基于MTF-SPCNN的小样本滚动轴承变工况故障诊断方法

焦孟萱; 雷春丽; 马淑珍; 薛林林; 史佳硕; 李建华

doi:10.13700/j.bh.1001-5965.2022.0927

基于MTF-SPCNN的小样本滚动轴承变工况故障诊断方法

doi: 10.13700/j.bh.1001-5965.2022.0927

1.
兰州理工大学机电工程学院，兰州 730050
2.
云南文山铝业有限公司，文山 663000

基金项目: 国家自然科学基金(51465035)；甘肃省自然科学基金(20JR5RA466)

详细信息

通讯作者:
E-mail：lclyq2004@163.com

中图分类号: TH133.33
计量
- 文章访问数: 273
- HTML全文浏览量: 109
- PDF下载量: 11
- 被引次数: 0
出版历程
- 收稿日期: 2022-11-17
- 录用日期: 2023-02-10
- 网络出版日期: 2023-02-28
- 整期出版日期: 2024-12-31

Fault diagnosis method of small sample rolling bearings under variable working conditions based on MTF-SPCNN

1.
School of Mechanical and Electronical Engineering，Lanzhou University of Technology，Lanzhou 730050，China
2.
Yunnan Wenshan Aluminum Co.，Ltd.，Wenshan 663000，China

Funds: National Natural Science Foundation of China (51465035); Natural Science Foundation of Gansu Province of China (20JR5RA466)

More Information

Corresponding author: E-mail：lclyq2004@163.com

摘要

摘要:
针对滚动轴承运行工况复杂及样本不足导致故障诊断精度较低的问题，提出一种基于马尔可夫转移场(MTF)与条纹池化卷积神经网络(SPCNN)的小样本滚动轴承变工况故障诊断方法。采用MTF将一维轴承信号转变为具有时间关联性的二维图像；提出条纹池化模块(SPM)并将其引入到网络中，不仅可以加强模型在长距离方向信息的捕捉能力，还可以有效提取远程空间特征；在最大池化层前添加SE注意力机制，增加有用信息的权重，提高模型训练速度，构建MTF-SPCNN模型；将MTF图像输入到MTF-SPCNN网络中进行训练，得到故障分类结果。运用美国凯斯西储大学及实验室滚动轴承MFS数据集验证所提方法在小样本变负载和变转速时的诊断效果，并对MFS数据集进行加噪处理，与其他智能算法进行对比，实验结果表明，所提方法具有更高的故障分类准确率、更强的泛化性能和抗干扰性能。
- 滚动轴承 /
- 故障诊断 /
- 马尔可夫转移场 /
- 条纹池化 /
- 卷积神经网络 /
- 小样本
Abstract:
In order to solve the problem of low fault diagnosis accuracy caused by complex operating conditions and insufficient samples of rolling bearings, a fault diagnosis method based on the Markov transition field (MTF) and the stripe pooling convolutional neural network (SPCNN) for small sample rolling bearings under variable working conditions was proposed. Firstly, one-dimensional bearing signals were transformed into two-dimensional images with time correlation by using MTF. Then, the stripe pooling module (SPM) was presented and introduced into the network, which could not only enhance the ability of the model to capture information in the long-distance direction but also effectively extract remote spatial features. Secondly, the channel attention mechanism, namely SE was added before the max pooling layer to increase the weight of useful information and improve the training speed of the model. The MTF-SPCNN model was thus constructed. Finally, the MTF images were input into the MTF-SPCNN for training, and fault classification results were obtained. The data sets of Case Western Reserve University and laboratory rolling bearings MFS were used to verify the validity of the proposed method in small samples with variable load and variable speed, and the MFS data sets were processed with noise added and compared with other intelligent algorithms. Experimental results show that the proposed method has higher fault recognition accuracy and stronger generalization performance and anti-interference performance.
- rolling bearing /
- fault diagnosis /
- Markov transition field /
- stripe pooling /
- convolutional neural network /
- small sample

HTML全文

图 1 SE注意力机制结构

Figure 1. Structure of SE attention mechanism

下载: 全尺寸图片幻灯片

图 2 条纹池化模块结构

Figure 2. SPM structure

下载: 全尺寸图片幻灯片

图 3 MTF-SPCNN模型

Figure 3. MTF-SPCNN model

下载: 全尺寸图片幻灯片

图 4 基于MTF-SPCNN的小样本滚动轴承变工况故障诊断流程

Figure 4. Fault diagnosis process of small sample rolling bearing under variable working conditions based on MTF-SPCNN

