断. 本文通信作者. E-mail: 2022036@tyust.edu.cn

石慧：太原科技大学电子信息工程学院教授, 博士生导师. 主要研究方向为复杂系统故障预测与健康管理. E-mail: huishi@stu.tyust.edu.cn

张晓红：太原科技大学经济与管理学院教授, 博士生导师. 主要研究方向为复杂系统故障预测与健康管理、大数据装备与制造. E-mail: zhangxh@stu.tyust.edu.cn

田娟：太原科技大学电子信息工程学院讲师. 主要研究方向为大数据驱动的智能故障诊断. E-mail: juantian@stu.tyust.edu.cn

计量
- 文章访问数: 92
- HTML全文浏览量: 27
- 被引次数: 0
出版历程
- 收稿日期: 2023-11-16
- 录用日期: 2024-08-07
- 网络出版日期: 2024-09-27

A Two-stage Domain Generalization Learning Framework for Fault Diagnosis of Bearings

XIE Gang^{1, 2
,},
HAN Qin^{1, 2
,},
NIE Xiao-Yin^{1, 2
,},
SHI Hui^{1, 2
,},
ZHANG Xiao-Hong^{1, 2
,},
TIAN Juan^{1, 2
,}

1.
Taiyuan University of Science and Technology, Taiyuan 030024
2.
Advanced Control and Industrial Intelligence Shanxi Provincial Key Laboratory, Taiyuan 030024

Funds: Supported by Major Science and Technology Project of Shanxi Province (202201090301013), Key Research and Development Plan of Shanxi Province (202102020101002, 202202100401002, 202202150401005), and Basic Research Program of Shanxi Province (20210302123206, 202203021211194, 202203021221142)

More Information

Author Bio:
XIE Gang　Professor and Ph.D. supervisor at Taiyuan University of Science and Technology. His research interest covers machine vision and image processing, big data-driven intelligent fault diagnosis

HAN Qin　Master student at the School of Electronic and Information Engineering, Taiyuan University of Science and Technology. She received her bachelor degree in Measurement and Control Technology and Instruments from Zhongyuan University of Technology in 2021. Her research interest covers big data-driven intelligent fault diagnosis

NIE Xiao-Yin　Lecturer at the School of Electronic Information Engineering, Taiyuan University of Science and Technology. Her research interest covers big data-driven intelligent fault diagnosis. Corresponding author of this paper

SHI Hui　Professor and Ph.D. supervisor at the School of Electronic Information Engineering, Taiyuan University of Science and Technology. Her research interest covers fault prediction and health management of complex systems

ZHANG Xiao-Hong　Professor and Ph.D. supervisor at the School of Economics and Management, Taiyuan University of Science and Technology. Her research interest covers fault prediction and health management of complex systems, big data equipment and manufacturing

TIAN Juan　Lecturer at the School of Electronic Information Engineering, Taiyuan University of Science and Technology. Her research interest covers big data-driven intelligent fault diagnosis

