一种用于轴承故障诊断的迁移学习模型

doi:10.13229/j.cnki.jdxbgxb20190493

摘要/Abstract

摘要：

基于人工神经网络的智能故障诊断方法能有效识别设备的健康状况，但传统人工神经网络需要大量有标签样本进行训练，这极大限制了人工神经网络在设备故障诊断中的应用，并且对于不同工况的适应性较差。为解决该问题，提出了一种基于迁移学习理论的轴承故障诊断模型，该模型由栈式稀疏自动编码器（SAE）和柔性最大值函数（Softmax）回归组成，引入高阶KL散度（HKL）训练域自适应能力，可从具有大量已知数据的工况迁移到仅有少量数据的相似工况中。当工况改变时，仅需少量数据进行训练即可适应新工况。采用凯斯西储大学公开的轴承实验数据集验证了该模型的有效性。

关键词: 故障诊断, 人工神经网络, 迁移学习, 自动编码器

Abstract:

Fault diagnosis technology can be used to detect the potential fault of the equipment by analyzing and detecting the signal, so as to ensure the operation safety and effectively improve the operation efficiency of the equipment. The bearings are widely used in rotating machinery and equipment. The fault of the bearings may seriously affect the normal operation of equipment, and inflict economic damage, even endanger the safety of staff. Therefore, it is of great theoretical and practical significance to monitor the health status of the bearing, find out the fault location and analyze its severity in time. In the actual engineering conditions, the operating environment and workspace of mechanical equipment are characterized by complexity and variability. The intelligent fault diagnosis method based on Artificial Neural Network (ANN) can effectively identify the health status of equipment, but the traditional ANN requires a large number of labeled samples for training, which greatly limits its application in equipment fault diagnosis. Also its adaptability to different working conditions is poor. In order to solve this problem, this article proposed a model of bearing fault diagnosis based on transfer learning theory. The model consists of stacked sparse AutoEncoder (SAE) and flexible maximum function (Softmax) regression. In this model, high order KL divergence (HKL) is used to train domain adaptive ability, which can transfer the working condition with a large number of known data to the similar condition with a small amount of data. Only a small amount of data is needed to train the model to adapt to the new working condition. The experimental data set of bearing from Case Western Reserve University was used to verify the effectiveness of the model.

Key words: fault diagnosis, artificial neural network, transfer learning, autoencoder

中图分类号:

TH17

张根保,李浩,冉琰,李裘进. 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.

Gen-bao ZHANG,Hao LI,Yan RAN,Qiu-jin LI. A transfer learning model for bearing fault diagnosis[J]. Journal of Jilin University(Engineering and Technology Edition), 2020, 50(5): 1617-1626.