下载: 全尺寸图片幻灯片

图 5 美国凯斯西储大学滚动轴承实验台

Figure 5. Rolling bearing test rig of the CWRU

下载: 全尺寸图片幻灯片

图 6 MFS机械故障模拟实验平台

Figure 6. Mechanical fault simulation test platform of MFS

下载: 全尺寸图片幻灯片

图 7 ER-16K深沟球轴承故障部位

Figure 7. Fault position of ER-16K deep-groove ball bearing

下载: 全尺寸图片幻灯片

图 8 不同数据标签下的MTF二维特征图

Figure 8. MTF two-dimensional characteristic maps under different data labels

下载: 全尺寸图片幻灯片

图 9 不同负载下箱型图

Figure 9. Boxplot under different loads

下载: 全尺寸图片幻灯片

图 10 不同样本数下混淆矩阵

Figure 10. Confusion matrix under different samples

下载: 全尺寸图片幻灯片

图 11 不同模型变转速时分类准确率

Figure 11. Classification accuracy of different models under variable speeds

下载: 全尺寸图片幻灯片

表 1 CWRU滚动轴承数据集

Table 1. CWRU rolling bearing data set

工况	损伤直径/mm	故障位置	标签	训练集样本数	测试集样本数	负载/kW
A	0	正常	0	40	100	0.7457 （1 hp）
	0.18	内圈	1	40	100
		外圈	2	40	100
		滚动体	3	40	100
	0.36	内圈	4	40	100
		外圈	5	40	100
		滚动体	6	40	100
B	0	正常	0	40	100	1.4914 （2 hp）
	0.18	内圈	1	40	100
		外圈	2	40	100
		滚动体	3	40	100
	0.36	内圈	4	40	100
		外圈	5	40	100
		滚动体	6	40	100
C	0	正常	0	40	100	2.2371 （3 hp）
	0.18	内圈	1	40	100
		外圈	2	40	100
		滚动体	3	40	100
	0.36	内圈	4	40	100
		外圈	5	40	100
		滚动体	6	40	100

下载: 导出CSV

表 2 MFS滚动轴承数据集

Table 2. MFS rolling bearing data set

工况	损伤直径/mm	故障位置	标签	训练集样本数	测试集样本数	转速/ （r·min⁻¹）
D	0	正常	0	40	100	1200
	0.6	内圈	1	40	100
		外圈	2	40	100
		滚动体	3	40	100
	1.2	内圈	4	40	100
		外圈	5	40	100
		滚动体	6	40	100
E	0	正常	0	40	100	1300
	0.6	内圈	1	40	100
		外圈	2	40	100
		滚动体	3	40	100
	1.2	内圈	4	40	100
		外圈	5	40	100
		滚动体	6	40	100
F	0	正常	0	40	100	1400
	0.6	内圈	1	40	100
		外圈	2	40	100
		滚动体	3	40	100
	1.2	内圈	4	40	100
		外圈	5	40	100
		滚动体	6	40	100

下载: 导出CSV

表 3 不同模型变负载时分类准确率

Table 3. Classification accuracy of different models under variable loads

样本数	模型	分类准确率/%						平均分类准确率/%
样本数	模型	A-B	A-C	B-A	B-C	C-A	C-B	平均分类准确率/%
10	MTF-SPCNN	99.47	97.47	99.59	97.96	98.07	99.04	98.60
	MTF-ICNN	98.76	96.90	98.04	96.86	97.04	98.19	97.63
	GADF-SPCNN	88.72	86.91	88.43	87.10	87.18	88.24	87.76
	DFCNN	98.66	95.86	98.53	96.14	97.70	98.04	97.49
	MSACNN	97.14	96.48	97.81	95.24	96.05	96.71	96.57
20	MTF-SPCNN	99.94	98.57	99.63	98.29	98.58	99.49	99.08
	MTF-ICNN	98.95	97.52	98.19	97.24	97.67	98.62	98.03
	GADF-SPCNN	89.24	87.67	89.99	87.95	88.17	89.71	88.79
	DFCNN	99.14	96.05	99.05	97.47	97.91	98.76	98.06
	MSACNN	97.48	96.81	97.95	97.56	97.48	97.43	97.45
30	MTF-SPCNN	99.97	98.92	99.68	98.57	98.70	99.67	99.25
	MTF-ICNN	99.01	98.09	98.39	98.14	98.33	99.05	98.50
	GADF-SPCNN	90.05	89.51	91.28	89.54	89.75	90.12	90.04
	DFCNN	99.24	97.34	99.29	97.62	98.33	99.37	98.53
	MSACNN	98.33	96.95	98.47	98.09	97.86	98.28	98.00
40	MTF-SPCNN	99.99	99.09	99.80	99.14	99.25	99.80	99.51
	MTF-ICNN	99.38	98.20	98.95	98.38	98.57	99.48	98.83
	GADF-SPCNN	90.33	90.19	91.72	90.09	90.28	90.24	90.48
	DFCNN	99.33	98.00	99.43	98.66	99.19	99.52	99.02
	MSACNN	99.29	98.14	99.24	98.28	98.24	99.14	98.72