摘要

摘要: 设备在实际运行过程中工况复杂多变, 导致振动信号分布存在较大差异. 现有的多数方法通过添加度量指标来约束特征提取过程, 提取源域和目标域的相似特征以解决从单一源域到目标域的诊断问题. 然而, 实际运行过程往往包含多个源域数据, 且目标域信息在不同源域中存在较大差异, 难以有效学习不同域之间的域不变特征. 针对上述问题, 提出了一种基于两阶段域泛化学习框架的轴承故障诊断方法. 在第一阶段, 利用大尺寸卷积特征提取模型对多视图振动信号进行预训练, 提取多个源域数据之间的初级故障特征. 在第二阶段, 将初级故障特征输入动静双态融合的时空图卷积模型中, 捕捉随时间变化的动态特征和全局时空特征. 通过两阶段的学习, 将多个源域的数据映射到一个共有特征空间, 提取判别性和泛化性特征. 实验结果表明, 该方法在多源域轴承故障诊断任务中具有较高的诊断精度和较强的泛化能力.
- 故障诊断 /
- 域泛化 /
- 多源域 /
- 多视图 /
- 时空关系
Abstract: During the actual operation of the equipment, the working conditions are complex and changeable, resulting in large differences in vibration signal distribution. Many existing methods constrain the feature extraction process by incorporating measurement metrics, aiming to extract similar features from both the source and target domains to address diagnostic problems from a single source domain to a target domain. However, the actual operational process often involves data from multiple source domains, and the target domain information exhibits significant differences across these various source domains, making it difficult to extract the domain invariant feature. In response to the above problems, this paper proposes a two-stage domain generalization learning framework for fault diagnosis of bearings. In the first stage, the large-scale convolutional feature extraction model is used to pre-train multi-view vibration signals to extract primary common features between multiple source domain data. In the second stage, the primary common features are input into the spatial-temporal graph convolutional model for dynamic and static two-state fusion combining dynamic and static states to capture the dynamic features and global spatiotemporal features that change over time. Through two-stage learning, data from multiple source domains are mapped to a common feature space, and discriminative and generalization features are extracted. Experimental results show that this method has high diagnostic accuracy and strong generalization ability in multi-source domain bearing fault diagnosis tasks.
- Fault diagnosis /
- domain generalization /
- multi-source /
- multi-view /
- spatial-temporal relationship

HTML全文

图 1 多源域到单目标域的域泛化

Fig. 1 Domain generalization from multi-source domains to single target domain

下载: 全尺寸图片幻灯片

图 2 两阶段域泛化学习框架

Fig. 2 The two-stage domain generalization learning framework

下载: 全尺寸图片幻灯片

图 3 大尺寸卷积特征提取模型

Fig. 3 Large-scale convolutional feature extraction model

下载: 全尺寸图片幻灯片

图 4 动静双态融合的时空图卷积模型

Fig. 4 Spatial-temporal graph convolutional model for dynamic and static two-state fusion

下载: 全尺寸图片幻灯片

图 5 动态节点连接

Fig. 5 The process of dynamic node connection

下载: 全尺寸图片幻灯片

图 6 静态节点连接

Fig. 6 The process of static node connection

下载: 全尺寸图片幻灯片

图 7 轴承故障模拟实验台

Fig. 7 Bearing fault simulation test bench

下载: 全尺寸图片幻灯片

图 8 故障细节

Fig. 8 Fault details

下载: 全尺寸图片幻灯片

图 9 动静双态融合的时空图卷积模型的特征可视化

Fig. 9 Feature visualization of spatial-temporal graph convolutional model for dynamic and static two-state fusion

下载: 全尺寸图片幻灯片

图 10 不同诊断方法的特征可视化

Fig. 10 Feature visualizations of different diagnosis methods

下载: 全尺寸图片幻灯片

表 1 不同域泛化任务的轴承数据说明

Table 1 Bearing data description of different domain generalization tasks

域泛化任务		源域			目标域
域泛化任务		转速(rpm)	总样本数	特征序列数	转速(rpm)	总样本数	特征序列数
任务1	B-C-D→A	600、900、1200	30000	29988	300	10000	9996
任务2	A-C-D→B	300、900、1200	30000	29988	600	10000	9996
任务3	A-B-D→C	300、600、1200	30000	29988	900	10000	9996
任务4	A-B-C→D	300、600、900	30000	29988	1200	10000	9996

