基于扩散模型和DenseNet的少样本齿轮箱故障诊断方法

doi:10.13229/j.cnki.jdxbgxb.20241268

吉林大学学报(工学版) ›› 2026, Vol. 56 ›› Issue (7): 1787-1797.doi: 10.13229/j.cnki.jdxbgxb.20241268

• 车辆工程·机械工程 • 上一篇

基于扩散模型和DenseNet的少样本齿轮箱故障诊断方法

曹洁^1,²(),陈志峰¹,王进花^1,³(),陈莉²

^1.兰州理工大学电气工程与信息工程学院，兰州 730050
^2.兰州城市学院信息工程学院，兰州 730070
^3.甘肃省制造信息工程研究中心，兰州 730050

收稿日期:2024-11-12 出版日期:2026-07-01 发布日期:2026-08-12
通讯作者: 王进花 E-mail:caoj@lut.edu.cn;wjh0615@lut.edu.cn
作者简介:曹洁（1966-），女，教授，博士. 研究方向：智能信息处理，机器视觉信息获取与处理.E-mail： caoj@lut.edu.cn
基金资助:
国家自然科学基金项目(62063020);国家自然科学基金项目(61763028);国家重点研发计划项目(2020YFB1713600);甘肃省自然科学基金项目(20JR5RA463);甘肃省重点研发计划项目(22YF7GA130)

Fault diagnosis method for gearbox with few samples based on diffusion model and DenseNet

Jie CAO^1,²(),Zhi-feng CHEN¹,Jin-hua WANG^1,³(),Li CHEN²

^1.College of Electrical and Information Engineering，Lanzhou University of Technology，Lanzhou 730050，China
^2.School of Information Engineering，Lanzhou City University，Lanzhou 730070，China
^3.Gansu Engineering Technology Research Center for Manufacturing Informatization，Lanzhou 730050，China

Received:2024-11-12 Online:2026-07-01 Published:2026-08-12
Contact: Jin-hua WANG E-mail:caoj@lut.edu.cn;wjh0615@lut.edu.cn

摘要/Abstract

摘要：

针对旋转机械故障诊断中的少样本问题，提出了基于一种扩散模型的行星齿轮箱故障诊断方法，通过将故障样本的时频图作为故障特征表示来进行故障分类。该方法首先使用连续小波变换将齿轮箱一维振动信号转换为时频图作为训练样本；其次，对扩散模型进行了优化改进，使其充分学习训练样本故障特征，以生成高质量故障样本；最后，构建了一种基于密集卷积网络的分类器，结合注意力机制增强分类性能，使用包含多种故障类型的测试集对分类器进行性能测试。结果表明：本文方法诊断准确率与充足样本条件下接近，准确率达到99%以上，并与现有方法进行了实验对比。另外，对小波变换方法及分类器的优化设置消融实验验证了其有效性。

关键词: 扩散模型, 故障诊断, 齿轮箱, 少样本, 注意力机制

Abstract:

Aiming at the problem of few samples in the fault diagnosis of rotating machinery， a fault diagnosis method for planetary gearboxes based on a diffusion model is proposed. The time-frequency diagrams of fault samples are used as the representation of fault features for fault classification. Firstly， the continuous wavelet transform is used to convert the one-dimensional vibration signals of the gearbox into time-frequency diagrams as training samples. Secondly， the diffusion model is optimized and improved to enable it to fully learn the fault features of the training samples and generate high-quality fault samples. Finally， a classifier based on a dense convolutional network is constructed， which combines the attention mechanism to enhance the performance of the classifier. The performance of the classifier is tested using a test set containing multiple fault types. The results show that the diagnostic accuracy of the proposed method is close to that under the condition of sufficient samples， reaching over 99%， and experimental comparisons with existing methods have been carried out. In addition， ablation experiments on the optimization settings of the wavelet transform method and the classifier are conducted to verify their effectiveness.

Key words: diffusion model, fault diagnosis, gears, small sample, attention mechanism

中图分类号:

TH132.425

曹洁,陈志峰,王进花,陈莉. 基于扩散模型和DenseNet的少样本齿轮箱故障诊断方法[J]. 吉林大学学报(工学版), 2026, 56(7): 1787-1797.

Jie CAO,Zhi-feng CHEN,Jin-hua WANG,Li CHEN. Fault diagnosis method for gearbox with few samples based on diffusion model and DenseNet[J]. Journal of Jilin University(Engineering and Technology Edition), 2026, 56(7): 1787-1797.

