A Multi-fault Diagnosis Method for Rolling Bearings Integrating Two-dimensional Convolutional and GRU

doi:10.3969/j.issn.1004-132X.2025.12.021

Abstract

Abstract:

Aiming at the problems of difficulty in diagnosing and classifying single or compound faults of rolling bearings under complex working conditions，a bearing fault diagnosis method was proposed by fusing two-dimensional convolutional neural network（2D-CNN）and GRU.Firstly，the 2D-CNN layer and GRU layer were used to extract the spatial and temporal features，and the batch normalization（BN） layer was introduced to prevent overfitting.Secondly，the spatial and temporal information features extracted by weight fusion were synthesized，and then the global average pooling layer was used instead of the flatten layer.Finally，the covariance matrix and t-SNE algorithm were used to visualize and analyze the model training processes and output the results by activation function Softmax classification.The model was verified by prognostics and health management（PHM） dataset and XJTU-SY dataset，and compared with other models，the good accuracy and generalization of the model were shown.

Key words: rolling bearing, convolutional neural network, gated recurrent neural network（GRU）, fault diagnosis and classification, visual analytics

摘要：

针对滚动轴承在复杂工况下的单一或复合故障诊断与分类困难的问题，提出了一种二维卷积神经网络（2D-CNN）与门控循环神经网络（GRU）融合的轴承故障诊断方法。首先采用2D-CNN层和GRU层提取空间特征与时序特征，并引入批量标准化（BN）层防止过拟合，然后通过权重融合综合二者提取的空间与时序信息特征，再使用全局平均池化层代替展平层，最后用协方差矩阵和t-SNE算法对模型训练过程进行可视化分析并通过激活函数Softmax分类输出结果。通过故障预测与健康管理（PHM）数据集和XJTU-SY数据集对模型进行验证，并和其他模型作对比，结果显示了所提模型良好的准确率和泛化性。

关键词: 滚动轴承, 卷积神经网络, 门控循环神经网络, 故障诊断与分类, 可视化分析法

CLC Number:

TH133

Xiong ZHANG, Lecong DONG, Wenqiang WANG, Weiying QU, Shuting WAN. A Multi-fault Diagnosis Method for Rolling Bearings Integrating Two-dimensional Convolutional and GRU[J]. China Mechanical Engineering, 2025, 36(12): 2978-2985.

张雄, 董乐聪, 王文强, 渠伟瀅, 万书亭. 融合二维卷积与门控循环神经网络的滚动轴承多故障诊断方法[J]. 中国机械工程, 2025, 36(12): 2978-2985.

Figures/Tables 27

Fig.1 Traditional CNN structure

Fig.2 GRU unit basic structure

Fig.3 MFCNN-GRU structure

Tab.1 MFCNN-GRU model parameters

模型结构	输出尺寸	参数设置	参数数量
Input	32×32×1
Conv2d_1	32×32×16	3×3×16	144
GRU_1	32×32×16	16	912
BN_1	32×32×16		64
Dropout_1	32×32×16	0.2
conv2d_1	32×32×32	3×3×32	4608
GRU_2	32×32×32	32	4800
BN_2	32×32×32		128
Dropout_2	32×32×32	0.2
Conv2d_2	32×32×64	3×3×64	18 432
GRU_3	32×32×64	64	18 816
BN_3	32×32×64		256
Dropout_3	32×32×64	0.2
Global Average Pooling	64
Softmax	10	10×1	650

Fig.4 Comparison of experimental results of different configurations of the two models

Fig.5 Fault diagnosis process

Fig.6 Subway bogie fault simulation test bench

Tab.2 Operating conditions

电机转频/Hz	横向负载/kN
20	0
	10
	$-$ 10
40	0
	10
	$-$ 10
60	0
	10
	$-$ 10

Tab.2 Operating conditions

电机转频/Hz	横向负载/kN
20	0
	10
	$-$ 10
40	0
	10
	$-$ 10
60	0
	10
	$-$ 10

Tab.3 Data label settings

数据集名称	数据类型	标签编号
M0_G0_LA0_RA0	正常数据	0
M0_G0_LA1_RA0	内圈故障	1
M0_G0_LA1+LA2+LA3+LA4_RA0	内圈、外圈、滚动体、保持架故障	2
M0_G0_LA2_RA0	外圈故障	3
M0_G0_LA3_RA0	滚动体故障	4
M0_G0_LA4_RA0	保持架故障	5
M0_G5_LA0_RA0	内圈故障	6
M0_G6_LA0_RA0	外圈故障	7
M0_G7_LA0_RA0	滚动体故障	8
M0_G8_LA0_RA0	保持架故障	9

