Sample augmentation method for fault diagnosis with high-frequency guided diffusion model

doi:10.11772/j.issn.1001-9081.2025060774

Journal of Computer Applications ›› 2026, Vol. 46 ›› Issue (7): 2119-2129.DOI: 10.11772/j.issn.1001-9081.2025060774

• Artificial intelligence • Previous Articles

Sample augmentation method for fault diagnosis with high-frequency guided diffusion model

Xiaohong QIU¹^,², Zixiang YU¹(), Yujie HU¹

^1.School of Industrial Software，Jiangxi University of Science and Technology，Nanchang Jiangxi 330013，China
^2.Nanchang Key Laboratory of Virtual Digital Engineering and Cultural Communication，Jiangxi University of Science and Technology，Nanchang Jiangxi 330013，China

Received:2025-07-15 Revised:2025-09-15 Accepted:2025-09-19 Online:2025-09-28 Published:2026-07-10
Contact: Zixiang YU
About author:QIU Xiaohong， born in 1967， Ph. D.， professor. His research interests include intelligent computing， fault diagnosis.
HU Yujie， born in 2002， M. S. candidate. His research interests include fault diagnosis， artificial intelligence.
Supported by:
Jiangxi Provincial Natural Science Foundation(20242BAB25073)

高频引导的扩散模型故障诊断样本增广方法

邱晓红¹^,², 于子祥¹(), 胡玉杰¹

^1.江西理工大学软件工程学院，南昌 330013
^2.南昌市虚拟数字工程与文化传播重点实验室（江西理工大学），南昌 330013

通讯作者: 于子祥
作者简介:邱晓红（1967—），男，江西南康人，教授，博士，主要研究方向：智能计算、故障诊断
胡玉杰（2002—），男，江西九江人，硕士研究生，主要研究方向：故障诊断、人工智能。
基金资助:
江西省自然科学基金资助项目(20242BAB25073)

Abstract

Abstract:

To address the issue of scarce fault samples in industrial fault diagnosis that limits the performance of deep learning models， as well as the training instability and mode collapse problems in the existing Generative Adversarial Network （GAN） -based data augmentation methods， a sample augmentation method for fault diagnosis with a high-frequency guided diffusion model was proposed. Firstly， the one-dimensional vibration signals were converted into two-dimensional gray-scale spectrograms using Short-Time Fourier Transform （STFT）， and a U-Net structure-based denoising diffusion probabilistic model was constructed. Then， a high-frequency guidance module was designed to enhance the extraction and fusion of high-frequency features via high-pass filtering and cross-attention mechanism， thereby preserving fine-grained texture information in fault signals. Meanwhile， an auxiliary classifier was introduced at each denoising step of the diffusion process to improve intra-class diversity and inter-class distinction of samples via class-conditional constraints. Finally， a multi-objective optimization strategy， combined with noise prediction and classifier guidance， was adopted to stabilize the generation process and enhance sample quality. Experimental results on two public industrial fault datasets showed that the classification model trained on samples generated by the proposed method achieved maximum diagnostic accuracies of 97.41% and 96.66%， respectively. Compared with the sub-optimal method on the two datasets， WGAN-GP （Wasserstein Generative Adversarial Network with Gradient Penalty） and Conditional Denoising Diffusion Probabilistic Model （CDDPM）， the improvements were 1.92 and 2.16 percentage points， respectively. The proposed method enhances fault diagnosis performance under few-shot conditions significantly and provides a stable and reliable solution to the class imbalance problem in industrial scenarios.

Key words: diffusion model, high-frequency guidance, auxiliary classifier, fault diagnosis, few-shot problem

