基于特征金字塔卷积循环神经网络的故障诊断方法

doi:10.16183/j.cnki.jsjtu.2021.001

摘要/Abstract

摘要：

变工况、变载荷设备部件不同故障的特征在信号中所占比例和位置不固定,且包括大量不同场景下的原始振动信号的多尺度复杂性.对此,提出一种基于特征金字塔网络(FPN)的卷积循环神经网络(CRNN)滚动轴承故障诊断方法.利用卷积神经网络(CNN)框架,并联CNN的卷积层和循环神经网络(RNN)中的长短时记忆(LSTM)层,形成新的CRNN,以充分利用CNN对空间域信息和RNN对时域信息的学习能力;在每一层中权值共享,减少网络参数;利用FPN构建全新特征图,输入一维信号和堆叠后形成的二维信号,对传感器采集的信号进行特征提取,实现故障诊断.利用行星齿轮箱进行故障试验,并进行5折交叉验证,该方法的诊断准确率平均值为99.20%,比基本神经网络模型至少高3.62%,表明该方法诊断精度高、鲁棒性强;利用凯斯西储大学轴承数据集进行验证,证明该方法具有良好的泛用性;利用t-SNE方法对模型的特征学习效果进行可视化分析,结果表明不同故障类别特征具有良好的聚类效果.

关键词: 卷积循环神经网络, 特征金字塔, 故障诊断, 特征可视化

Abstract:

Aimed at the problems that the proportion and position of different fault characteristics of equipment components under variable working conditions and variable load in the signal are not fixed, and include the multi-scale complexity of the original vibration signal in a large number of different scenarios, a convolutional recurrent neural network (CRNN) rolling bearing fault diagnosis method based on feature pyramid network (FPN) was proposed. Using the convolution neural network (CNN) framework, the convolution layer of CNN and the long and short-term memory (LSTM) layer of recurrent neural network (RNN) were connected in parallel to form a new CRNN, so as to make full use of the learning ability of CNN to spatial domain information and RNN to time domain information. The weights were shared in each layer to reduce network parameters. A novel feature map was constructed using FPN, and one-dimensional signal and two-dimensional signal formed after stacking were input to extract the feature of the signal collected by the sensor, and realize fault diagnosis. The average diagnostic accuracy of this method is 99.20%, which is at least 3.62% higher than that of the basic neural network model, indicating that this method has a high diagnostic accuracy and a strong robustness. Using the bearing data set of Case Western Reserve University, it is proved that the method has a good universality. The t-SNE method was used to visually analyze the feature learning effect of the model. The results show that different fault category features have good clustering effect.

Key words: convolutional recurrent neural network (CRNN), feature pyramid, fault diagnosis, feature visualization

中图分类号:

刘秀丽, 徐小力. 基于特征金字塔卷积循环神经网络的故障诊断方法[J]. 上海交通大学学报, 2022, 56(2): 182-190.

LIU Xiuli, XU Xiaoli. A Fault Diagnosis Method Based on Feature Pyramid CRNN Network[J]. Journal of Shanghai Jiao Tong University, 2022, 56(2): 182-190.