Fig.7 Comparison of the accuracy of different feature extraction layers

Tab.4 Comparison of accuracy for different batch sizes

批次大小	时间/s	训练集准确率/%	测试集准确率/%
32	167	100	98.78
64	120	100	98.90
128	115	100	99.99
256	121	100	99.76

Fig.8 Accuracy curve of the MFCNN-GRU model

Fig.9 Loss curves of the MFCNN-GRU model

Fig.10 Test set confusion matrix

Fig.11 The covariance matrix of the output features of the second layer feature extraction layer of the MFCNN-GRU model

Fig.12 The covariance matrix of the output features of the second layer feature extraction layer of the CNN-LSTM model

Fig.13 Process visualization

Fig.14 Comparison of the accuracy of different models

Fig.15 Comparison of the loss of different modelsMFCNN-GRU

Tab.5 The accuracy of each model under different partition ratios

模型名称	划分比例
模型名称	7∶3	6∶4	8∶2
MFCNN-GRU	99.99	99.95	99.97
AlexNet	91.46	90.58	91.50
GhostNet	87.21	85.43	86.77
MoblieNetV2	80.52	84.31	79.95

Fig.16 Comparison of the accuracy of each model with different noises

Tab.6 XJTU-SY bearing dataset case settings

工况序号	转速/（r·min^-1）	径向力/kN
1	2100	12
2	2250	11
3	2400	10

Tab.7 XJTU-SY bearing dataset label settings

工况	数据集名称	故障部位	标签编号
1	Bearing 1_1	正常	0
1	Bearing 1_1	外圈	1
1	Bearing 1_4	保持架	2
1	Bearing 1_5	内圈、外圈	3
2	Bearing 2_1	内圈	4
2	Bearing 2_3	保持架	5
2	Bearing 2_5	外圈	6
3	Bearing 3_1	外圈	7
3	Bearing 3_2	内圈、滚动体、保持架、外圈	8
3	Bearing 3_3	内圈	9