下载: 导出CSV

表 2 结构参数设计

Table 2 Structural parameter design

网络结构			输入尺寸	输出尺寸	参数设置	填充方式
第一阶段	输入层		64×3×1024	64×3×1024×1	/	/
	卷积层(C1)		64×3×1024×1	64×3×13×64	kernel_size=400, filter=64, stride=50	same
	最大池化层(P2)		64×3×13×64	64×3×2×64	pool_size=5, strides=5	valid
	Dropout层(D3)		64×3×2×64	64×3×2×64	p=0.5	/
	展平层(F4)		64×3×2×64	64×3×128	/	/
	全连接层(F5)		64×3×128	64×384	/	/
	故障分类器		64×384	64×5	/	/
第二阶段	输入层		64×3×128	64×3×128	/	/
	添加时间步		64×3×128	64×5×3×128	/	/
	动态图卷积层		64×5×3×128	64×5×3×64	k=3, filter=64, $ \lambda =0.001 $	valid
	时序卷积层		64×5×3×64	64×5×3×64	filter=64, stride=(1,1), kernel=(3,1)	same
	静态图卷积层		64×5×3×128	64×5×3×64	k=3, filter=64	valid
	时序卷积层		64×5×3×64	64×5×3×64	filter=64, stride=(1,1), kernel=(3,1)	same
	全连接层		64×5×3×64, 64×5×3×64	64×5×3×128	/	/
	展平层		64×5×3×128	64×1920	/	/
	故障分类	全连接层	64×1920	64×64	/	/
	故障分类	故障分类器	64×64	64×5	/	/
	域分类	梯度反转层	64×1920	64×1920	$ \beta =0.01 $	/
		全连接层	64×1920	64×64	/	/
		域分类器	64×64	64×4	/	/

下载: 导出CSV

表 3 五种健康状态分类结果统计表

Table 3 Table of classification results of five health states

健康状态	准确率(%)	F1得分(%)	精确率(%)	召回率(%)
滚子故障	97.05±2.32	97.39±2.02	95.95±2.87	99.96±0.01
内圈故障	99.06±0.10	99.06±0.10	99.80±0.01	98.33±0.57
内外圈故障	92.27±1.05	97.61±1.31	99.98±0.01	96.34±0.07
外圈故障	99.15±0.04	99.15±0.01	98.44±0.18	99.20±0.50
正常状态	99.98±0.01	99.98±0.01	99.97±0.01	99.99±0.01
平均值	98.77±0.36	98.77±0.38	98.83±0.53	98.77±0.32

下载: 导出CSV

表 4 不同视图数据组合对性能的影响统计表

Table 4 Table of the impact of different multi-view data combinations on performance

不同视图组合	准确率(%)	F1得分(%)	精确率(%)	召回率(%)
$ xz $	90.75±0.50	90.24±0.62	92.06±0.76	90.42±0.61
$ xy $	92.31±0.71	92.67±0.62	93.18±0.45	92.31±0.71
$ yz $	94.78±2.23	94.57±2.55	95.70±1.33	94.76±2.54
$ xyz $	98.77±0.36	98.77±0.38	98.83±0.53	98.77±0.32

下载: 导出CSV

表 5 动静双态融合的时空图卷积模型不同组成结构对性能的影响统计表

Table 5 Table of the impact of different structures of the spatial-temporal graph convolutional model for dynamic and static two-state fusion

动态时空图卷积模块		静态时空图卷积模块		准确率(%)	F1 得分(%)	精确率(%)	召回率(%)
动态图卷积层	时序卷积层	静态图卷积层	时序卷积层	准确率(%)	F1 得分(%)	精确率(%)	召回率(%)
√	×	√	×	90.36±0.63	90.32±0.52	90.87±2.40	90.36±0.37
×	√	×	√	92.58±4.35	92.66±4.13	93.57±2.01	92.25±4.35
×	×	√	√	90.69±3.12	90.75±3.24	92.35±1.05	90.70±4.49
√	√	×	×	90.35±0.51	90.34±0.51	91.91±0.56	90.35±0.59
√	√	√	√	98.77±0.36	98.77±0.38	98.83±0.53	98.77±0.32

下载: 导出CSV

表 6 不同特征可视化效果的量化指标统计表

Table 6 Table of quantitative indicators of different feature visualization effects