摘要：

针对工业故障诊断中故障样本稀缺导致深度学习模型性能受限，以及现有的基于生成对抗网络（GAN）的数据增强方法存在训练不稳定和模式崩溃的问题，提出一种高频引导的扩散模型故障诊断样本增广方法。首先，将一维振动信号通过短时傅里叶变换（STFT）转换为二维灰度谱图，并构建基于U-Net结构的去噪扩散概率模型；其次，设计高频引导模块，通过高通滤波和交叉注意力机制增强高频特征的提取与融合，以保留故障信号中的细粒度纹理信息；同时，在扩散过程的每个去噪步骤中引入辅助分类器，以通过类别条件约束增强样本的类内多样性和类间区分性；最后，采用噪声预测和分类器引导相结合的多目标优化策略，以稳定生成过程并提升样本质量。在2个公开工业故障数据集上的实验结果表明，本文方法生成的样本用于训练分类模型后的诊断准确率最高分别达到97.41%和96.66%，相较于2个数据集上的次优方法WGAN-GP （Wasserstein Generative Adversarial Network with Gradient Penalty）和条件去噪扩散概率模型（CDDPM）分别提升了1.92和2.16个百分点。本文方法有效提高了小样本条件下的故障诊断性能，可为工业场景中的样本不平衡问题提供稳定可靠的解决方案。

关键词: 扩散模型, 高频引导, 辅助分类器, 故障诊断, 小样本问题

CLC Number:

TP277

Xiaohong QIU, Zixiang YU, Yujie HU. Sample augmentation method for fault diagnosis with high-frequency guided diffusion model[J]. Journal of Computer Applications, 2026, 46(7): 2119-2129.

邱晓红, 于子祥, 胡玉杰. 高频引导的扩散模型故障诊断样本增广方法[J]. 《计算机应用》唯一官方网站, 2026, 46(7): 2119-2129.

