基于小波变换和改进残差神经网络的故障诊断方法

doi:10.12305/j.issn.1001-506X.2026.06.23

系统工程与电子技术 ›› 2026, Vol. 48 ›› Issue (6): 2031-2041.doi: 10.12305/j.issn.1001-506X.2026.06.23

基于小波变换和改进残差神经网络的故障诊断方法

袁胜智¹(), 王少蕾¹, 李静¹, 肖任楷¹, 谭浩天², 赖建伟¹^,*, 许江涛²

1. 海军工程大学兵器工程学院，湖北武汉 430033
2. 哈尔滨工程大学航天与建筑工程学院，黑龙江哈尔滨 150001

收稿日期:2025-01-22 修回日期:2025-06-11 出版日期:2026-06-25 发布日期:2026-03-16
通讯作者: 赖建伟 E-mail:yuanszhi_hjgcdx@sina.com
作者简介:袁胜智（1977—），男，副教授，博士，主要研究方向为自动测试与故障诊断
王少蕾（1981—），男，讲师，博士，主要研究方向为发射工程及装备保障
李　静（1982—），男，讲师，博士，主要研究方向为飞行器导航制导与控制、智能非线性控制理论
肖任楷（2001—），男，硕士研究生，主要研究方向为智能算法
谭浩天（2000—），男，博士研究生，主要研究方向为飞行器智能控制算法
许江涛（1975—），男，教授，主要研究方向为飞行器动力学、制导与控制

Fault diagnosis method based on wavelet packet transform and improved residual neural network

Shengzhi YUAN¹(), Shaolei WANG¹, Jing LI¹, Renkai XIAO¹, Haotian TAN², Jianwei LAI¹^,*, Jiangtao XU²

1. College of Weaponry Engineering，Naval University of Engineering，Wuhan 430033，China
2. College of Aerospace and Civil Engineering，Harbin Engineering University，Harbin 150001，China

Received:2025-01-22 Revised:2025-06-11 Online:2026-06-25 Published:2026-03-16
Contact: Jianwei LAI E-mail:yuanszhi_hjgcdx@sina.com

摘要/Abstract

摘要：

传统深度神经网络在处理大批量、多类别、多程度故障样本数据时，需增加网络层数来提升表征能力，从而提高诊断精度，但网络层数的增加易引起梯度消失，导致识别准确率提升困难。结合残差神经网络与连续小波变换方法，并使用线性指数单元改进残差神经网络，提高网络在处理大批量故障样本数据时的识别精度上限。结果表明，使用线性指数单元的激活函数能够有效改善残差神经网络的训练稳定性与预测精度，在具有42种不同类别与程度的多传感器机电系统故障数据集中，所提方法相比未使用残差层和线性指数单元的同规模神经网络识别正确率至少提高13%。因此，所提方法通过融合连续小波变换与线性指数单元改进的残差神经网络，可显著抑制梯度消失问题，有效突破传统深度神经网络的识别精度瓶颈，为复杂机电系统多故障智能诊断提供了一种高可靠性的解决方案。

关键词: 深度学习, 故障诊断, 连续小波变换, 残差神经网络

Abstract:

Traditional deep neural networks need to increase the number of network layers to enhance the characterization ability and improve the network fault diagnosis accuracy when dealing with large-volume, multi-category, multi-degree fault sample data, but the increase in network depth can easily lead to problems such as gradient disappearance, which results in a low upper limit of recognition accuracy. A method that combines the wavelet packet transform with the residual neural network is presented to enhance the recognition accuracy of large-volume motor fault sample data. The residual neural network is improved using an exponential linear unit. The results demonstrate that employing exponential linear unit significantly enhances residual neural network training stability and prediction accuracy. On a 42-types multi-sensor electromechanical fault dataset, the proposed method achieves at least 13% higher recognition accuracy than comparable networks without residual layers and exponential linear unit. Therefore, the proposed method, which integrates the continuous wavelet transform with a residual neural network enhanced by linear exponential units, significantly mitigates the vanishing gradient problem and effectively overcomes the accuracy limitations of traditional deep neural networks, thereby providing a highly reliable solution for the intelligent diagnosis of multiple faults in complex electromechanical systems.

Key words: deep learning, fault diagnosis, wavelet packet transform, residual neural network

中图分类号:

TP 183

袁胜智, 王少蕾, 李静, 肖任楷, 谭浩天, 赖建伟, 许江涛. 基于小波变换和改进残差神经网络的故障诊断方法[J]. 系统工程与电子技术, 2026, 48(6): 2031-2041.