量化指标	$ {\boldsymbol{X}}_{text}^{{{s}_{q}}} $	$ {\boldsymbol{X}}_{DGCN}^{{{s}_{q}}} $	$ {\boldsymbol{X}}_{DTCN}^{{{s}_{q}}} $	$ {\boldsymbol{X}}_{GCN}^{{{s}_{q}}} $	$ {\boldsymbol{X}}_{TCN}^{{{s}_{q}}} $	$ {{{\boldsymbol{F}}}^{{{s}_{q}}}} $
轮廓系数	0.0828	0.1134	0.4807	0.1920	0.5021	0.5431
CH指数	1194.49	2336.10	13642.92	5208.58	14917.97	17765.84

下载: 导出CSV

表 7 不同方法的诊断结果统计表

Table 7 Table of statistics of diagnostic results of different methods

评估指标	域泛化任务B-C-D→A
评估指标	CNN	WDCNN	GCN	TCN	DAGCN	CVC-Net	所提方法
准确率(%)	60.24±9.14	63.20±6.51	59.77±8.56	62.64±5.38	85.01±6.75	86.31±5.25	95.14±1.75
F1得分(%)	64.01±6.86	65.70±5.30	54.85±5.17	57.37±7.54	83.30±8.21	85.11±4.50	95.81±0.86
精确率(%)	63.61±8.43	61.86±6.46	57.38±6.52	61.15±8.64	90.98±3.83	80.04±3.59	95.95±1.98
召回率(%)	61.13±7.25	63.19±5.43	59.07±7.89	62.32±7.95	85.37±9.47	86.67±5.58	95.03±0.85

下载: 导出CSV

表 8 不同方法特征可视化效果的量化指标统计表

Table 8 Table of quantitative indicators of feature visualization effects of different methods

量化指标	CNN	WDCNN	GCN	TCN	DAGCN	CVC-Net	所提方法
轮廓系数	0.2164	0.1926	0.1843	0.2390	0.4440	0.4602	0.5082
CH指数	3920.27	3484.86	3483.51	4795.91	12051.26	11544.79	14780.77

下载: 导出CSV

参考文献(32)