下载: 导出CSV

表 4 不同模型抗噪性能分析

Table 4. Analysis of anti-noise performance of different models

信噪比/dB	模型	分类准确率/%						平均分类准确率/%
信噪比/dB	模型	D-E	D-F	E-D	E-F	F-D	F-E	平均分类准确率/%
0	MTF-SPCNN	88.14	88.01	88.15	87.43	87.86	88.87	88.08
	MTF-ICNN	62.01	62.62	60.43	65.66	62.09	61.24	62.34
	GADF-SPCNN	57.90	53.34	55.72	58.00	56.43	56.35	56.29
	DFCNN	58.77	61.15	55.73	61.61	61.06	58.04	59.39
	MSACNN	67.71	65.05	66.29	64.86	65.05	64.95	65.65
2	MTF-SPCNN	93.29	92.43	93.15	94.86	93.58	93.05	93.39
	MTF-ICNN	72.38	74.05	73.09	70.71	73.33	73.57	72.86
	GADF-SPCNN	60.29	61.59	61.19	61.43	62.14	63.14	61.63
	DFCNN	73.51	71.82	72.56	70.39	73.13	74.50	72.65
	MSACNN	75.24	78.66	74.09	71.23	76.43	78.48	75.69
4	MTF-SPCNN	95.05	96.62	96.86	95.25	96.01	95.57	95.89
	MTF-ICNN	80.48	82.86	80.01	84.83	86.43	81.98	82.77
	GADF-SPCNN	70.86	72.29	72.43	71.91	71.79	72.50	71.96
	DFCNN	83.15	81.27	82.37	80.11	84.56	79.72	81.86
	MSACNN	82.38	82.67	87.81	85.55	85.28	82.00	84.28
6	MTF-SPCNN	96.35	97.07	97.86	96.66	97.62	96.72	97.05
	MTF-ICNN	91.38	91.58	92.05	92.28	90.34	91.86	91.58
	GADF-SPCNN	81.72	81.57	81.99	81.79	81.43	80.95	81.58
	DFCNN	90.18	90.01	91.63	91.08	90.56	90.39	90.64
	MSACNN	90.81	90.95	90.67	90.77	90.95	91.24	90.90

下载: 导出CSV

参考文献(24)