Shengzhi YUAN, Shaolei WANG, Jing LI, Renkai XIAO, Haotian TAN, Jianwei LAI, Jiangtao XU. Fault diagnosis method based on wavelet packet transform and improved residual neural network[J]. Systems Engineering and Electronics, 2026, 48(6): 2031-2041.

图/表 15

图1

表1

图2

图3

图4

表2

图5

表3

表4

图6

表5

图7

图8

图9

表6

参考文献 31

1	LIU H L, DONG J C, WANG T Y, et al. The digital manufacturing equipment and development of high speed and high precision with monitoring and intelligent maintenance[J]. Key Engineering Materials, 2016, 693, 1948- 1953. doi: 10.4028/www.scientific.net/KEM.693.1948
2	PARMAR U, PANDYA D H. Experimental investigation of cylindrical bearing fault diagnosis with SVM[J]. Materials Today: Proceedings, 2021, 44, 1286- 1290.
3	GOYAL D, CHOUDHARY A, PABLA B S, et al. Support vector machines based non-contact fault diagnosis system for bearings[J]. Journal of Intelligent Manufacturing, 2020, 31, 1275- 1289. doi: 10.1007/s10845-019-01511-x
4	钟张豪, 丑永新, 侯千红. 基于随机森林的电机异音故障诊断方法[J]. 盐城工学院学报: 自然科学版, 2023, 36 (2): 37- 43.
	ZHONG Z H, CHOU Y X, HOU Q H. Fault diagnosis method of motor abnormal sound based on random forest[J]. Journal of Yancheng Institute of Technology （Natural Science Edition）, 2023, 36 (2): 37- 43.
5	李兵, 韩睿, 何怡刚, 等. 改进随机森林算法在电机轴承故障诊断中的应用[J]. 中国电机工程学报, 2020, 4, 1310- 1319.
	LI B, HAN R, HE Y G, et al. Applications of the improved random forest algorithm in fault diagnosis of motor bearings[J]. Proceedings of the CSEE, 2020, 4, 1310- 1319.
6	袁建虎, 韩涛, 唐建, 等. 基于小波时频图和CNN的滚动轴承智能故障诊断方法[J]. 机械设计与研究, 2017, 33 (2): 93- 97.
	YUAN J H, HAN T, TANG J, et al. An approach to intelligent fault diagnosis of rolling bearing using wavelet time-frequency representations and CNN[J]. Machine Design and Research, 2017, 33 (2): 93- 97.
7	HUANG X D, PANG X W. Review of intelligent device fault diagnosis based on deep learning[J]. Computer Science, 2023, 50 (5): 93- 102.
8	HUANG T, ZHANG Q, TANG X A, et al. A novel fault diagnosis method based on CNN and LSTM and its application in fault diagnosis for complex systems[J]. Artificial Intelligence Review, 2022, 55, 1289- 1315. doi: 10.1007/s10462-021-09993-z
9	MENH W, ZHAO P, SHI Y, et al. Intelligent fault diagnosis of mechanical engineering using NLF-LSTM optimized deep learning model[J]. Optimization and Engineering, 2024, 28, 1- 22.
10	KHAN T, ALEKHYA P, SESHADRINATH J. Incipient inter-turn fault diagnosis in induction motors using CNN and LSTM based methods[C]//Proc. of the IEEE Industry Applications Society Annual Meeting, 2018.
11	AZAMFAR M, SINGH J, BRAVO-IMAZ I, et al. Multisensor data fusion for gearbox fault diagnosis using 2-D convolutional neural network and motor current signature analysis[J]. Mechanical Systems and Signal Processing, 2020, 144, 106861. doi: 10.1016/j.ymssp.2020.106861
12	PARK C H, KIM H, LEE J, et al. A feature inherited hierarchical convolutional neural network （FI-HCNN） for motor fault severity estimation using stator current signals[J]. Korean Society for Precision Engineering, 2021, 8, 1253- 1266.
13	LV J H. Research on mechanical fault diagnosis and prediction technology based on deep learning[J]. Transactions on Computer Science and Intelligent Systems Research, 2024, 4, 112- 117. doi: 10.62051/vbkk3b17
14	YAN S, SHAO H D, WANG J, et al. LiConv Former: a lightweight fault diagnosis framework using separable multiscale convolution and broadcast self-attention[J]. Expert Systems with Applications, 2023, 237, 121338.