图/表 21

图1

图2

图3

表1

网络模型参数"

模型	输出尺寸	网络结构
卷积层	$64 × 64$	$7 × 7 c o n v, 步长 2$
池化层	$32 × 32$	$3 × 3 最大池化, 步长 2$
EMA分支1	$32 × 1$	$沿高度 H 维度平均池化, 保留宽度为 1$
	$1 × 32$	$沿宽度 W 维度平均池化, 保留高度为 1$
	$32 × 32$	$1 × 1 c o n v, 通道数 64, 分组 16, 每组 4 通道$
	$32 × 32$	$组归一化, 每组 4 通道$
EMA分支2	$32 × 32$	$3 × 3 c o n v, 通道数保持 64$
融合分支1、2	$32 × 32$	—
密集层1	$32 × 32$	$1 × 1 3 × 3 × 6$
过渡层1	$32 × 32$	$1 × 1 c o n v$
过渡层1	$16 × 16$	$2 × 2 平均池化, 步长 2$
密集层2	$16 × 16$	$1 × 1 3 × 3 × 12$
过渡层2	$16 × 16$	$1 × 1 c o n v$
过渡层2	$8 × 8$	$2 × 2 平均池化, 步长 2$
密集层3	$8 × 8$	$1 × 1 3 × 3 × 48$
过渡层3	$8 × 8$	$1 × 1 c o n v$
过渡层3	$4 × 4$	$2 × 2 平均池化, 步长 2$
密集层4	$4 × 4$	$1 × 1 3 × 3 × 32$
分类层	$1 × 1$	$7 × 7 全局平均池化$
分类层	$5$	$S o f t m a x$

表1

图4

图5

图6

表2

行星齿轮箱参数"