[1]	刘建昌, 权贺, 于霞, 何侃, 李镇华. 基于参数优化VMD和样本熵的滚动轴承故障诊断. 自动化学报, 2022, 48(03): 808−819 Liu Jian-Chang, Quan He, Yu Xia, He Kan, Li Zhen-Hua. Rolling bearing fault diagnosis based on parameter optimization vmd and sample entropy. Acta Automatica Sinica, 2022, 48(03): 808−819
[2]	Xin G, Zhong Q T, Jin Y Q, Li Z, Chen Y F, Li Y Fu, et al. Autonomous bearing fault diagnosis based on fault-induced envelope spectrum and moving peaks-over-threshold approach. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 1−12
[3]	Ma Z P, Fu L, Dun G, Tan D P, Xu F, Zhang L B. A robust domain distribution alignment discriminative network driven by physical samples for rotor-bearing fault diagnosis. Knowledge-Based Systems, 2024, 300(27): 1−15
[4]	Pu H X, Zhang K, An Y Y. Restricted sparse networks for rolling bearing fault diagnosis. IEEE Transactions on Industrial Informatics, 2023, 19(11): 11139−11149 doi: 10.1109/TII.2023.3243929
[5]	范苍宁, 刘鹏, 肖婷, 赵巍, 唐降龙. 深度域适应综述: 一般情况与复杂情况. 自动化学报, 2021, 47(25): 515−548 Fan Cang-Ning, Liu Peng, Xiao Ting, Zhao Wei, Tang Jiang-Long. A review of deep domain adaptation: General situation and complex situation. Acta Automatica Sinica, 2021, 47(25): 515−548
[6]	Wang J D, Lan C L, Liu C, Ouyang Y D, Qin T, Lu W, et al. Generalizing to unseen domains: A survey on domain generalization. IEEE transactions on knowledge and data engineering, 2022, 35(8): 8052−8072
[7]	Huo C R, Xu W Y, Jiang Q S, Shen Y H, Zhu Q X, Zhang Q K. An unsupervised transfer learning approach for rolling bearing fault diagnosis based on dual pseudo-label screening. Structural Health Monitoring, 2024, 23(4): 2288−2309 doi: 10.1177/14759217231206579
[8]	An Y Y, Zhang K, Chai Y, Liu Q, Huang X H. Bearing fault diagnosis under variable working conditions base on contrastive domain adaptation method. IEEE Transactions on Instrumentation and Measurement, 2022, 71: 1−11
[9]	An Y Y, Zhang K, Chai Y, Zhu Z Q, Liu Q. Gaussian mixture variational based transformer domain adaptation fault diagnosis method and its application in bearing fault diagnosis. IEEE Transactions on Industrial Informatics, 2022, 20(1): 615−625
[10]	Ding Y F, Jia M P, Zhuang J C, Cao Y D, Zhao X L, Lee C G. Deep imbalanced domain adaptation for transfer learning fault diagnosis of bearings under multiple working conditions. Reliability Engineering & System Safety, 2023, 230: 108890
[11]	叶楠, 常佩泽, 张露予, 王嘉. 基于改进后半监督深度信念网络的多工况轴承故障诊断研究. 机械工程学报, 2021, 57(15): 80−90 doi: 10.3901/JME.2021.15.080 Ye Nan, Chang Pei-Ze, Zhang Lu-Yu, Wang Jia. Research on multi-condition bearing fault diagnosis based on improved semi-supervised deep belief network. Journal of Mechanical Engineering, 2021, 57(15): 80−90 doi: 10.3901/JME.2021.15.080
[12]	黄星华, 吴天舒, 杨龙玉, 胡友强, 柴毅. 一种面向旋转机械的基于Transformer特征提取的域自适应故障诊断. 仪器仪表学报, 2022, 43(11): 210−218 Huang Xing-Hua, Wu Tian-Shu, Yang Long-Yu, Hu You-Qiang, Chai Yi. Domain adaptive fault diagnosis based on transformer feature extraction for rotating machinery. Chinese journal of Scientific Instrument, 2022, 43(11): 210−218
[13]	陈仁祥, 唐林林, 胡小林, 胡友强, 柴毅. 不同转速下基于深度注意力迁移学习的滚动轴承故障诊断方法. 振动与冲击, 2022, 41(12): 95−101+195 Chen Ren-Xiang, Tang Lin-Lin, Hu Xiao-Lin, Hu You-Qiang, Chai Yi. A rolling bearing fault diagnosis method based on deep attention transfer learning at different rotations. Journal of Vibration and Shock, 2022, 41(12): 95−101+195
[14]	Chen Z Y, Liao Y X, Li J P, Huang R Y, Xu L, Jin G, et al. A multi-source weighted deep transfer network for open-set fault diagnosis of rotary machinery. IEEE Transactions on Cybernetics, 2022, 53(3): 1982−1993
[15]	吕鹏飞, 闫云聚, 荔越. 基于马氏距离的改进核Fisher化工故障诊断研究. 自动化学报, 2020, 46(11): 2379−2391 Lv Peng-Fei, Yan Yun-Ju, Li Yue. Research on fault diagnosis of improved kernel fisher based on mahalanobis distance in the field of chemical industry. Acta Automatica Sinica, 2020, 46(11): 2379−2391