图/表 21

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

表1

表2

图11

表3

图12

表4

图13

表5

图14

表6

图15

参考文献 20

[1]	辛玉, 李舜酩, 王金瑞, 等. 基于迭代经验小波变换的齿轮故障诊断方法[J]. 仪器仪表学报, 2018, 39(11):79-86.
	XIN Yu, LI Shunming, WANG Jinrui, et al. Gear fault diagnosis method based on iterative empirical wavelet transform[J]. Chinese Journal of Scientific Instrument, 2018, 39(11):79-86.
[2]	陈旭阳, 韩振南, 宁少慧. 自适应改进双树复小波变换的齿轮箱故障诊断[J]. 振动、测试与诊断, 2019, 39(5):1016-1022.
	CHEN Xuyang, HAN Zhennan, NING Shaohui. Gearbox fault diagnosis based on adaptive modified dual-tree complex wavelet transform[J]. Journal of Vibration, Measurement & Diagnosis, 2019, 39(5):1016-1022.
[3]	朱喜华, 李颖晖, 周飞帆, 等. 基于改进EMD算法的永磁同步电机故障特征提取[J]. 微电机, 2011, 44(2):65-69.
	ZHU Xihua, LI Yinghui, ZHOU Feifan, et al. Feature extraction for PMSM based on ameliorated EMD arithmetic[J]. Micromotors, 2011, 44(2):65-69.
[4]	高冠琪, 黄伟国, 李宁, 等. 基于时频挤压和阶比分析的变转速轴承故障检测方法[J]. 振动与冲击, 2020, 39(3):205-210.
	GAO Guanqi, HUANG Weiguo, LI Ning, et al. Fault detection method for varying rotating speed bearings based on time-frequency squeeze and order analysis[J]. Journal of Vibration and Shock, 2020, 39(3):205-210.
[5]	陈剑, 夏康, 黄凯旋, 等. 基于VMD相对能量熵和自适应ARMA模型的轴承性能退化趋势动态预警[J]. 电子测量与仪器学报, 2020, 34(8):116-123.
	CHEN Jian, XIA Kang, HUANG Kaixuan, et al. Dynamic prediction of bearing performance degradation trend based on VMD relative energy entropy and adaptive ARMA model[J]. Journal of Electronic Measurement and Instrumentation, 2020, 34(8):116-123.
[6]	王华庆, 任帮月, 宋浏阳, 等. 基于终止准则改进K-SVD字典学习的稀疏表示特征增强方法[J]. 机械工程学报, 2019, 55(7):35-43.
	WANG Huaqing, REN Bangyue, SONG Liuyang, et al. Sparse representation method based on termination criteria improved K-SVD dictionary learning for feature enhancement[J]. Journal of Mechanical Engineering, 2019, 55(7):35-43.
[7]	唐立力. 基于粗糙遗传BP神经网络的滚动轴承故障诊断[J]. 机械工程与自动化, 2018(3):138-140.
	TANG Lili. Fault diagnosis on rolling bearing based on rough genetic BP neural network[J]. Mechanical Engineering & Automation, 2018(3):138-140.
[8]	吴伟, 郑娟. 基于BP神经网络的齿轮故障诊断[J]. 机械研究与应用, 2015, 28(1):88-90.
	WU Wei, ZHENG Juan. Gear fault diagnosis based on BP neural network[J]. Mechanical Research & Application, 2015, 28(1):88-90.
[9]	张建, 李艳军, 曹愈远, 等. 免疫支持向量机用于航空发动机磨损故障诊断[J]. 北京航空航天大学学报, 2017, 43(7):1419-1425.
	ZHANG Jian, LI Yanjun, CAO Yuyuan, et al. Immune SVM used in wear fault diagnosis of aircraft engine[J]. Journal of Beijing University of Aeronautics and Astronautics, 2017, 43(7):1419-1425.
[10]	贺立敏, 王岘昕, 韩冰. 基于随机森林和支持向量机的船舶柴油机故障诊断[J]. 中国航海, 2017, 40(2):29-33.
	HE Limin, WANG Xianxin, HAN Bing. Fault diagnosis of marine diesel engine based on random forest and support vector machine[J]. Navigation of China, 2017, 40(2):29-33.
[11]	CHENG J, WANG P S, LI G, et al. Recent advances in efficient computation of deep convolutional neural networks[J]. Frontiers of Information Technology & Electronic Engineering, 2018, 19(1):64-77.
[12]	GU J X, WANG Z H, KUEN J, et al. Recent advances in convolutional neural networks[J]. Pattern Recognition, 2018, 77:354-377. doi: 10.1016/j.patcog.2017.10.013 URL
[13]	KRIZHEVSKY A, SUTSKEVER I, HINTON G. Imagenet classification with deep convolutional neural networks[J]. Advances in Neural Information Processing Systems, 2017, 60(6):84-90.
[14]	曲建岭, 余路, 袁涛, 等. 基于一维卷积神经网络的滚动轴承自适应故障诊断算法[J]. 仪器仪表学报, 2018, 39(7):134-143.
	QU Jianling, YU Lu, YUAN Tao, et al. Adaptive fault diagnosis algorithm for rolling bearings based on one-dimensional convolutional neural network[J]. Chinese Journal of Scientific Instrument, 2018, 39(7):134-143.
[15]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. Computer Science, 2014, 32(5):86-93.
[16]	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.
[17]	韩涛, 袁建虎, 唐建, 等. 基于MWT和CNN的滚动轴承智能复合故障诊断方法[J]. 机械传动, 2016, 40(12):139-143.
	HAN Tao, YUAN Jianhu, TANG Jian, et al. An approach of intelligent compound fault diagnosis of rolling bearing based on MWT and CNN[J]. Journal of Mechanical Transmission, 2016, 40(12):139-143.
[18]	WU C Z, JIANG P C, FENG P Z, et al. Faults diagnosis method for gearboxes based on a 1-D convolutional neural network[J]. Journal of Vibration and Shock, 2018, 37(22):51-56.
[19]	李东东, 王浩, 杨帆, 等. 基于一维卷积神经网络和Soft-Max分类器的风电机组行星齿轮箱故障检测[J]. 电机与控制应用, 2018, 45(6):80-87.
	LI Dongdong, WANG Hao, YANG Fan, et al. Fault detection of wind turbine planetary gear box using 1D convolution neural networks and soft-max classifier[J]. Electric Machines & Control Application, 2018, 45(6):80-87.
[20]	LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection [C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Hawaii, USA: IEEE, 2017: 936-944.