图/表 11

表1

参考文献 26

1	Yin J L, Wang W Y, Man Z H, et al. Statistical modeling of gear vibration signals and its application to detecting and diagnosing gear faults[J]. Information Sciences, 2014, 259: 295-303.
2	Li W, Zhu Z C, Jiang F, et al. Fault diagnosis of rotating machinery with a novel statistical feature extraction and evaluation method[J]. Mechanical Systems and Signal Processing, 2015, 50/51: 414-426.
3	Shen C Q, Wang D, Kong F R, et al. Fault diagnosis of rotating machinery based on the statistical parameters of wavelet packet paving and a generic support vector regressive classifier[J]. Measurement, 2013, 46(4): 1551-1564.
4	马辉, 车迪, 牛强, 等. 基于深度神经网络的提升机轴承故障诊断研究[J]. 计算机工程与应用, 2019(16): 123-129.
	Ma Hui, Che Di, Niu Qiang, et al. Fault diagnosis of elevator bearing based on deep neural network[J]. Computer Engineering and Applications, 2019(16): 123-129.
5	刘文朋, 廖英英, 杨绍普, 等. 一种基于多点峭度谱和最大相关峭度解卷积的滚动轴承故障诊断方法[J]. 振动与冲击, 2019, 38(2): 146-151, 163.
	Liu Wen-peng, Liao Ying-ying, Yang Shao-pu, et al. Fault diagnosis of rolling bearings based onmultipoint kurtosis spectrums and the maximum correlated kurtosis deconvolution method[J]. Journal of Vibration and Shock, 2019, 38(2): 146-151, 163.
6	郝研, 王太勇, 万剑, 等. 基于经验模式分解和广义维数的机械故障诊断[J]. 吉林大学学报: 工学版, 2012, 42(2): 392-396.
	Hao Yan, Wang Tai-yong, Wan Jian, et al. Mechanical fault diagnosis based on empirical mode decomposition and generalized dimension[J]. Journal of Jilin University(Engineering and Technology Edition), 2012, 42(2): 392-396.
7	高立新, 张建宇, 崔玲丽, 等. 基于小波分析的低速重载设备故障诊断[J]. 机械工程学报, 2005, 41(12): 222-227.
	Gao Li-xin, Zhang Jian-yu, Cui Ling-li, et al. Research on fault diagnosis technology of low speed and heavy duty equipments based on wavelet analysis[J]. Chinese Journal of Mechanical Engineering, 2005, 41(12): 222-227.
8	李文军, 张洪坤, 程秀生. 基于小波和神经网络的传感器故障诊断[J]. 吉林大学学报: 工学版, 2004, 344(3): 491-495.
	Li Wen-jun, Zhang Hong-kun, Cheng Xiu-sheng. Sensor fault diagnosis based on wavelet and neural network[J]. Journal of Jilin University(Engineering and Technology Edition), 2004, 34(3): 491-495.
9	Jia F, Lei Y G, Lin J, et al. Deep neural networks: a promising tool for fault characteristic mining and intelligent diagnosis of rotating machinery with massive data[J]. Mechanical Systems and Signal Processing, 2016, 72/73: 303-315.
10	牟亮, 王凯, 李彦, 等. 层叠P阶多项式主成分分析在轴承故障诊断中的应用[J]. 振动与冲击, 2019, 38(2): 25-32.
	Mu Liang, Wang Kai, Li Yan, et al. Bearing fault diagnosis based on the stacked P-order polynomial principal component analysis[J]. Journal of Vibration and Shock, 2019, 38(2): 25-32.
11	程利军, 张英堂, 李志宁, 等. 基于时频分析及阶比跟踪的曲轴轴承故障诊断研究[J]. 振动与冲击, 2012, 31(19): 73-78.
	Cheng Li-jun, Zhang Ying-tang, Li Zhi-ning, et al. Research on fault diagnose of main bearing based on time-frequency analysis and order tracking[J]. Journal of Vibration and Shock, 2012, 31(19): 73-78.
12	陶新民, 徐晶, 刘兴丽, 等. 基于最大小波奇异谱的轴承故障诊断方法[J]. 振动、测试与诊断, 2010, 30(1): 78-82.
	Tao Xin-min, Xu Jing, Liu Xing-li, et al. Fault diagnosis of bearing using maximum wavelet singular spectrum[J]. Journal of Vibration, Measurement & Diagnosis, 2010, 30(1): 78-82.
13	庄福振, 罗平, 何清, 等. 迁移学习研究进展[J]. 软件学报, 2015, 26(1): 26-39.
	Zhuang Fu-zhen, Luo Ping, He Qing, et al. Survey on transfer learning research[J]. Journal of Software, 2015, 26(1): 26-39.
14	Zhang R, Tao H Y, Wu L F, et al. Transfer learning with neural networks for bearing fault diagnosis in changing working conditions[J]. IEEE Access, 2017(5): 14347-14357.
15	Han D M, Liu Q G, Fan W G. A new image classification method using CNN transfer learning and web data augmentation[J]. Expert Systems with Applications, 2018, 95: 43-56.
16	Chen D M, Yang S, Zhou F N. Incipient fault diagnosis based on DNN with transfer learning[C]∥International Conference on Control, Automation and Information Sciences, Hangzhou, China, 2018: 303-308.
17	Karsten M B, Arthur G, Malte R, et al. Integrating structured biological data by kernel maximum mean discrepancy[J]. Bioinformatics, 2006, 22(14): 49-57.
18	Lu W N, Liang B, Cheng Y, et al. Deep model based domain adaptation for fault diagnosis[J]. IEEE Transactions on Industrial Electronics, 2017, 64(3): 2296-2305.
19	Wen L, Gao L, Li X Y. A new deep transfer learning based on sparse auto-encoder for fault diagnosis[J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2019, 49(1): 136-144.
20	Ding Z M, Fu Y. Robust transfer metric learning for image classification[J]. IEEE Transactions on Image Processing, 2017, 26(2): 660-670.
21	Jiang L, Ge Z Q, Song Z H. Semi-supervised fault classification based on dynamic sparse stacked auto-encoders model[J]. Chemometrics and Intelligent Laboratory Systems, 2017, 168: 72-83.
22	Zheng Y L, Wang T Z, Xin B, et al. A sparse autoencoder and softmax regression based diagnosis method for the attachment on the blades of marine current turbine[J]. Sensors, 2019, 19(4): 826.
23	Qian W W, Li S M, Wang J R. A new transfer learning method and its application on rotating machine fault diagnosis under variant working conditions[J]. IEEE Access, 2018（6）: 1-11.
24	朱冰, 蒋渊德, 邓伟文, 等. 基于KL散度的驾驶员驾驶习性非监督聚类[J]. 汽车工程, 2018, 40(11): 1317-1323.
	Zhu Bing, Jiang Yuan-de, Deng Wei-wen, et al. Unsupervised clustering of driving styles based on KL divergence[J]. Automotive Engineering, 2018, 40(11): 1317-1323.
25	Case Western Reserve University. Bearing data center[DB/OL]. [2019-04-28]. ﬁle
26	Yang B, Lei Y G, Jia F, et al. An intelligent fault diagnosis approach based on transfer learning from laboratory bearings to locomotive bearings[J]. Mechanical Systems and Signal Processing, 2019, 122: 692-706.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

