深度在线小波极限学习在旋转机械故障诊断中的应用

摘要: 由于旋转机械故障诊断模型训练时间长，容易过拟合以及传统的极限学习机只能处理批量数据，实效性差等问题。提出一种基于深度在线小波极限学习机的旋转机械故障诊断方法。将自编码器的思想引入小波极限学习机中，堆叠形成WELM-AE，将底层的故障特征向更加抽象的高级特征转换。再采用在线极限学习机作为顶层分类器进行故障识别。实验结果验证：该算法在旋转机械故障诊断上的可行性，继承了极限学习机训练速度快的特点，相较于BP、SVM、SAE、CNN有更高的准确率。
- 旋转机械 /
- 故障诊断 /
- 深度小波极限学习机自编码器 /
- 在线极限学习机
Abstract: Due to the long training time of the rotating machinery fault diagnosis model, it is easy to overfit and the traditional extreme learning machine can only handle batch data, and the effectiveness is poor. A fault diagnosis method for rotating machinery based on deep online wavelet extreme learning machine is proposed. Introducing the idea of autoencoder into the wavelet extreme learning machine, stacking to form WELM-AE converts the underlying fault features to more abstract and advanced features. And the online extreme learning machine is used as the top-level classifier for fault identification. The experimental results verify the feasibility of the method in the fault diagnosis of rotating machinery, inheriting the characteristics of the fast training speed of the extreme learning machine, and having a higher accuracy rate than BP, SVM, SAE and CNN.
- rotating machinery /
- fault diagnosis /
- deep wavelet extreme learning machine autoencoder /
- online extreme learning machine

HTML全文

图 1 WELM-AE结构图

Figure 1. WELM-AE structure diagram

下载: 全尺寸图片幻灯片

图 2 DWOSELM结构图

Figure 2. DWOSELM structure diagram

下载: 全尺寸图片幻灯片

图 3 不同激活函数10次实验平均准确率

Figure 3. 3 Average accuracy of 10 experiments with different activation functions

下载: 全尺寸图片幻灯片

图 4 不同小波函数10次实验平均诊断准确率

Figure 4. Average diagnostic accuracy of 10 experiments with different wavelet functions

下载: 全尺寸图片幻灯片

图 5 优化算法适应度曲线

Figure 5. Fitness curve of the optimization algorithm

下载: 全尺寸图片幻灯片

图 6 各模型分类评价结果

Figure 6. Evaluation results for each model classification

下载: 全尺寸图片幻灯片

图 7 诊断结果混淆矩阵

Figure 7. The confusion matrix of the diagnostic results

下载: 全尺寸图片幻灯片

图 8 各模型分类评价结果

Figure 8. Evaluation results for each model classification

下载: 全尺寸图片幻灯片

图 9 诊断结果混淆矩阵

Figure 9. The confusion matrix of the diagnostic results

下载: 全尺寸图片幻灯片

表 1 隐含层参数选择

Table 1. Selection of hidden layer parameters

隐含层参数选择	准确率/%	时间/s
N₁=10, N₂=10	89.17	3.15
N₁=10, N₂=20	89.72	3.62
N₁=10, N₂=30	89.72	4.07
N₁=20, N₂=20	93.89	4.20
N₁=20, N₂=30	93.61	4.94
N₁=30, N₂=30	95.00	5.80
N₁=30, N₂=40	94.44	6.42
N₁=10, N₂=10, N₃=10	90.28	3.53
N₁=10, N₂=20, N₃=30	90.56	5.45
N₁=20, N₂=20, N₃=20	93.89	4.98
N₁=30, N₂=30, N₃=30	95.28	6.19
N₁=40, N₂=40, N₃=40	94.44	8.92

下载: 导出CSV

表 2 齿轮箱故障类别

Table 2. Gearbox fault categories

运行状态	类别
正常	1
点蚀故障	2
点磨(大齿轮点蚀和小齿轮磨损)	3
断齿故障	4
断磨(大齿轮断齿和小齿轮磨损)	5
磨损故障	6

下载: 导出CSV

表 3 诊断结果对比

Table 3. Comparison of diagnostic results

模型	平均准确率/%	训练时间/s
SVM	90.73	0.30
BP	91.24	0.20
ELM	92.67	0.19
SAE	91.85	9.19
CNN	93.29	17.19
DELM	92.59	2.95
本文方法	95.35	5.32

下载: 导出CSV

表 4 滚动轴承故障类别

Table 4. Rolling bearing fault categories

运行状态	类别标签
正常	1
滚动体故障	2
内圈故障	3
外圈故障	4

下载: 导出CSV

表 5 诊断结果对比

Table 5. Comparison of diagnostic results

模型	平均准确率/%	训练时间/s
SVM	92.56	0.21
BP	90.73	0.19
ELM	93.67	0.16
SAE	91.85	6.50
CNN	92.78	14.23
DELM	90.36	1.82
本文方法	96.73	4.74

下载: 导出CSV

参考文献(15)