参数	数值
太阳轮	28
环形齿轮	100
行星齿轮齿数（数量）	36（4）
啮合频率	（175/8） $f r$
太阳轮故障频率	（25/8） $f r$

表2

表3

图7

图8

表4

表5

图9

表6

图10

表7

表8

图11

图12

图13

参考文献 27

[1]	Peng H, Zhang H, Fan Y S, et al. A review of research on wind turbine bearings' failure analysis and fault diagnosis[J]. Lubricants, 2023, 11(1): No.14.
[2]	Su Y, Meng L, Kong X, et al. Small sample fault diagnosis method for wind turbine gearbox based on optimized generative adversarial networks[J]. Engineering Failure Analysis, 2022, 140: No.106573.
[3]	Shen C, Wang J, Chen J, et al. Gearbox fault diagnosis for wind turbines based on data augmentation using improved generative adversarial networks[J]. Proceedings of the International Conference on Electrical Materials and Power Equipment(ICEMPE), Chongqing, China, 2021.
[4]	张玺君, 尚继洋, 余光杰, 等. 基于注意力的多尺度卷积神经网络轴承故障诊断[J]. 吉林大学学报:工学版, 2024, 54(10): 3009-3017.
	Zhang Xi-jun, Shang Ji-yang, Yu Guang-jie, et al. Bearing fault diagnosis based on attention for multi-scale convolutional neural network[J]. Journal of Jilin University(Engineering and Technology Edition), 2024, 54(10): 3009-3017.
[5]	Misbah I, Lee C K M, Keung K L. Fault diagnosis in rotating machines based on transfer learning: literature review[J]. Knowledge-Based Systems, 2024, 283: No.111158.
[6]	邵海东, 李伟, 刘翊, 等. 时变转速下基于双阈值注意力生成对抗网络和小样本的转子-轴承系统故障诊断[J]. 机械工程学报, 2023, 59(12): 215-224.
	Shao Hai-dong, Li Wei, Liu Yi, et al. Fault diagnosis of rotor-bearing system under time-varying speeds by using dual-threshold attention-embedded GAN and small samples[J]. Journal of Mechanical Engineering, 2023, 59(12): 215-224.
[7]	Xie S, Cheng W, Nie Z, et al. Intelligent fault diagnosis of bearings under variable working conditions and small samples with generative adversarial network[C]∥Prognostics and Health Management Conference, London, United Kingdom, 2022: 162-168.
[8]	Liang P, Deng C, Yuan X, et al. A deep capsule neural network with data augmentation generative adversarial networks for single and simultaneous fault diagnosis of wind turbine gearbox[J]. ISA Transactions, 2023, 135: 462-475.
[9]	Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets[C]∥Advances in Neural Information Processing Systems, Montreal,Canda, 2014.
[10]	Ho J, Jain A, Abbeel P. Denoising diffusion probabilistic models[C]∥Advances in Neural Information Processing Systems, Online, 2020: 6840-6851.
[11]	Song J, Meng C, Ermon S. Denoising diffusion implicit models[J/OL].[2020-12-20]. .
[12]	Chen P, Xu C, Ma Z, et al. A mixed samples-driven methodology based on denoising diffusion probabilistic model for identifying damage in carbon fiber composite structures[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-11.
[13]	Huang T, Gao Y, Li Z, et al. A hybrid deep learning framework based on diffusion model and deep residual neural network for defect detection in composite plates[J]. Applied Sciences, 2023, 13(10): No.5843.
[14]	Yang X, Ye T, Yuan X, et al. A novel data augmentation method based on denoising diffusion probabilistic model for fault diagnosis under imbalanced data[J]. IEEE Transactions on Industrial Informatics, 2024,20(5): 7820-7831.
[15]	Croitoru F A, Hondru V, Ionescu R T, et al. Diffusion models in vision: a survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(9): 10850-10869.
[16]	刘子昌, 白永生, 李思雨, 等. 基于小波时频图与Swin Transformer的柴油机故障诊断方法[J]. 系统工程与电子技术, 2023, 45(9): 2986-2998.
	Liu Zi-chang, Bai Yong-sheng, Li Si-yu, et al. Diesel engine fault diagnosis method based on wavelet time-frequency diagram and Swin Transformer[J]. Systems Engineering and Electronics, 2023, 45(9): 2986-2998.
[17]	Ye C, Wang J, Peng C, et al. Novel cross-domain fault diagnosis method based on model-agnostic meta-learning embedded in adaptive threshold network[J]. Measurement, 2023, 222: No.113677.
[18]	Wang W, Bao J, Zhou W, et al. SinDiffusion: learning a diffusion model from a single natural image[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025,47(5):3412-3423.
[19]	Lv H, Chen J, Pan T, et al. Attention mechanism in intelligent fault diagnosis of machinery: a review of technique and application[J]. Measurement, 2022, 199: No.111594.
[20]	Ouyang D, He S, Zhang G, et al. Efficient multi-scale attention module with cross-spatial learning[C]∥IEEE International Conference on Acoustics, Speech and Signal Processing(ICASSP), Rhodes Island, Greece, 2023: 1-5.
[21]	Huang G, Liu Z, Van Der Maaten L, et al. Densely connected convolutional networks[C]∥IEEE Conference on Computer Vision and Pattern Recognition, Hawaii State, America, 2017: 4700-4708.
[22]	徐浩, 沙刚刚, 李腾腾, 等. 基于二维连续小波变换与数据融合技术的逆有限元法-伪激励法结构损伤识别方法[J]. 复合材料学报, 2021, 38(10): 3554-3562.
	Xu Hao, Sha Gang-gang, Li Teng-teng, et al. Structure damage identification method of inverse finite element method-pseudo-excitation method based on 2D continuous wavelet transform and data fusion technology[J]. Acta Materiae Compositae Sinica, 2021, 38(10) :3554-3562.
[23]	陈佛计, 朱枫, 吴清潇, 等. 生成对抗网络及其在图像生成中的应用研究综述[J]. 计算机学报, 2021, 44(2): 347-369.
	Chen Fu-ji, Zhu Feng, Wu Qing-xiao, et al. A survey about image generation with generative adversarial nets[J]. Chinese Journal of Computers, 2021, 44(2): 347-369.
[24]	Ibrahem H, Salem A, Kang H S. Exploration of semantic label decomposition and dataset size in semantic indoor scenes synthesis via optimized residual generative adversarial networks[J]. Sensors, 2022,22(21): No.8306.
[25]	Liu D, Cui L, Cheng W. A review on deep learning in planetary gearbox health state recognition: Methods, applications, and dataset publication[J]. Measurement Science and Technology, 2023,35(1): No. 012002.
[26]	Wang X, Jiang H, Wu Z, et al. Adaptive variational autoencoding generative adversarial networks for rolling bearing fault diagnosis[J]. Advanced Engineering Informatics, 2023, 56: No.102027.
[27]	Meng Z, He H, Cao W, et al. A novel generation network using feature fusion and guided adversarial learning for fault diagnosis of rotating machinery[J]. Expert Systems with Applications, 2023, 234: No.121058.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

标签	损伤种类	标签	损伤种类
1	齿轮断齿	4	齿根裂纹
2	健康状态	5	齿轮磨损
3	缺齿

标签	FID	LPIPS	MMD
平均值	49.894 7	0.186 1	0.144 8
1	54.775 2	0.264 6	0.200 7
2	47.824 5	0.144 2	0.138 9
3	46.171 4	0.149 7	0.125 9
4	45.745 4	0.124 9	0.113 2
5	54.957 2	0.247 2	0.145 4

