Please wait a minute...
工程设计学报  2024, Vol. 31 Issue (2): 137-150    DOI: 10.3785/j.issn.1006-754X.2024.03.212
机械设计理论与方法     
基于多判别器辅助分类器生成对抗网络的故障诊断方法研究
叶子汉1,2(),王中华1,2(),姜潮1,2,吕新1,2,张哲1,2
1.湖南大学 整车先进设计制造技术全国重点实验室,湖南 长沙 410082
2.湖南大学 机械与运载工程学院,湖南 长沙 410082
Research on fault diagnosis method based on multi-discriminator auxiliary classifier generative adversarial network
Zihan YE1,2(),Zhonghua WANG1,2(),Chao JIANG1,2,Xin Lü1,2,Zhe ZHANG1,2
1.State Key Laboratory of Advanced Design and Manufacturing Technology for Vehicle, Hunan University, Changsha 410082, China
2.College of Mechanical and Vehicle Engineering, Hunan University, Changsha 410082, China
 全文: PDF(2872 KB)   HTML
摘要:

在强冲击、强辐射、极高温等极端恶劣的工作环境下,机械设备的故障模式复杂多样,获得充足且有效的故障数据变得非常困难甚至难以实现,以致故障诊断的准确性受限,后续检修维护方案难以有效制定。针对这一问题,提出了一种多判别器辅助分类器生成对抗网络的数据增强算法。通过设置3个判别器、1个生成器并添加独立的分类器,构建了新的辅助分类器生成对抗网络模型。针对在该模型训练中存在的不稳定性问题,通过引入Wasserstein距离构造新的损失函数,并采用稳定性更具优势的单边软约束正则化项替换原有的L2梯度惩罚项来解决模型崩溃问题;在此基础上,采用高效通道注意力机制来进一步提高模型的特征提取能力。将所提出的模型应用于扩充机械设备故障数据集,辅助深度学习智能诊断模型的训练。多个故障数据集扩充实验表明,与现有模型相比,新模型所生成数据的质量更高,故障诊断的准确率也得到进一步提高,因此具有较高的应用价值。

关键词: 多判别器辅助分类器生成对抗网络高效通道注意力机制Lipschitz(利普希茨)约束数据增强故障诊断    
Abstract:

In extremely harsh working environments such as strong impacts, intense radiation and extremely high temperature, the fault modes of mechanical equipment are complex and varied, and it is very difficult to obtain sufficient and effective fault data, even difficult to achieve, so that the accuracy of fault diagnosis is limited, and subsequent maintenance and repair programs are difficult to be effectively developed.To solve this problem, a data enhancement algorithm for multi-discriminator auxiliary classifier generative adversarial network was proposed. By setting up 3 discriminators, 1 generator and adding independent classifier, a new auxiliary classifier generative adversarial network model was constructed. Aiming at the instability issue in the model's training, the Wasserstein distance was introduced to construct a new loss function, and the unilateral soft constraint regularization term with more stability was used to replace the original L2 gradient penalty term to solve the problem of model collapse. Building on this, an efficient channel attention mechanism was adopted to further improve the model's feature extraction capability. The proposed model was applied to extend the fault data set of mechanical equipment to assist the training of deep learning intelligent diagnosis model. Multiple fault data set expansion experiments showed that compared with the existing model, the new model could generate higher quality data, and the accuracy of fault diagnosis was further improved, so it had high application value.

Key words: multi-discriminator auxiliary classifier generative adversarial network    efficient channel attention mechanism    Lipschitz penalty    data augmentation    fault diagnosis
收稿日期: 2023-10-30 出版日期: 2024-04-26
CLC:  TP 277  
基金资助: 国防基础科研计划资助项目(JCKY2020110C105);国家自然科学基金资助项目(52205262);整车先进设计制造技术全国重点实验室开放基金资助项目(32175001)
通讯作者: 王中华     E-mail: 240014070@qq.com;wangzh0946@hnu.edu.cn
作者简介: 叶子汉(1998—),男,湖南株洲人,硕士,从事复杂装备故障诊断研究,E-mail: 240014070@qq.com, https://orcid.org/0009-0007-5011-2448
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章  
叶子汉
王中华
姜潮
吕新
张哲

引用本文:

叶子汉,王中华,姜潮,吕新,张哲. 基于多判别器辅助分类器生成对抗网络的故障诊断方法研究[J]. 工程设计学报, 2024, 31(2): 137-150.

