基于一维WConv-BiLSTM的轴承故障诊断算法

doi:10.16180/j.cnki.issn1007-7820.2021.04.012

摘要/Abstract

摘要：

针对不同信噪比噪声干扰下的轴承故障诊断问题,文中建立了一种基于一维宽卷积核卷积神经网络和双向长短记忆神经网络的轴承故障诊断模型。向该模型输入轴承振动信号,通过短时傅里叶变换将振动信号转化为时频图。然后利用首层为宽卷积核的卷积神经网络和长短记忆神经网络分别提取其空间与时间特征,并结合全连接层实现分类。为增强抗噪性,模型采用数据增强、mini-batch和批量标准化的方法。试验采用CWRU的轴承故障数据集,通过添加不同信噪比噪声构造噪声干扰试验。试验表明该模型有较好的抗噪性,在信噪比为2 dB的噪声干扰下仍能取得98%以上的故障识别准确率。

关键词: 轴承故障诊断, 卷积神经网络, 宽卷积核, 双向长短记忆神经网络, 短时傅里叶变换, 抗噪性

Abstract:

In view of bearing fault diagnosis under different signal-to-noise ratio noise interference, a bearing fault diagnosis model based on one-dimensional wide convolution kernel convolutional neural network and bidirectional long short-term memory neural network is established. The bearing vibration signals are used as input to the proposed model, and the short-time Fourier transform is used to transform the vibration signal into the time-frequency figures. Convolutional neural networks with wide convolution kernels and bidirectional long short-term memory neural networks are applied to extract the spatial and temporal characteristics of data, and classification is achieved through dense layers. Data augmentation, mini-batch, and batch normalization methods are utilized to enhance the anti-noise performance of the model. The test uses the CWRU bearing fault data set, and adds noise with different SNR to construct noise interference tests. The experiments show that proposed model has great noise resistance, and can achieve 98% accuracy of fault recognition with a signal-to-noise ratio of 2 dB.

Key words: bearing fault diagnosis, CNN, wide convolution kernels, BiLSTM, short-time Fourier transform, noise immunity

中图分类号:

TP389.1

闫书豪,乔美英. 基于一维WConv-BiLSTM的轴承故障诊断算法[J]. 电子科技, 2021, 34(4): 75-82.

YAN Shuhao,QIAO Meiying. Bearing Fault Diagnosis Algorithm Based on One-Dimensional WConv-BiLSTM[J]. Electronic Science and Technology, 2021, 34(4): 75-82.