Figures/Tables 23

References 35

[1]	Ni Q， Ji J C， Halkon B， et al. Physics-Informed Residual Network （PIResNet） for rolling element bearing fault diagnostics ［J］. Mechanical Systems and Signal Processing， 2023， 200： No.110544.
[2]	Xu Y， Li S， Yan X， et al. Multiattention-based feature aggregation convolutional networks with dual focal loss for fault diagnosis of rotating machinery under data imbalance conditions ［J］. IEEE Transactions on Instrumentation and Measurement， 2024， 73： No.3506211.
[3]	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.
[4]	LeCun Y， Bengio Y， Hinton G. Deep learning ［J］. Nature， 2015， 521（7553）： 436-444.
[5]	Li H， Wu X， Liu T， et al. Rotating machinery fault diagnosis based on typical resonance demodulation methods： a review ［J］. IEEE Sensors Journal， 2023， 23（7）： 6439-6459.
[6]	陈艳霞，李鑫明，王志勇，等.基于LSTM-CNN-Attention模型的电力设施非周期巡视决策方法［J］.计算机应用， 2023， 43（）： 291-297.
	Chen Yanxia， Li Xinming， Wang Zhiyong， et al. Non-periodic inspection decision method of power facility based on LSTM-CNN-Attention model ［J］. Journal of Computer Applications， 2023， 43（）： 291-297.
[7]	Gao B， Xu J， Zhang Z， et al. Marine diesel engine piston ring fault diagnosis based on LSTM and improved beluga whale optimization ［J］. Alexandria Engineering Journal， 2024， 109： 213-228.
[8]	Wu G， Ji X， Yang G， et al. Signal-to-image： rolling bearing fault diagnosis using ResNet family deep-learning models ［J］. Processes， 2023， 11（5）： No.1527.
[9]	Zhang C， Wang F， Zhou D， et al. A CNN-based fault diagnosis method of multi-function integrated RF system using frequency domain scanning with Lasso regression ［J］. Knowledge-Based Systems， 2025， 309： No.112836.
[10]	王妍，王新发，王延峰，等.基于改进天鹰优化算法优化LSTM的滚动轴承故障诊断方法［J］.振动与冲击， 2024， 43（23）： 144-154.
	Wang Yan， Wang Xinfa， Wang Yanfeng， et al. A rolling bearing fault diagnosis method based on improved Aquila optimization algorithm to optimize LSTM ［J］. Journal of Vibration and Shock， 2024， 43（23）： 144-154.
[11]	Huang X， Xie T， Wu J， et al. Deep continuous convolutional networks for fault diagnosis ［J］. Knowledge-Based Systems， 2024， 292： No.111623.
[12]	Wang Q， Sun Z， Zhu Y， et al. A fault diagnosis method based on an improved diffusion model under limited sample conditions ［J］. PLoS ONE， 2024， 19（9）： No.e0309714.
[13]	Liu H， Liu Z， Jia W， et al. A novel imbalanced data classification method based on weakly supervised learning for fault diagnosis ［J］. IEEE Transactions on Industrial Informatics， 2022， 18（3）： 1583-1593.
[14]	Zhang T， Chen J， Li F， et al. Intelligent fault diagnosis of machines with small & imbalanced data： a state-of-the-art review and possible extensions ［J］. ISA Transactions， 2022， 119： 152-171.
[15]	Li X， Wu X， Wang T， et al. Fault diagnosis method for imbalanced data based on adaptive diffusion models and generative adversarial networks ［J］. Engineering Applications of Artificial Intelligence， 2025， 147： No.110410.
[16]	Fu Z， Liu Z， Ping S， et al. TRA-ACGAN： a motor bearing fault diagnosis model based on an auxiliary classifier generative adversarial network and transformer network ［J］. ISA Transactions， 2024， 149： 381-393.
[17]	周建民，沈熙闻，刘露露.结合WGAN-GP与CNN-SVM的滚动轴承故障红外诊断［J］.激光与红外， 2024， 54（3）： 416-422.
	Zhou Jianming， Shen Xiwen， Liu Lulu. Infrared diagnosis of rolling bearing faults combined with WGAN-GP and CNN-SVM ［J］. Laser and Infrared， 2024， 54（3）： 416-422.
[18]	Wang Y， Li D， Li L， et al. A novel deep learning framework for rolling bearing fault diagnosis enhancement using VAE-augmented CNN model ［J］. Heliyon， 2024， 10（15）： No.e35407.
[19]	Li X， Xiao B， Chen X， et al. MSDF-VAE： a cloud-edge collaborative method for fault diagnosis based on transfer learning ［J］. IEEE Internet of Things Journal， 2025， 12（12）： 22393-22403.
[20]	Yan K， Su J， Huang J， et al. Chiller fault diagnosis based on VAE-enabled generative adversarial networks ［J］. IEEE Transactions on Automation Science and Engineering， 2022， 19（1）： 387-395.
[21]	Yang G， Zhong Y， Yang L， et al. Fault diagnosis of harmonic drive with imbalanced data using generative adversarial network ［J］. IEEE Transactions on Instrumentation and Measurement， 2021， 70： No.3519911.
[22]	Ping Z， Wang D， Zhang Y， et al. Few-shot aero-engine bearing fault diagnosis with denoising diffusion based data augmentation ［J］. Neurocomputing， 2025， 622： No.129327.
[23]	Sohl-Dickstein J， Weiss E， Maheswaranathan N， et al. Deep unsupervised learning using nonequilibrium thermodynamics ［C］// ICML 2015. New York： JMLR.org， 2015： 2256-2265.
[24]	Mueller P N. Attention-enhanced conditional-diffusion-based data synthesis for data augmentation in machine fault diagnosis ［J］. Engineering Applications of Artificial Intelligence， 2024， 131： No.107696.
[25]	吴静远，舒启林，王耿，等.基于扩散模型的滚动轴承小样本故障诊断研究［J］.组合机床与自动化加工技术， 2024（12）： 117-122.
	Wu Jingyuan， Shu Qilin， Wang Geng， et al. Research on small-sample fault diagnosis of rolling bearings based on diffusion model ［J］. Modular Machine Tool and Automatic Manufacturing Technique， 2024（12）： 117-122.
[26]	Zhu J. Synthetic data generation by diffusion models ［J］. National Science Review， 2024， 11（8）： No.nwae276.
[27]	Zhao Y， Sheng T， Li D. Data augmentation fault diagnosis of rolling machinery using condition denoising diffusion probabilistic model and improved CNN ［J］. IEEE Transactions on Instrumentation and Measurement， 2025， 74： No.3517712.
[28]	Ahsan M， Salah M M. Similarity index of the STFT-based health diagnosis of variable speed rotating machines ［J］. Intelligent Systems with Applications， 2023， 20： No.200270.
[29]	谢苗，孟庆爽，李博，等.融合短时傅里叶变换和卷积神经网络的托辊故障诊断方法［J］.工程设计学报， 2024， 31（5）： 565-574.
	Xie Miao， Meng Qingshuang， Li Bo， et al. Idler fault diagnosis method fusing short-time Fourier transform and convolutional neural network ［J］. Chinese Journal of Engineering Design， 2024， 31（5）： 565-574.
[30]	Huang G， Liu Z， van der Maaten L， et al. Densely connected convolutional networks ［C］// CVPR 2017. Piscataway： IEEE， 2017： 2261-2269.
[31]	Smith W A， Randall R B. Rolling element bearing diagnostics using the Case Western Reserve University data： a benchmark study ［J］. Mechanical Systems and Signal Processing， 2015， 64/65： 100-131.
[32]	Wang C， Huang C， Zhang L， et al. Denoising diffusion implicit model combined with TransNet for rolling bearing fault diagnosis under imbalanced data ［J］. Sensors， 2024， 24（24）： No.8009.
[33]	Zhang T， Lin J， Jiao J， et al. An interpretable latent denoising diffusion probabilistic model for fault diagnosis under limited data ［J］. IEEE Transactions on Industrial Informatics， 2024， 20（8）： 10354-10365.
[34]	Shmelkov K， Schmid C， Alahari K. How good is my GAN？［C］// ECCV 2018. Cham： Springer， 2018： 218-234.
[35]	Liu G， Wu L. Incremental bearing fault diagnosis method under imbalanced sample conditions ［J］. Computers and Industrial Engineering， 2024， 192： No.110203.