Fig.17 Accuracy curve of XJTU-SY dataset

Fig.18 Loss curve of XJTU-SY dataset

Fig.19 XJTU-SY dataset confusion matrix

Fig.20 Process visualization of XJTU-SY dataset

References 19

[1]	赵志宏，李春秀，窦广鉴，等. 基于MTF-CNN的轴承故障诊断研究［J］. 振动与冲击， 2023， 42（2）：126-131.
	ZHAO Zhihong， LI Chunxiu， DOU Guangjian， et al. Research on Bearing Fault Diagnosis Based on MTF-CNN ［J］. Vibration and Shock， 2023， 42（2）：126-131.
[2]	杨平，苏燕辰，张振. 基于卷积胶囊网络的滚动轴承故障诊断研究［J］. 振动与冲击， 2020， 39（4）：55-62.
	YANG Ping， SU Yanchen， ZHANG Zhen. Research on Fault Diagnosis of Rolling Bearing Based on Convolutional Capsule Network ［J］. Vibration and Shock， 2020， 39（4）：55-62.
[3]	李俊卿，刘静. 结合卷积神经网络和迁移学习的电机轴承故障诊断方法［J］. 华北电力大学学报（自然科学版）， 2023， 50（1）：76-83.
	LI Junqing， LIU Jing. Fault Diagnosis Method of Motor Bearing Based on Convolutional Neural Network and Transfer Learning ［J］. Journal of North China Electric Power University （Natural Science Edition）， 2023， 50（1）：76-83.
[4]	赵靖，杨绍普，李强，等. 一种残差注意力迁移学习方法及其在滚动轴承故障诊断中的应用［J］. 中国机械工程， 2023， 34（3）：332-343.
	ZHAO Jing， YANG Shaopu， LI Qiang， et al. A New Transfer Learning Method with Residual Attention and Its Applications on Rolling Bearing Fault Diagnosis ［J］. China Mechanical Engineering， 2023， 34（3）：332-343.
[5]	陈保家，陈学力，沈保明，等. CNN-LSTM深度神经网络在滚动轴承故障诊断中的应用［J］. 西安交通大学学报， 2021， 55（6）：28-36.
	CHEN Baojia， CHEN Xueli， SHEN Baoming， et al. Application of CNN-LSTM Deep Neural Network in Fault Diagnosis of Rolling Bearings ［J］. Journal of Xi'an Jiaotong University， 2021， 55（6）：28-36.
[6]	邹筱瑜，孙国庆，王忠宾，等. 基于时频融合深度网络的矿用钻机轴承故障诊断［J］. 中国机械工程， 2024， 35（8）：1405-1413.
	ZOU Xiaoyu， SUN Guoqing， WANG Zhongbin， et al. Bearing Fault Diagnosis of Mining Drilling Rig with Time-frequency-fused Deep Network ［J］. China Mechanical Engineering， 2024， 35（8）：1405-1413.
[7]	宫文峰，陈辉，张泽辉，等. 基于改进卷积神经网络的滚动轴承智能故障诊断研究［J］. 振动工程学报， 2020， 33（2）：400-413.
	GONG Wenfeng， CHEN Hui， ZHANG Zehui， et al. Research on Intelligent Fault Diagnosis of Rolling Bearings Based on Improved Convolutional Neural Network ［J］. Journal of Vibration Engineering， 2020， 33（2）：400-413.
[8]	FU Guanghua， WEI Qingjuan， YANG Yongsheng. Bearing Fault Diagnosis with Parallel CNN and LSTM ［J］. Mathematical Biosciences and Engineering， 2024， 21（2）：2385-2406.
[9]	WU Chunming， ZHENG Shupeng. Fault Diagnosis Method of Rolling Bearing Based on MSCNN-LSTM ［J］. Computers， Meterials & Continua， 2024， 79（3）：4395-4411.
[10]	刘玉鑫，武文博，张雄，等. 基于 HHO-CNN 的轴承故障诊断方法研究［J］. 河北大学学报（自然科学版）， 2023， 43（6）：571-583.
	LIU Yuxin， WU Wenbo， ZHANG Xiong， et al. Research on Bearing Fault Diagnosis Method Based on HHO-CNN ［J］. Journal of Hebei University （Natural Science Edition）， 2023， 43（6）：571-583.
[11]	CHEN Guangyi， TANG Gang， ZHU Zhixiao. VKCNN：an Interpretable Variational Kernel Convolutional Neural Network for Rolling Bearing Fault Diagnosis ［J］. Advanced Engineering Informatics， 2024， 62（B）：102705.
[12]	LIU Fang， LIANG Chen， GUO Zhihao， et al. Fault Fiagnosis of Rolling Bearings under Varying Speeds Based on Gray Level Cooccurrence Matrix and DCCNN ［J］. Measurement， 2024， 235：114955.
[13]	CHO K， van MERRIENBOER B， GULCEHRE C， et al. Leaning Phrase Representations Using RNN Encoder-Decoder for Statistical Machine Translation ［J］. ArXiv， 2014， 1406：1724-1734.
[14]	刘建伟，宋志妍. 循环神经网络研究综述［J］. 控制与决策， 2022， 37（11）：2753-2768.
	LIU JIANwei， SONG Zhiyan. A Review of Recurrent Neural Networks ［J］. Control and Decision， 2022， 37（11）：2753-2768.
[15]	唐贵基，周威，王晓龙，等. 基于变维GRU-BiLSTM神经网络模型的滚动轴承寿命预测［J］. 中国工程机械学报， 2022， 20（6）：498-503.
	TANG Guiji， ZHOU Wei， WANG Xiaolong， et al. Life Prediction of Rolling Bearings Based on Variable Dimensionality GRU-BiLSTM Neural Network Model ［J］. Proceedings of the China Society of Engineering Machinery， 2022， 20（6）：498-503.
[16]	朱云贵，郭瑜，陈鑫，等. 优化AR模型的滚动轴承故障IAS信号诊断方法［J］. 振动工程学报， 2024， 37（12）：2141-2147.
	ZHU Yungui， GUO Yu， CHEN Xin， et al. Diagnostic Method for Rolling Bearing Fault IAS Signal Based on Optimized AR Model ［J］. Journal of Vibration Engineering， 2024， 37（12）：2141-2147.
[17]	Yin Shuxin， Chen Zengxu. Research on Compound Fault Diagnosis of Bearings Using an Improved DRSN-GRU Dual-channel Model ［J］. IEEE Sensors Journal， 2024， 24（21）：35304-35311.
[18]	DING Ao， QIN Yong， WANG Biao， et al. Evolvable Graph Neural Network for System-level Incremental Fault Diagnosis of Train Transmission Systems ［J］. Mechanical Systems and Signal Processing， 2024， 210：111-175.
[19]	WANG Biao， LEI Yaguo， LI Naipeng， et al. A Hybrid Prognostics Approach for Estimating Remaining Useful Life of Rolling Element Bearings ［J］. IEEE Transactions on Reliability， 2020， 69（1）：401-412.