15	WANG G, LU M F. Multiscale deep subspace clustering network with hierarchical fusion mechanism for mechanical fault diagnosis[J]. IEEE Trans. on Instrumentation and Measurement, 2024, 73, 1- 15.
16	YEMI H N. Deep learning architectures enabling sophisticated feature extraction and representation for complex data analysis[J]. International Journal of Innovative Science and Research Technology, 2024, 9 (10): 2290- 2300.
17	LIU C. Research on image classification leveraging deep convolutional neural networks and visual cognition[J]. Applied and Computational Engineering, 2024, 32 (1): 200- 209. doi: 10.54254/2755-2721/32/20230212
18	HU Y H, HUBER A, ANUMULA J, et al. Overcoming the vanishing gradient problem in plain recurrent networks[EB/OL]. https://arxiv.org/abs/1801.06105.
19	WANG S B, ZHANG X D, LIU D R, et al. Unified batch normalization: identifying and alleviating the feature condensation in batch normalization and a unified framework [EB/OL]. [2024−12−22]. https://arxiv.org/abs/2311.15993.
20	ZHONG J, CHEN H Y, CHAO W L. Making batch normalization great in federated deep learning[EB/OL]. [2024−12−22]. https://arxiv.org/abs/2303.06530.
21	HE K M, ZHANG X Y, REN S, et al. Deep residual learning for image recognition[C]// Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770−778.
22	刘飞, 陈仁文, 邢凯玲, 等. 基于迁移学习与深度残差网络的滚动轴承快速故障诊断算法[J]. 振动与冲击, 2022, 41 (3): 154- 164.
	LIU F, CHEN R W, XING K L, et al. Fast fault diagnosis algorithm for rolling bearing based on transfer learning and deep residual network[J]. Journal of Vibration and Shock, 2022, 41 (3): 154- 164.
23	YAN X F, CHEN S C, JIN S M, et al. Research on fault identification method of high-voltage circuit breaker based on wavelet packet dispersion entropy feature extraction[J]. Journal of Physics: Conference Series, 2025, 2935 (1): 012022. doi: 10.1088/1742-6596/2935/1/012022
24	LIU R Z, WANG X B, KUMAR A, et al. WPD-enhanced deep graph contrastive learning data fusion for fault diagnosis of rolling bearing[J]. Micromachines, 2023, 14 (7): 1467. doi: 10.3390/mi14071467
25	ZHOU X, TANG X Z, LIANG W H. A novel analog circuit fault diagnosis method based on multi-channel 1D-resnet and wavelet packet transform[J]. Analog Integrated Circuits and Signal Processing, 2024, 121 (1): 25- 38.
26	WU G G, JI X R, YANG G L, et al. Signal-to-image: rolling bearing fault diagnosis using ResNet family deep-learning models[J]. Processes, 2023, 11 (5): 1527. doi: 10.3390/pr11051527
27	WANG S H, SATAPATHY S C, XIE M X, et al. Retracted article: ELUCNN for explainable COVID-19 diagnosis[J]. Soft Computing, 2024, 28 (S2): 455. doi: 10.1007/s00500-023-07813-w
28	FENG H S, YANG C H. PolyLU: a simple and robust polynomial-based linear unit activation function for deep learning[J]. IEEE Access, 2023, 11, 101347- 101358. doi: 10.1109/ACCESS.2023.3315308
29	YE Q, LIU C H. An unsupervised deep feature learning model based on parallel convolutional autoencoder for intelligent fault diagnosis of main reducer[J]. Computational Intelligence and Neuroscience, 2021, 8922656.
30	张洪亮, 余其源, 秦超群, 等. 基于信息融合及双连接注意力残差网络的轴承故障诊断[J]. 振动与冲击, 2023, 42 (20): 114- 123.
	ZHANG H L, YU Q Y, QIN C Q, et al. Bearing fault diagnosis based on double connected attention residual network and information fusion[J]. Journal of Vibration and Shock, 2023, 42 (20): 114- 123.
31	ZHANG K, ZUO W M, CHEN Y J, et al. Beyond a Gaussian denoiser: residual learning of deep CNN for image denoising[J]. IEEE Trans. on Image Processing, 2017, 26 (7): 3142- 3155. doi: 10.1109/TIP.2017.2662206