故障类别	状态	尺寸/mil	转速/（r·min^-1）	样本数
1	正常		1 797	928
2	内圈故障	7	1 797	928
3	内圈故障	14	1 797	928
4	内圈故障	21	1 797	928
5	外圈故障	7	1 797	928
6	外圈故障	14	1 797	928
7	外圈故障	21	1 797	928
8	滚动体故障	7	1 797	928
9	滚动体故障	14	1 797	928
10	滚动体故障	21	1 797	928

故障类别	状态	尺寸/mil	转速/（r·min^-1）	样本数
1	正常		1 797	928
2	内圈故障	7	1 797	928
3	内圈故障	14	1 797	928
4	内圈故障	21	1 797	928
5	外圈故障	7	1 797	928
6	外圈故障	14	1 797	928
7	外圈故障	21	1 797	928
8	滚动体故障	7	1 797	928
9	滚动体故障	14	1 797	928
10	滚动体故障	21	1 797	928

方法	训练轮次	批次大小	学习率	优化器
ACGAN	10 000	64	0.000 2	Adam
WGAN-GP	10 000	64	0.000 2	Adam
VAE	1 000	64	0.000 2	Adam
DDIM	500	32	0.001 0	AdamW
IVQ-LDM	2 000	20	0.001 0	Adam
CDDPM	300	64	0.000 1	Adam
本文方法	500	32	0.001 0	AdamW

方法	训练轮次	批次大小	学习率	优化器
ACGAN	10 000	64	0.000 2	Adam
WGAN-GP	10 000	64	0.000 2	Adam
VAE	1 000	64	0.000 2	Adam
DDIM	500	32	0.001 0	AdamW
IVQ-LDM	2 000	20	0.001 0	Adam
CDDPM	300	64	0.000 1	Adam
本文方法	500	32	0.001 0	AdamW

指标	描述
SSIM	衡量2个样本之间的结构相似性。越接近于1，表明2个样本相似度越高
HF-SSIM	针对图像高频成分的结构相似性度量，能精确捕捉细节差异。越接近于1，表明图像高频成分信息越接近
MMD	衡量2个概率分布在核希尔伯特空间中的距离。越接近于0，表明二者相似度越高
JSD	在本文应用场景下用于衡量生成样本与真实样本在归一化时频能量分布上的差异。数值越接近于0，表示生成样本在时频结构上的物理意义越接近真实样本

Sample augmentation method for fault diagnosis with high-frequency guided diffusion model