Zihan YE,Zhonghua WANG,Chao JIANG,Xin Lü,Zhe ZHANG. Research on fault diagnosis method based on multi-discriminator auxiliary classifier generative adversarial network[J]. Chinese Journal of Engineering Design, 2024, 31(2): 137-150.

链接本文:

https://www.zjujournals.com/gcsjxb/CN/10.3785/j.issn.1006-754X.2024.03.212        https://www.zjujournals.com/gcsjxb/CN/Y2024/V31/I2/137

图1  GAN和ACGAN框架
图2  MDACGAN框架
网络结构
Discriminator0Conv2D(32,3,2,same)+MaxPool2D(2,1,same)
Conv2D(64,3,2,same)+MaxPool2D(2,1,same)
Conv2D(128,3,2,same)+MaxPool2D(2,1,same)
Conv2D(256,3,2,same)+MaxPool2D(2,1,same)
GlobalAvgpool2D()
Dense(1)
Discriminator1Conv2D(64,3,2,same)+MaxPool2D(2,1,same)
Conv2D(128,3,2,same)+MaxPool2D(2,1,same)
Conv2D(256,3,2,same)+MaxPool2D(2,1,same)
Conv2D(512,3,2,same)+MaxPool2D(2,1,same)
GlobalAvgpool2D()
Dense(1)
Discriminator2Flatten()
Dense(256)+Dropout(0.2)
Dense(128)+Dropout(0.2)
Dense(64)+Dropout(0.2)
Dense(1)
GeneratorDense(8192)
Conv2Dtranspose(128,5,2,same)+BN()
Conv2Dtranspose(64,5,2,same)+BN()
Conv2Dtranspose(32,5,2,same)+BN()
Conv2Dtranspose(1,5,2,same)
ClassifierConv2D(32,3,2,same)+BN()+MaxPool2D(2,2,same)
Conv2D(64,3,2,same)+BN()+MaxPool2D(2,2,same)
Conv2D(128,3,2,same)+BN()+MaxPool2D(2,2,same)
Conv2D(256,3,2,same)+BN()+MaxPool2D(2,2,same)
ECA_block()
GlobalAvgpool2D()
Dense(8)
Optimizer of Discriminator0RMSprop(0.0002,0.5)
Optimizer of Discriminator1RMSprop(0.0002,0.5)
Optimizer of Discriminator2RMSprop(0.0002,0.5)
Optimizer of GeneratorRMSprop(0.0002,0.5)
Optimizer of ClassifierAdam(0.00002,0.5)
表1  MDACGAN模型的结构
标签故障类型
012k_Drive_End_B007_0
112k_Drive_End_B021_1
212k_Drive_End_IR007_0
312k_Drive_End_IR021_1
412k_Drive_End_OR007@3_0
512k_Drive_End_OR007@6_0
612k_Drive_End_OR007@12_0
712k_Drive_End_OR021@3_1
表2  实验1的数据标签及故障类型
图3  实验1的部分故障数据灰度图
图4  不同模型对CWRU轴承故障数据集生成的样本
模型类别
01234567
ACGAN532.10532.11532.10532.11532.01532.10532.11532.11
ACWGAN-GP179.08326.34488.76171.87236.83199.66220.80532.00
MDACGAN34.3179.47220.2562.2868.2082.7188.68147.58
表3  CWRU轴承故障数据集真实样本与生成样本之间的 DFI 值
模型类别
01234567
ACGAN4.330 54.873 74.661 95.337 84.445 54.761 75.637 85.071 2
ACWGAN-GP0.340 60.412 00.419 70.343 70.702 40.352 20.603 90.635 6
MDACGAN0.149 00.145 60.152 30.153 20.112 40.146 30.155 30.155 7
表4  CWRU轴承故障数据集真实样本与生成样本之间的 DMM 值
图5  CWRU轴承故障数据集生成样本和训练样本 t-SNE可视化结果
类型数据集样本总数/个
测试集数据集0100
训练集数据集150(0)
数据集2100(50)
数据集3150(100)
数据集4200(150)
数据集5250(200)
数据集6350(300)
数据集7200(0)
表5  CWRU轴承故障数据集扩充和划分
模型准确率/%
数据集1数据集2数据集3数据集4数据集5数据集6数据集7
ACWGAN-GP78.3789.3792.6291.1293.8594.8799.75
MDACGAN78.3790.2593.7595.6297.1398.0099.75
表6  CWRU轴承故障数据集分类结果
标签工况故障类型失效位置
01Bearing1_1外圈
1Bearing1_4保持架
22Bearing2_1内圈
3Bearing2_2外圈
43Bearing3_3内圈
5Bearing3_5外圈
表7  实验2的数据标签及故障类型
图6  不同模型对 XJTU-SY数据集生成的样本
模型类别
012345
ACWGAN-GP270.87223.04220.8469.56116.89111.72
MDACGAN141.37178.31144.8760.16110.6566.07
表8  XJTU-SY数据集真实样本与生成样本之间的 DFI 值
模型类别
012345
ACWGAN-GP0.233 80.352 90.310 90.239 00.339 30.213 3
MDACGAN0.164 80.198 20.160 20.115 00.178 90.122 8