[1]	张英堂, 马超, 尹刚, 等. 基于多极限学习机在线集成的柴油机故障诊断方法研究[J]. 车用发动机, 2012(6): 85-89. https://www.cnki.com.cn/Article/CJFDTOTAL-CYFD201206021.htm ZHANG Y T, MA C, YIN G, et al. Diesel engine fault diagnosis based on online ensemble of multiple extreme learning machine[J]. Vehicle Engine, 2012(6): 85-89. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-CYFD201206021.htm
[2]	XIA M, LI T, XU L, et al. Fault diagnosis for rotating machinery using multiple sensors and convolutional neural networks[J]. IEEE／ASME Transactions on Mechatronics, 2018, 23(1): 101-110. doi: 10.1109/TMECH.2017.2728371
[3]	JIAN Y F, QING X G, HE L, et al. Fault diagnosis of motor bearing based on deep learning[J]. Advances in Mechanical Engineering, 2019, 11(9): 1-9.
[4]	熊景鸣, 潘林, 朱昇, 等. DBN与PSO-SVM的滚动轴承故障诊断[J]. 机械科学与技术, 2019, 38(11): 1726-1731. doi: 10.13433/j.cnki.1003-8728.20190040 XIONG J M, PAN L, ZHU S, et al. Bearing fault diagnosis based on deep belief networks and particle swarm optimization support vector machine[J]. Mechanical Science and Technology for Aerospace Engineering, 2019, 38(11): 1726-1731. (in Chinese) doi: 10.13433/j.cnki.1003-8728.20190040
[5]	CAMBRIA E, HUANG G B, KASUN L L C, et al. Extreme learning machines [trends & controversies][J]. IEEE Intelligent Systems, 2013, 28(6): 30-59. doi: 10.1109/MIS.2013.140
[6]	TANG J X, DENG C W, HUANG G B. Extreme learning machine for multilayer perceptron[J]. IEEE Transactions on Neural Networks and Learning Systems, 2016, 27(4): 809-821. doi: 10.1109/TNNLS.2015.2424995
[7]	蔡念, 胡匡祜, 李淑宇, 等. 小波神经网络及其应用[J]. 中国体视学与图像分析, 2001, 6(4): 239-245. https://www.cnki.com.cn/Article/CJFDTOTAL-ZTSX200104013.htm CAI N, HU K H, LI S Y, et al. Wavelet neural network and its application[J]. Chinese Journal of Stereology and Image Analysis, 2001, 6(4): 239-245. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-ZTSX200104013.htm
[8]	俞阿龙. 基于遗传小波神经网络的机器人腕力传感器动态建模研究[J]. 物理学报, 2008, 57(6): 3385-3390. https://www.cnki.com.cn/Article/CJFDTOTAL-WLXB200806016.htm YU A L. Research on the dynamic modeling based on genetic wavelet neural network for the robot wrist force sensor[J]. Acta Physica Sinica, 2008, 57(6): 3385-3390. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-WLXB200806016.htm
[9]	缪志强, 王耀南. 基于径向小波神经网络的混沌系统鲁棒自适应反演控制[J]. 物理学报, 2012, 61(3): 030503. https://www.cnki.com.cn/Article/CJFDTOTAL-WLXB201203012.htm MIAO Z Q, WANG Y N. Robust adaptive radial wavelet neural network control for chaotic systems using backstepping design[J]. Acta Physica Sinica, 2012, 61(3): 030503. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-WLXB201203012.htm
[10]	LIANG N Y, HUANG G B, SARATCHANDRAN P, et al. A fast and accurate online sequential learning algorithm for feedforward networks[J]. IEEE Transactions on Neural Networks, 2006, 17(6): 1411-1423.
[11]	陈哲, 冯天瑾. 小波分析与神经网络结合的研究进展[J]. 电子科学学刊, 2000, 22(3): 496-504. https://www.cnki.com.cn/Article/CJFDTOTAL-DZYX200003023.htm CHEN Z, FENG T J. Research advances on combinatin of wavelet analysis and neural networks[J]. Journal of Electronics, 2000, 22(3): 496-504. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-DZYX200003023.htm
[12]	HINTON G E, ZEMEL R S. Autoencoders, Minimum description length and Helmholtz free energy[C]／／Proceedings of the 6th International Conference on Neural Information Processing Systems. Denver: Morgan Kaufmann Publishers Inc., 1993: 3-10.
[13]	谢群辉, 田青. 权重随机正交化的极速非线性判别分析网络[J]. 计算机技术与发展, 2018, 28(1): 23-27. https://www.cnki.com.cn/Article/CJFDTOTAL-WJFZ201801005.htm XIE Q H, TIAN Q. Nonlinear discriminant analysis networks with random and orthogonalized input weights[J]. Computer Technology and Development, 2018, 28(1): 23-27. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-WJFZ201801005.htm
[14]	WANG C, PEDRYCZ W, YANG J B, et al. Wavelet frame-based fuzzy C-means clustering for segmenting images on graphs[J]. IEEE Transactions on Cybernetics, 2020, 50(9): 3938-3949.
[15]	于惠鸣, 张智晟, 龚文杰, 等. 基于深度递归神经网络的电力系统短期负荷预测模型[J]. 电力系统及其自动化学报, 2019, 31(1): 112-116. https://www.cnki.com.cn/Article/CJFDTOTAL-DLZD201901020.htm YU H M, ZHANG Z S, GONG W J, et al. Short-term load forecasting model of power system based on deep recurrent neural network[J]. Proceedings of the CSU-EPSA, 2019, 31(1): 112-116. (in Chinese) https://www.cnki.com.cn/Article/CJFDTOTAL-DLZD201901020.htm