高频引导的扩散模型故障诊断样本增广方法

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 23

References 35

Related Articles 15

Recommended Articles

Metrics

方法	SSIM	HF-SSIM	MMD	JSD
ACGAN	0.296	0.165	2.144	0.403
WGAN-GP	0.487	0.371	1.405	0.223
VAE	0.307	0.149	2.013	0.387
DDIM	0.417	0.321	1.239	0.256
CDDPM	0.391	0.315	1.458	0.218
IVQ-LDM	0.364	0.279	1.770	0.284
本文方法	0.503	0.418	0.967	0.182

方法	GAN-train	GAN-test
原始数据	98.26±0.43
ACGAN	81.16±1.09	85.27±1.69
WGAN-GP	93.46±0.96	92.67±1.34
VAE	84.30±0.65	82.39±1.01
DDIM	90.77±0.90	91.33±0.49
CDDPM	92.17±1.21	91.40±0.84
IVQ-LDM	88.67±1.12	88.42±0.87
本文方法	95.57±0.63	95.71±0.75

实验序号	故障样本	生成样本	正常样本	比例
1	4	0	400	1∶100
2	4	4	400	1∶50
3	4	36	400	1∶10
4	4	76	400	1∶5
5	4	196	400	1∶2
6	4	396	400	1∶1

方法	SSIM	HF-SSIM	MMD	JSD
ACGAN	0.248	0.217	2.410	0.473
WGAN-GP	0.391	0.339	1.104	0.243
VAE	0.271	0.157	2.235	0.350
DDIM	0.406	0.344	1.429	0.276
SMOTEENN	0.387	0.329	1.575	0.291
IVQ-LDM	0.363	0.309	1.813	0.285
CDDPM	0.389	0.327	1.307	0.270
本文方法	0.453	0.392	0.917	0.193

方法	GAN-train	GAN-test
原始数据	97.60±0.29
ACGAN	75.69±1.03	80.12±2.10
WGAN-GP	92.19±0.74	93.67±0.88
VAE	78.34±0.81	83.46±0.68
DDIM	91.73±0.94	91.33±0.72
SMOTEENN	90.39±1.00	93.82±0.59
CDDPM	94.20±0.72	92.47±0.59
IVQ-LDM	92.31±0.66	91.87±0.74
本文方法	96.23±0.49	95.59±0.53

方法	1∶100	1∶50	1∶10	1∶5	1∶2	1∶1
ACGAN	49.35	58.97	65.71	72.17	78.32	82.50
WGAN-GP	49.89	64.54	74.12	81.60	89.43	94.31
VAE	48.47	59.33	63.80	75.47	80.68	85.16
DDIM	48.91	62.76	75.39	80.61	87.76	91.09
SMOTEENN	49.10	61.69	74.37	82.26	88.13	92.16
CDDPM	48.61	63.58	74.41	83.41	91.80	94.50
IVQ-LDM	49.30	61.36	72.50	79.24	89.65	92.50
本文方法	49.38	66.42	76.83	85.47	92.59	96.66

实验序号	灰度谱图	辅助分类器	高频引导模块	SSIM	HF-SSIM	MMD	JSD
1	√	×	×	0.410	0.317	1.430	0.254
2	√	×	√	0.463	0.384	1.037	0.207
3	√	√	×	0.429	0.341	1.253	0.221
4	√	√	√	0.503	0.418	0.967	0.182

实验序号	灰度谱图	辅助分类器	高频引导模块	训练时长/h	准确率/%
1	×	×	×	2.71	93.48
2	×	√	√	2.34	95.79
3	×	√	×	2.30	93.67
4	×	×	√	2.87	94.40
5	√	√	×	2.58	95.03
6	√	×	√	3.62	96.29
7	√	√	√	2.96	97.41