[1]	CHENG J S, YANG Y, YANG Y. A rotating machinery fault diagnosis method based on local mean decomposition[J]. Digital Signal Processing, 2012, 22(2): 356-366. doi: 10.1016/j.dsp.2011.09.008
[2]	WANG H Q, KE Y L, LUO G G, et al. Compressed sensing of roller bearing fault based on multiple down-sampling strategy[J]. Measurement Science and Technology, 2016, 27(2): 025009. doi: 10.1088/0957-0233/27/2/025009
[3]	JIANG H K, XIA Y, WANG X D. Rolling bearing fault detection using an adaptive lifting multiwavelet packet with a dimension spectrum[J]. Measurement Science and Technology, 2013, 24(12): 125002. doi: 10.1088/0957-0233/24/12/125002
[4]	袁彩艳, 孙洁娣, 温江涛, 等. 多域信息融合结合改进残差密集网络的轴承故障诊断[J]. 振动与冲击, 2022, 41(4): 200-208. YUAN C Y, SUN J D, WEN J T, et al. Bearing fault diagnosis based on information fusion and improved residual dense networks[J]. Journal of Vibration and Shock, 2022, 41(4): 200-208 (in Chinese).
[5]	HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504-507. doi: 10.1126/science.1127647
[6]	MIKOLOV T, KARAFIÁT M, BURGET L, et al. Recurrent neural network based language model[C]//Proceedings of the 11th Annual Conference of the International Speech Communication Association. Toulouse: ISCA-INST Speech Communication Association, 2010: 1045-1048.
[7]	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
[8]	XU Y, MO T, FENG Q W, et al. Deep learning of feature representation with multiple instance learning for medical image analysis[C]//Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing. Piscataway: IEEE Press, 2014: 1626-1630.
[9]	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
[10]	XU Y N, YANG L. Based on improved CNN bearing fault detection[J]. Journal of Physics: Conference Series, 2022, 2171(1): 012073. doi: 10.1088/1742-6596/2171/1/012073
[11]	赵小强, 张亚洲. 利用改进卷积神经网络的滚动轴承变工况故障诊断方法[J]. 西安交通大学学报, 2021, 55(12): 108-118. ZHAO X Q, ZHANG Y Z. Improved CNN-based fault diagnosis method for rolling bearings under variable working conditions[J]. Journal of Xi’an Jiaotong University, 2021, 55(12): 108-118(in Chinese).
[12]	邓飞跃, 吕浩洋, 顾晓辉, 等. 基于轻量化神经网络Shuffle-SENet的高速动车组轴箱轴承故障诊断方法[J]. 吉林大学学报(工学版), 2022, 52(2): 474-482. DENG F Y, LU H Y, GU X H, et al. Fault diagnosis method of high-speed emu axle box bearing based on lightweight neural network Shuffle-SENet[J]. Journal of Jilin University (Engineering and Technology Edition), 2022, 52(2): 474-482(in Chinese).
[13]	APPANA D K, PROSVIRIN A, KIM J M. Reliable fault diagnosis of bearings with varying rotational speeds using envelope spectrum and convolution neural networks[J]. Soft Computing, 2018, 22(20): 6719-6729. doi: 10.1007/s00500-018-3256-0
[14]	MAO G, ZHANG Z Z, QIAO B, et al. Fusion domain-adaptation CNN driven by images and vibration signals for fault diagnosis of gearbox cross-working conditions[J]. Entropy, 2022, 24(1): 119. doi: 10.3390/e24010119
[15]	梁睿君, 冉文丰, 余传粮, 等. 基于CWT-CNN的齿轮箱运行故障状态识别[J]. 航空动力学报, 2021, 36(12): 2465-2473. LIANG R J, RAN W F, YU C L, et al. Recognition of gearbox operation fault state based on CWT-CNN[J]. Journal of Aerospace Power, 2021, 36(12): 2465-2473(in Chinese).
[16]	谷玉海, 朱腾腾, 饶文军, 等. 基于EMD二值化图像和CNN的滚动轴承故障诊断[J]. 振动、测试与诊断, 2021, 41(1): 105-113. GU Y H, ZHU T T, RAO W J, et al. Rolling bearing fault diagnosis based on EMD binary image and CNN[J]. Journal of Vibration, Measurement & Diagnosis, 2021, 41(1): 105-113(in Chinese).
[17]	WANG Z G, OATES T. Encoding time series as images for visual inspection and classification using tiled convolutional neural networks[C]//Proceedings of the 29th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2015: 40-46.
[18]	WANG Z G, OATES T. Spatially encoding temporal correlations to classify temporal data using convolutional neural networks[EB/OL]. (2015-09-24)[2023-02-01].
[19]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 7132-7141.
[20]	HOU Q B, ZHANG L, CHENG M M, et al. Strip pooling: Rethinking spatial pooling for scene parsing[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 4002-4011.
[21]	ZHANG T C, CHEN J L, LI F D, et al. Intelligent fault diagnosis of machines with small & imbalanced data: A state-of-the-art review and possible extensions[J]. ISA Transactions, 2022, 119: 152-171. doi: 10.1016/j.isatra.2021.02.042
[22]	JIAO J Y, ZHAO M, LIN J, et al. A comprehensive review on convolutional neural network in machine fault diagnosis[J]. Neurocomputing, 2020, 417: 36-63. doi: 10.1016/j.neucom.2020.07.088
[23]	ZHANG J Q, SUN Y, GUO L, et al. A new bearing fault diagnosis method based on modified convolutional neural networks[J]. Chinese Journal of Aeronautics, 2020, 33(2): 439-447. doi: 10.1016/j.cja.2019.07.011
[24]	丁雪, 邓艾东, 李晶, 等. 基于多尺度和注意力机制的滚动轴承故障诊断[J]. 东南大学学报(自然科学版), 2022, 52(1): 172-178. DING X, DENG A D, LI J, et al. Fault diagnosis of rolling bearing based on multi-scale and attention mechanism[J]. Journal of Southeast University (Natural Science Edition), 2022, 52(1): 172-178(in Chinese).