方法	FID	MMD
WGAN?GP	150.21	0.214
AVAEGAN^［26］	113.41	0.181
UDWGAN^［27］	70.21	0.148
本文方法	49.89	0.145

实验序号	真实样本数量	生成样本数量	测试样本数量	准确率/%
1	40	0	1 000	98.074
2	40	20	1 000	98.872
3	40	40	1 000	99.240
4	40	60	1 000	99.120
5	40	80	1 000	99.370
6	30	0	1 000	96.748
7	30	20	1 000	97.300
8	30	40	1 000	98.466
9	30	60	1 000	98.108
10	30	80	1 000	99.098
11	20	0	1 000	96.072
12	20	20	1 000	98.216
13	20	40	1 000	96.962
14	20	60	1 000	96.902
15	20	80	1 000	98.228
16	10	0	1 000	89.666
17	10	20	1 000	94.192
18	10	40	1 000	96.404
19	10	60	1 000	96.066
20	10	80	1 000	97.354

标签	精准率	召回率	Specificity
1	0.996	0.999	0.999
2	0.996	0.999	0.999
3	0.998	0.98	1.0
4	0.984	0.996	0.996
5	1.0	1.0	1.0

基于扩散模型和DenseNet的少样本齿轮箱故障诊断方法

Fault diagnosis method for gearbox with few samples based on diffusion model and DenseNet

RICH HTML

PDF (PC)

摘要/Abstract

引用本文

使用本文

图/表 21

参考文献 27

相关文章 15

Metrics

本文评价

推荐阅读 0

模型	批量大小	参数量/Mb
Res Net	32	25.6
ResNeXt	32	25
Dense Net	32	7.98
SE?Dense Net	32	8.2
本文方法	32	8.1

[1]	刘昕晖,陈晫群,吕岩. 基于深度学习的工业故障诊断综述[J]. 吉林大学学报(工学版), 2026, 56(7): 1759-1779.
[2]	缪君,严杰,杜荣华,李磊,储珺. 基于双向特征融合的物体位姿估计方法[J]. 吉林大学学报(工学版), 2026, 56(2): 523-532.
[3]	周求湛,李新萌,沈皓庆子,武慧南,李媛媛,荣静,胡春华,刘萍萍. 基于注意力机制的不平衡数据的非侵入式负荷分解[J]. 吉林大学学报(工学版), 2026, 56(1): 239-246.
[4]	冯志刚,任梦媛,董冰,于明月. 基于多频带特征图和改进SqueezeNet的滚动轴承故障诊断[J]. 吉林大学学报(工学版), 2026, 56(1): 96-108.
[5]	兰巍,周政,王冠宇,王伟,张苗苗. 基于机器学习的汽车设计智能拟合方法[J]. 吉林大学学报(工学版), 2025, 55(9): 2858-2863.
[6]	霍震,金立生,华强,贺阳. 基于边缘特征引导的智能汽车语义分割方法[J]. 吉林大学学报(工学版), 2025, 55(9): 3032-3041.
[7]	庄珊娜,王君帅,白晶,杜京瑾,王正友. 基于三维卷积与自注意力机制的视频行人重识别[J]. 吉林大学学报(工学版), 2025, 55(7): 2409-2417.
[8]	冯志刚,王首起,于明月. 基于变分模态提取及轻量级网络的滚动轴承故障诊断[J]. 吉林大学学报(工学版), 2025, 55(6): 1883-1891.
[9]	薛雅丽,俞潼安,崔闪,周李尊. 基于级联嵌套U-Net的红外小目标检测[J]. 吉林大学学报(工学版), 2025, 55(5): 1714-1721.
[10]	张河山,范梦伟,谭鑫,郑展骥,寇立明,徐进. 基于改进YOLOX的无人机航拍图像密集小目标车辆检测[J]. 吉林大学学报(工学版), 2025, 55(4): 1307-1318.
[11]	才华,王玉瑶,付强,马智勇,王伟刚,张晨洁. 基于注意力机制和特征融合的语义分割网络[J]. 吉林大学学报(工学版), 2025, 55(4): 1384-1395.
[12]	张兰芳,李根泽,刘婷宇,余博. 局部多车影响下跟驰行为机理及建模[J]. 吉林大学学报(工学版), 2025, 55(3): 963-973.
[13]	李扬,李现国,苗长云,徐晟. 基于双分支通道先验和Retinex的低照度图像增强算法[J]. 吉林大学学报(工学版), 2025, 55(3): 1028-1036.
[14]	王娜,崔月磊,李杨,王子从. 基于小波包对数能量图的滚动轴承故障诊断方法[J]. 吉林大学学报(工学版), 2025, 55(2): 494-502.
[15]	刘展鹏,马振超. 考虑隐私信息安全性的极少样本条件指纹图像细节加密算法[J]. 吉林大学学报(工学版), 2025, 55(2): 687-692.