[1]	Liwan YAO, Hailong LIU, Zhangfan ZENG. Frequency-domain driven and diffusion-based fusion for sonar image enhancement algorithm [J]. Journal of Computer Applications, 2026, 46(6): 1947-1955.
[2]	Ronghui ZHAO, Chao DENG, Zidong YU. Construction and application of knowledge graph for fault diagnosis of key components of aviation equipment [J]. Journal of Computer Applications, 2026, 46(5): 1604-1613.
[3]	Min CHEN, Xiaolin QIN, Shaohan LI, Hao YANG, Taohong LI. Review of deep learning applications in severe convective weather prediction [J]. Journal of Computer Applications, 2026, 46(3): 980-992.
[4]	Peirong SHAO, Suzhen LIN, Yanbo WANG. Human-centric detail-enhanced virtual try-on method [J]. Journal of Computer Applications, 2026, 46(3): 915-923.
[5]	Yan HU, Peng LI, Shuyan CHENG. Adversarial purification method based on directly guided diffusion model [J]. Journal of Computer Applications, 2026, 46(3): 821-829.
[6]	Yuan JIA, Deyu YUAN, Yuquan PAN, Anran WANG. Watermarking method for diffusion model output [J]. Journal of Computer Applications, 2026, 46(1): 161-168.
[7]	Xingjie FENG, Xingpeng BIAN, Xiaorong FENG, Xinglong WANG. Incremental missing value imputation algorithm for time series based on diffusion model [J]. Journal of Computer Applications, 2025, 45(8): 2582-2591.
[8]	Qiang LI, Shaoxiong BAI, Yuan XIONG, Wei YUAN. Privacy preserving localization of surveillance images based on large vision models [J]. Journal of Computer Applications, 2025, 45(3): 832-839.
[9]	Meng WANG, Daqian ZHANG, Bingyan ZHOU, Qianying MA, Jidong LYU. Fault diagnosis method for train control on-board interface equipment of CTCS-3 based on temporal knowledge graph completion [J]. Journal of Computer Applications, 2025, 45(2): 677-684.
[10]	Xinran XU, Shaobing ZHANG, Miao CHENG, Yang ZHANG, Shang ZENG. Bearings fault diagnosis method based on multi-pathed hierarchical mixture-of-experts model [J]. Journal of Computer Applications, 2025, 45(1): 59-68.
[11]	Jie WU, Ansi ZHANG, Maodong WU, Yizong ZHANG, Congbao WANG. Overview of research and application of knowledge graph in equipment fault diagnosis [J]. Journal of Computer Applications, 2024, 44(9): 2651-2659.
[12]	Chenyang LI, Long ZHANG, Qiusheng ZHENG, Shaohua QIAN. Multivariate controllable text generation based on diffusion sequences [J]. Journal of Computer Applications, 2024, 44(8): 2414-2420.
[13]	Jinsong XU, Ming ZHU, Zhiqiang LI, Shijie GUO. Location control method for generated objects by diffusion model with exciting and pooling attention [J]. Journal of Computer Applications, 2024, 44(4): 1093-1098.
[14]	Yusheng LIU, Xuezhong XIAO. High-fidelity image editing based on fine-tuning of diffusion model [J]. Journal of Computer Applications, 2024, 44(11): 3574-3580.
[15]	Jian ZHANG, Peiyuan CHENG, Siyu SHAO. Rotary machine fault diagnosis based on improved residual convolutional auto-encoding network and class adaptation [J]. Journal of Computer Applications, 2022, 42(8): 2440-2449.

方法	不同样本比例下的诊断准确率/%
方法	1∶100	1∶50	1∶10	1∶5	1∶2	1∶1
ACGAN	53.48	58.83	71.39	73.66	81.53	83.82
WGAN-GP	52.81	62.30	81.72	89.32	92.77	95.49
VAE	53.70	59.37	70.42	78.78	85.10	88.36
DDIM	51.97	61.39	75.04	84.21	90.76	94.12
CDDPM	53.13	64.61	80.07	89.57	91.80	94.48
IVQ-LDM	52.40	60.78	76.36	87.51	89.65	92.83
本文方法	52.23	69.58	81.39	90.61	94.49	97.41

方法	不同样本比例下的诊断准确率/%
方法	1∶100	1∶50	1∶10	1∶5	1∶2	1∶1
ACGAN	53.48	58.83	71.39	73.66	81.53	83.82
WGAN-GP	52.81	62.30	81.72	89.32	92.77	95.49
VAE	53.70	59.37	70.42	78.78	85.10	88.36
DDIM	51.97	61.39	75.04	84.21	90.76	94.12
CDDPM	53.13	64.61	80.07	89.57	91.80	94.48
IVQ-LDM	52.40	60.78	76.36	87.51	89.65	92.83
本文方法	52.23	69.58	81.39	90.61	94.49	97.41