参数	取值
太阳轮齿数	24
行星轮齿数	24
内齿圈齿数	72
行星轮系模数/mm	0.6
压力角/(°)	20
轴承滚子数	8

行星齿轮箱状态	训练集数量	测试集数量	标签
正常	3200	800	1
滚动体缺失	3200	800	2
齿面磨损	3200	800	3
行星齿轮与断齿	3200	800	4
行星齿轮与齿面磨损	3200	800	5

编号	网络层		结构参数	算法参数			输出			补零
编号	网络层		结构参数	核数量/个	步长	循环/次	实际		语义	补零
1	复合层1	卷积层	2×2×1@32	—	2×2	—	24×	128	22×126	是
		LSTM	—	3072	—	32	24×	128	24×128	否
	池化层		2×2	—	1×1	—	24×	128	22×126	是
2	池化层		2×2	—	2×2	—	12×	64	11×63	否
3	复合层2	卷积层	2×2×32@32	—	2×2	—	12×	64	10×62	是
		LSTM	—	768	—	32	12×	64	11×63	否
		池化层	2×2		1×1		12×	64	10×62	是
4	池化层		2×2	—	2×2	—	6×	32	10×62	否
5	全连接层		—	1024	—	—	1024×	1	—	否
6	分类器		—	4	—	—	4×	1	—	否

组号	数据	ε/%
组号	数据	K=1	K=2	K=3	K=4	K=5
1	训练	99.78	99.69	99.44	98.71	99.37
	测试	98.53	99.01	97.79	99.57	98.62
2	训练	100	99.47	99.79	99.19	99.01
	测试	98.93	98.97	98.88	99.65	98.98
3	训练	99.43	99.72	98.83	99.55	99.57
	测试	99.59	97.45	100	99.30	99.21

轴承状态	训练集数量	测试集数量	标签
正常	800	200	6
轴承内圈故障	800	200	7
滚动体故障	800	200	8
轴承外圈中心方向	800	200	9
轴承外圈正交方向	800	200	10