表9  XJTU-SY数据集真实样本与生成样本之间的 DMM 值
图7  XJTU-SY数据集生成样本和训练样本 t-SNE可视化结果
模型准确率/%
数据集1数据集2数据集3数据集4数据集5数据集6数据集7
ACWGAN-GP75.0083.1683.5087.3390.1292.1699.66
MDACGAN75.0085.8688.2189.6691.8395.5099.66
表10  XJTU-SY数据集分类结果
标签试验序号测试的轴承失效位置
01轴承3内圈
1轴承4滚动体
22轴承1外圈
3轴承2
43轴承1
5轴承3外圈
6轴承4
表11  实验3的数据标签及故障类型
图8  不同模型对IMS数据集生成的样本
模型类别
0123456
ACWGAN-GP255.29148.21515.00195.62173.14237.86106.51
MDACGAN120.0023.40337.67141.6188.0190.9277.35
表12  IMS数据集真实样本与生成样本之间的 DFI 值
模型类别
0123456
ACWGAN-GP0.366 80.323 60.321 60.344 20.311 20.332 40.324 5
MDACGAN0.180 50.141 50.213 30.134 20.144 30.197 00.116 3
表13  IMS数据集真实样本与生成样本的 DMM 值
图9  IMS数据集生成样本和训练样本 t-SNE可视化结果
模型准确率/%
数据集1数据集2数据集3数据集4数据集5数据集6数据集7
ACWGAN-GP75.2886.8588.7191.0089.0191.8597.28
MDACGAN75.2886.7188.8592.7193.1495.5797.28
表14  IMS数据集分类结果
1 ZHONG J, HUANG Y. Time-frequency representation based on an adaptive short-time Fourier transform[J]. IEEE Transactions on Signal Processing, 2010, 58(10): 5118-5128.
2 TSE P W, PENG Y H, YAM R. Wavelet analysis and envelope detection for rolling element bearing fault diagnosis: Their effectiveness and flexibilities [J]. Journal of Vibration and Acoustics, 2001, 123(3): 303-310.
3 HUANG N E, SHEN Z, LONG S R, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis[J]. Proceedings of the Royal Society of London Series A: Mathematical, Physical and Engineering Sciences, 1998, 454: 903-995.
4 DRAGOMIRETSKIY K, ZOSSO D. Variational mode decomposition[J]. IEEE Transactions on Signal Processing, 2013, 62(3): 531-544.
5 ZHANG W, ZHANG F, CHEN W, et al. Fault state recognition of rolling bearing based fully convolutional network[J]. Computing in Science & Engineering, 2019, 21(5): 55-63.
6 WANG D, GUO Q, SONG Y, et al. Application of multiscale learning neural network based on CNN in bearing fault diagnosis[J]. Journal of Signal Processing Systems, 2019, 91(10): 1205-1217.
7 HE M, HE D. Deep learning based approach for bearing fault diagnosis[J]. IEEE Transactions on Industry Applications, 2017, 53(3): 3057-3065.
8 RUAN D, SONG X, GÜHMANN C, et al. Collaborative optimization of CNN and GAN for bearing fault diagnosis under unbalanced datasets[J/OL]. Lubricants, 2021, 9(10): 105[2023-10-20]. .
9 XIAO D, HUANG Y, QIN C, et al. Transfer learning with convolutional neural networks for small sample size problem in machinery fault diagnosis[J]. Proceedings of the Institution of Mechanical Engineers Part C: Journal of Mechanical Engineering Science, 2019, 233(14): 5131-5143.
10 崔旭浩,郗欣甫,孙以泽.数据驱动的经编机横移机构故障检测方法研究[J].工程设计学报,2022,29(3):263-271. doi:10.3785/j.issn.1006-754X.2022.00.030
CUI X H, XI X F, SUN Y Z. Research on data-driven fault detection method of traverse mechanism of warp knitting machine[J]. Chinese Journal of Engineering Design, 2022, 29 (3): 263-271.
doi: 10.3785/j.issn.1006-754X.2022.00.030
11 REN S, ZHU W, LIAO B, et al. Selection-based resampling ensemble algorithm for nonstationary imbalanced stream data learning[J]. Knowledge-Based Systems, 2019, 163: 705-722.