[16]	Li X Y, Zhang Z, Gao L, Wen L. A new semi-supervised fault diagnosis method via deep coral and transfer component analysis. IEEE Transactions on Emerging Topics in Computational Intelligence, 2021, 6(3): 690−699
[17]	Li X Q, Jiang H K, Xie M, Wang T Q, Wang R X, Wu Z H. A reinforcement ensemble deep transfer learning network for rolling bearing fault diagnosis with multi-source domains. Advanced Engineering Informatics, 2022, 51: 101480 doi: 10.1016/j.aei.2021.101480
[18]	Yang B, Xu S C, Lei Y G, Lee C G, Stewart E, Roberts C. Multi-source transfer learning network to complement knowledge for intelligent diagnosis of machines with unseen faults. Mechanical Systems and Signal Processing, 2022, 162: 108095 doi: 10.1016/j.ymssp.2021.108095
[19]	Wang J D, Lan C L, Liu C, Ouyang Y D, Qin T, Lu W, et al. Generalizing to unseen domains: A survey on domain generalization. IEEE Transactions on Knowledge and Data Engineering, 2022, 35(8): 8052−8072
[20]	Bhatti U A, Tang H, Wu G L, Marjan S, Hussain A. Deep learning with graph convolutional networks: An overview and latest applications in computational intelligence. International Journal of Intelligent Systems, 2023, 2023(1): 8342104
[21]	Liu X L, Hou F, Qin H, Hao A M. Multi-view multi-scale cnns for lung nodule type classification from ct images. Pattern Recognition, 2018, 77: 262−275 doi: 10.1016/j.patcog.2017.12.022
[22]	Jia Z Y, Lin Y F, Wang J, Ning X J, He Y L, Zhou R H, et al. Multi-view spatial-temporal graph convolutional networks with domain generalization for sleep stage classification. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2021, 29: 1977−1986 doi: 10.1109/TNSRE.2021.3110665
[23]	Huang R Y, Li J P, Liao Y X, Chen J B, Wang Z, Li W H. Deep adversarial capsule network for compound fault diagnosis of machinery toward multidomain generalization task. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1−11
[24]	Zhao D F, Liu S L, Miao Z H, Zhang H L, Dou W. Subdomain adaptation joint attention network enabled two-stage strategy towards few-shot fault diagnosis of LRE turbopump. Advanced Engineering Informatics, 2024, 60: 102366 doi: 10.1016/j.aei.2024.102366
[25]	Luo F, Poslad S, Budanese E. Temporal convolutional networks for multiperson activity recognition using a 2-d lidar. IEEE Internet of Things Journal, 2020, 7(8): 7432−7442 doi: 10.1109/JIOT.2020.2984544
[26]	Guo Q G, Li J, Zhou F D, Li G L, Lin J L. An open-set fault diagnosis framework for mmcs based on optimized temporal convolutional network. Applied Soft Computing, 2023, 133: 109959 doi: 10.1016/j.asoc.2022.109959
[27]	Yang Y J, Zhang T X, Li G Y, Kim T, Wang G H. An unsupervised domain adaptation model based on dual-module adversarial training. Neurocomputing, 2022, 475: 102−111 doi: 10.1016/j.neucom.2021.12.060
[28]	Bagirov A M, Aliguliyev R M, Sultanova N. Finding compact and well-separated clusters: Clustering using silhouette coefficients. Pattern Recognition, 2023, 135: 109144 doi: 10.1016/j.patcog.2022.109144
[29]	Gonzalez K, Misra S. Unsupervised learning monitors the carbon-dioxide plume in the subsurface carbon storage reservoir. Expert Systems with Applications, 2022, 201: 117216 doi: 10.1016/j.eswa.2022.117216
[30]	Gao D W, Huang K, Zhu Y S, Zhu L P, Yan K, Ren Z J, et al. Semi-supervised small sample fault diagnosis under a wide range of speed variation conditions based on uncertainty analysis. Reliability Engineering & System Safety, 2024, 242: 109746
[31]	Li T, Zhao Z, Sun C, Yan R, Chen X. Domain adversarial graph convolutional network for fault diagnosis under variable working conditions. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1−10
[32]	Li X, Wang Y, Yao J, Li M, Gao Z. Multi-sensor fusion fault diagnosis method of wind turbine bearing based on adaptive convergent viewable neural networks. Reliability Engineering & System Safety, 2024, 245: 109980