符号	意义	符号	意义
$X s$ 、 $X t$	源域/目标域数据集	$y s i$ 、 $y t i$	源域/目标域样本的标签
$x s$ 、 $x t$	源域/目标域样本	$N i n$ 、 $N o u t$	自动编码器输入/输出维数
$μ s n$ 、 $μ t n$	源域/目标域数据集的n阶矩	$k s$ 、 $k t$	源域/目标域样本数
$h s (i)$ 、 $h t (i)$	源域/目标域在第i层编码器的编码特征矩阵	$δ$	HKL散度的最大阶数

工况（域）	载荷（HP）	转速/ （r·min^-1）	样本数	数据点数（每个样本）	健康状态
Ⅰ	0	1797	5000	1200	10
Ⅱ	1	1772	5000	1200	10
Ⅲ	2	1750	5000	1200	10
Ⅴ	3	1730	5000	1200	10

健康状态	方法1	方法2	方法3	方法4	方法5	方法6
平均	55.15	81.99	89.46	96.11	97.14	99.47
N	100.00	100.00	100.00	100.00	100.00	100.00
IF18	50.97	80.54	88.52	95.18	96.97	99.34
IF36	43.48	75.68	82.94	94.73	96.00	99.19
IF54	49.69	79.37	85.66	95.54	96.39	99.19
OF18	50.39	80.26	87.93	92.89	97.28	99.58
OF36	52.87	81.39	89.32	97.19	97.67	99.69
OF54	51.96	78.34	90.11	96.46	97.05	99.44
BF18	47.84	80.96	87.44	95.33	96.31	99.34
BF36	51.95	81.14	91.28	97.17	97.22	99.46
BF54	52.34	82.25	91.39	96.58	96.54	99.45

训练样本	源域样本容量	目标域样本容量	健康状态数
甲	4000	1000	10
乙	4000	750	10
丙	4000	500	10

方法	训练样本	Ⅰ-Ⅱ	Ⅰ-Ⅲ	Ⅰ-Ⅴ	Ⅱ-Ⅰ	Ⅱ-Ⅲ	Ⅱ-Ⅴ	Ⅲ-Ⅰ	Ⅲ-Ⅱ	Ⅲ-Ⅴ	Ⅴ-Ⅰ	Ⅴ-Ⅱ	Ⅴ-Ⅲ	平均
TSAE	甲	89.46	88.33	87.91	88.97	88.77	88.19	88.39	88.82	88.44	88.09	88.24	88.37	88.50
	乙	87.94	86.42	84.61	84.21	83.89	84.77	85.62	84.95	83.94	83.71	84.56	83.74	84.86
	丙	84.21	83.79	81.33	81.07	80.42	82.04	83.74	81.86	80.69	81.36	82.01	80.97	81.96
SAE+ Softmax+ KL	甲	96.11	90.94	92.44	97.38	96.51	92.68	97.91	96.99	98.00	91.36	93.98	98.74	95.25
	乙	95.83	90.61	92.14	97.05	96.18	92.30	97.66	96.72	97.08	91.19	93.59	98.66	94.92
	丙	94.69	89.11	91.09	96.52	95.44	91.00	96.92	95.53	96.07	90.76	91.99	97.48	93.88
FTNN	甲	97.14	97.49	89.16	98.00	98.64	98.28	97.51	98.41	99.31	92.94	97.99	98.18	96.92
	乙	96.94	95.18	89.02	97.88	98.21	97.99	97.23	98.19	99.26	92.83	97.88	98.10	96.56
	丙	94.29	94.68	88.33	96.68	97.64	97.69	95.90	97.73	97.88	92.42	97.39	97.66	95.69
SAE+ Softmax+ HKL	甲	99.47	99.35	99.62	99.19	98.98	98.47	99.29	98.96	99.64	99.60	98.39	99.51	99.21
	乙	99.11	99.27	99.49	98.86	98.42	98.44	99.10	98.59	99.43	99.14	97.98	99.34	98.93
	丙	98.94	99.16	99.27	97.87	97.71	97.31	98.94	97.87	98.93	98.06	97.54	98.22	98.32