图/表 15

图1

图2

图3

图4

图5

图6

表1

图7

图8

表2

图9

图10

表3

表4

图11

参考文献 17

[1]	Rabah A, Abdelhafid K, Azeddine B , et al. Rolling bearing fault diagnosis based on an improved denoising method using the complete ensemble empirical mode decomposition and the optimized thresholding operation[J]. IEEE Sensors Journal, 2018,18(17):7166-7172.
[2]	王建国, 陈帅, 张超. 噪声参数最优ELMD与LS-SVM在轴承故障诊断中的应用与研究[J]. 振动与冲击, 2017,36(5):72-78,86.
	Wang Jianguo, Chen Shuai, Zhang Chao. Application of noise parametric optimization with ELMD and LS-SVM in bearing fault diagnosis[J]. Journal of Vibration & Shock, 2017,36(5):72-78,86.
[3]	余忠潇, 郝如江. 基于小波包和LCD的滚动轴承故障诊断[J]. 组合机床与自动化加工技术, 2019(10):120-123.
	Xu Zhongxiao, Hao Rujiang. Application of bearing fault diagnosis based on wavelet packet and LCD[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2019(10):120-123.
[4]	向丹, 岑健. 基于EMD熵特征融合的滚动轴承故障诊断方法[J]. 航空动力学报, 2015,30(5):1149-1155.
	Xiang Dan, Cen Jian. Method of roller bearing fault diagnosis based on feature fusion of EMD entropy[J]. Journal of Aerospace Power, 2015,30(5):1149-1155.
[5]	昝涛, 庞兆亮, 王民 , 等. 基于VMD的滚动轴承早期故障诊断方法[J]. 北京工业大学学报, 2019,45(2):103-110.
	Zan Tao, Pang Zhaoliang, Wang Min , et al. Early fault diagnosis method of rolling bearings based on VMD[J]. Journal of Beijing University of Technology, 2019,45(2):103-110.
[6]	Sheng L, Yue S, Lanyong Z. A novel fault diagnosis method based on noise-assisted MEMD and functional neural fuzzy network for rolling element bearings[J]. IEEE Access, 2018(6):27048-27068.
[7]	王新, 闫文源. 基于变分模态分解和SVM的滚动轴承故障诊断[J]. 振动与冲击, 2017,36(18):252-256.
	Wang Xin, Yan Wenyuan. Fault diagnosis of roller bearings based on the variational mode decomposition and SVM[J]. Journal of Vibration & Shock, 2017,36(18):252-256.
[8]	Liu R N, Yang B Y , et al. Artificial intelligence for fault diagnosis of rotating machinery:A review[J]. Mechanical Systems and Signal Processing, 2018,10(8):33-47.
[9]	梁蒙蒙, 周涛, 夏勇 , 等. 基于随机化融合和CNN的多模态肺部肿瘤图像识别[J]. 南京大学学报(自然科学), 2018,54(4):775-785.
	Liang Mengmeng, Zhou Tao, Xia Yong , et al. Multimodal lung tumor image recognition based on randomized fusion and CNN[J]. Journal of Nanjing University (Natural Science), 2018,54(4):775-785.
[10]	王纪军, 靖慧, 冯曙明 , 等. 基于Faster R-CNN的仓库视频监控目标检测方法研究[J]. 信息技术, 2019,43(7):92-96.
	Wang Jijun, Jing Hui, Feng Shuming , et al. Object detection method of video monitoring in power warehouse based on Faster R-CNN[J]. Information Technology, 2019,43(7):92-96.
[11]	Zhang W, Zhang F, Chen W , et al. Fault state recognition of rolling bearing based fully convolutional network[J]. Computing in Science & Engineering, 2018,21(5):55-63.
[12]	Zhang W, Peng G L, Li C , 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):425-433.
[13]	胡晓依, 荆云建, 宋志坤 , 等. 基于CNN-SVM的深度卷积神经网络轴承故障识别研究[J]. 振动与冲击, 2019,38(18):173-178.
	Hu Xiaoyi, Jing Yunjian, Song Zhikun , et al. Bearing fault identification by using deep convolution neural networks based on CNN-SVM[J]. Journal of Vibration and Shock, 2019,38(18):173-178.
[14]	李恒, 张氢, 秦仙蓉 , 等. 基于短时傅里叶变换和卷积神经网络的轴承故障诊断方法[J]. 振动与冲击, 2018,37(19):124-131.
	Li Heng, Zhang Qing, Qin Xianrong , et al. Fault diagnosis method for rolling bearings based on short-time Fourier transform and convolution neural network[J]. Journal of Vibration and Shock, 2018,37(19):124-131.
[15]	李业良, 张二华, 唐振民. 基于混合式注意力机制的语音识别研究[J]. 计算机应用研究, 2020,37(1):131-134.
	Li Yeliang, Zhang Erhua, Tang Zhenmin. Research on speech recognition based on hybrid attention mechanism[J]. Application Research of Comporters, 2020,37(1):131-134.
[16]	张悦, 胡春燕. 基于有记忆递归神经网络的脑电特征情感识别研究[J]. 电子科技, 2020,33(11):67-72.
	Zhang Yue, Hu Chunyan. Research on emotion recognition of EEG features based on the long short-term memory neural network[J]. Electronic Science and Technology, 2020,33(11):67-72.
[17]	Yildirim Ô. A novel wavelet sequences based on deep bidirectional LSTM network model for ECG signal classification[J]. Computers in Biology & Medicine, 2018,19(6):189-202.

标签	故障等级	故障直径 /mm	Load B /HP	Load C /HP	Load D /HP
1	正常	0	1	2	3
2	内圈轻微	0.18	1	2	3
3	内圈中等	0.36	1	2	3
4	内圈严重	0.54	1	2	3
5	外圈轻微	0.18	1	2	3
6	外圈中等	0.36	1	2	3
7	外圈严重	0.54	1	2	3
8	滚珠轻微	0.18	1	2	3
9	滚珠中等	0.36	1	2	3
10	滚珠严重	0.54	1	2	3

层	类型	输出尺寸	卷积核尺寸/步长	神经元数量	卷积核数量	Dropout	激活函数	池化	批量标准化
1	Input_F	65×65	-	-	-	-	-	-	N
2	WConv	65×16	32×1/1	-	16	0.85	RELU	2	Y
3	1D-Conv	32×32	2×1/1	-	32	0.85	RELU	2	Y
4	1D-Conv	16×64	2×1/1	-	64	0.85	RELU	1	Y
5	BiLSTM	40×1	-	(40,40)	-		Tanh	-	N
6	Dense1	100×1	-	100	-	0.9	-	-	N
7	Dense2	10	-	10	-	-	Softmax	-	N

模型	SNR
模型	-4 dB	-2 dB	0 dB	2 dB	4 dB	6 dB	8 dB	10 dB
CNN	34.95%	48.25%	56.51%	69.52%	71.13%	88.43%	91.63%	92.87%
CNN-LSTM	54.65%	61.95%	78.35%	85.95%	90.63%	91.52%	91.56%	93.13%
WConvBiLSTM	73.33%	79.48%	85.97%	88.48%	92.79%	93.48%	94.87%	95.9%
Mini-batch\ BN-Mode	78.24%	81.26%	89.78%	92.34%	94.12%	95.13%	95.29%	97.16%
DE-Mode	85.23%	92.48%	95.97%	98.48%	98.79%	99.48%	99.87%	99.9%

分组	负载/HP	训练集	测试集
A-B	0~1	A	B
B-D/D-B	1~2/2~1	B/D	D/B
B-AB	1~012	B	AB
B-ABCD	1~0123	B	ABCD