层类型	具体参数
输入层	输入数据二维图
卷积层1（Conv1）	卷积核大小7×7，步长2
最大池化层（Maxpool）	池化窗口大小3×3，步长2
卷积层2（Conv2）	$ \left[\begin{array}{l}3 \times 3,64 \\3 \times 3,64\end{array}\right] \times 2 $，步长2
卷积层3（Conv3）	$ \left[\begin{array}{l}3 \times 3,128 \\3 \times 3,128\end{array}\right] \times 2 $，步长2
卷积层4（Conv4）	$ \left[\begin{array}{l}3 \times 3,256 \\3 \times 3,256\end{array}\right] \times 2 $，步长2
卷积层5（Conv5）	$ \left[\begin{array}{l}3 \times 3,512 \\3 \times 3,512\end{array}\right] \times 2 $，步长2
全连接层（Fc）	输入维度512，输出维度42

信号状况	样本集数量	标签
正常数据	2900	0
不平衡故障	2900	1
水平平行错位	2900	2
垂直平行错位	2900	3
内轨道故障	2900	4
外轨道故障	2900	5

项目	训练耗时	检测耗时
CNN网络训练	4376	8.55
ResNet网络训练	5408	10.62
Transformer网络训练	58101	116.22
ConvNext网络训练	57665	115.33

项目	计算占用/GB
CNN网络训练	14.2
ResNet网络训练	14.6
Transformer网络训练	14.1
ConvNext网络训练	10.9

训练样本数据量	小波降噪+ MLP	小波降噪+ CNN	小波频域分解降噪+CNN	小波频域分解降噪+ResNet
原数据大小	12.07	58.85	77.85	92.95
4倍降采样	15.09	63.26	75.33	92.32
8倍降采样	16.54	60.50	73.25	92.25

基于小波变换和改进残差神经网络的故障诊断方法

Fault diagnosis method based on wavelet packet transform and improved residual neural network

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 31

相关文章 15

编辑推荐

Metrics

本文评价

层名称	ReLU激活概率	ELU激活概率
Conv1	0.522388	1
Conv2-Block1	0.872158	1
Conv2-Block2	0.712385	1
Conv3-Block1	0.730315	1
Conv3-Block2	0.503255	1
Conv4-Block1	0.657838	1
Conv4-Block2	0.50858	1
Conv5-Block1	0.657683	1
Conv5-Block2	0.494633	1

[1]	陈垣光, 刘晓凯, 肖思瑶, 张顺生, 崔红元, 陈晓莹, 刘莹. 低信噪比下面向时空特征融合的雷达目标智能检测方法[J]. 系统工程与电子技术, 2026, 48(5): 1523-1531.
[2]	杨泽莹, 陈卓, 徐志明, 艾小锋, 吴其华, 王玲. 空间目标ISAR图像部件正向标记方法[J]. 系统工程与电子技术, 2026, 48(4): 1165-1173.
[3]	林炅, 高朋, 洪文兴. 物理信息神经网络在空天领域的应用与展望[J]. 系统工程与电子技术, 2026, 48(4): 1264-1282.
[4]	程鲲, 雷洪涛, 吕志轩. 改进YOLOv8的轻量级无人机跟踪方法[J]. 系统工程与电子技术, 2026, 48(3): 737-750.
[5]	张龙, 朱连宏, 杨波, 雷震, 冯轩铭. MFA-Net：一种面向复杂对抗环境的反舰导弹智能识别多模态自适应融合网络[J]. 系统工程与电子技术, 2026, 48(3): 894-907.
[6]	田弘宇, 葛松虎, 郭宇, 崔中普, 梁潇. 低干噪比下多特征融合的通信干扰识别方法[J]. 系统工程与电子技术, 2026, 48(3): 1061-1071.
[7]	包玉刚, 贾皓翔, 赵旦峰. 基于高阶递归网络的单幅图像去雨滴模型[J]. 系统工程与电子技术, 2026, 48(1): 12-21.
[8]	陈海青, 汪刘应, 刘顾, 王龙, 葛超群, 陈孟州. 合成孔径雷达图像去噪算法研究进展[J]. 系统工程与电子技术, 2026, 48(1): 94-105.
[9]	李开明, 张袁鹏, 罗迎, 代肖楠. 弹道导弹雷达目标识别研究进展[J]. 系统工程与电子技术, 2025, 47(9): 2870-2889.
[10]	季策, 马相宇, 牟晓宇, 赵家毅. TS-GRU-VTA：基于深度学习的车辆信道估计方案[J]. 系统工程与电子技术, 2025, 47(9): 3093-3098.
[11]	闫小伟, 凌冲, 石胜斌. 地表未爆子弹药快速检测系统设计与实现[J]. 系统工程与电子技术, 2025, 47(8): 2639-2645.
[12]	张琬滢, 高由兵, 李泽一, 李鹏飞, 张伟. 基于双路径特征融合网络的背景信号抑制算法[J]. 系统工程与电子技术, 2025, 47(7): 2406-2413.
[13]	张新征, 闫梦可, 朱晓林. 噪声伪标签容忍的半监督SAR目标识别[J]. 系统工程与电子技术, 2025, 47(6): 1796-1805.
[14]	陈肖楠, 王奕首, 卿新林. 基于混合驱动的航空发动机气路故障诊断技术综述[J]. 系统工程与电子技术, 2025, 47(6): 1880-1892.
[15]	姜智杰, 宋恒, 胡楠, 段兰茜, 曹平. 隧道环境毫米波雷达目标识别与分类算法[J]. 系统工程与电子技术, 2025, 47(5): 1453-1460.