12 陈保家,陈学力,肖文荣,等.小样本下滚动轴承故障的多源域迁移诊断方法[J].电子测量与仪器学报,2022, 36(2): 219-228.
CHEN J B, CHEN X L, XIAO W R, et al. Multi-source domain transfer diagnosis method for rolling bearing faults under small samples[J]. Journal of Electronic Measurement and Instrumentation, 2022, 36(2): 219-228.
13 吴定会,方钦,吴楚宜.基于数据生成与迁移学习的轴承小样本故障诊断[J].机械传动,2020,44(11):139-144.
WU D H, FANG Q, WU C Y. Bearing small sample fault diagnosis based on data generation and transfer learning [J]. Journal of Mechanical Transmission, 2020, 44(11): 139-144.
14 GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial nets[C]//Proceedings of the 27th International Conference on Neural Information Processing Systems. Montreal, Canada: MIT Press, 2014: 2672-2680.
15 GAO X, DENG F, YUE X. Data augmentation in fault diagnosis based on the Wasserstein generative adversarial network with gradient penalty[J]. Neurocomputing, 2020, 396: 487-494.
16 HAN T, LIU C, YANG W, et al. A novel adversarial learning framework in deep convolutional neural network for intelligent diagnosis of mechanical faults[J]. Knowledge-Based Systems, 2019, 165: 474-487.
17 ZHOU F, YANG S, FUJITA H, et al. Deep learning fault diagnosis method based on global optimization GAN for unbalanced data[J]. Knowledge-Based Systems, 2020, 187: 104837.
18 LIU S, CHEN J, QU C, et al. LOSGAN: Latent optimized stable GAN for intelligent fault diagnosis with limited data in rotating machinery[J]. Measurement Science and Technology, 2021, 32(4): 045101.
19 ODENA A, OLAH C, SHLENS J. Conditional image synthesis with auxiliary classifier GANS[EB/OL]. [2023-10-20]. .
20 SHAO S, WANG P, YAN R. Generative adversarial networks for data augmentation in machine fault diagnosis [J]. Computers in Industry, 2019, 106: 85-93.
21 卢锦玲,张祥国,张伟,等.基于改进辅助分类生成对抗网络的风机主轴承故障诊断[J].电力系统自动化,2021, 45(7):148-154. doi:10.7500/AEPS20200415002
LU J L, ZHANG X G, ZHANG W, et al. Fault diagnosis of main bearing of wind turbine based on improved auxiliary classifier generative adversarial network[J]. Automation of Electric Power System, 2021, 45(7): 148-154.
doi: 10.7500/AEPS20200415002
22 HUANG N, CHEN Q, CAI G, et al. Fault diagnosis of bearing in wind turbine gearbox under actual operating conditions driven by limited data with noise labels[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 1-10.
23 何强,唐向红,李传江,等.负载不平衡下小样本数据的轴承故障诊断[J].中国机械工程,2021,32(10):1164-1171. doi:10.3969/j.issn.1004-132X.2021.10.004
HE Q, TANG X H, LI C J, et al. Bearing fault diagnosis method based on small sample data under unbalanced loads[J]. China Mechanical Engineering, 2021, 32(10): 1164-1171.
doi: 10.3969/j.issn.1004-132X.2021.10.004
24 YANG Z K, BU L P, WANG T, et al. Fire image generation based on ACGAN[C]//The 31th Chinese Control and Decision Conference, Nanchang, Jiangxi, Jun., 3-5, 2019.
25 DHARANYA V, JOSEPH RAJ A N, GOPI V P. Facial expression recognition through person-wise regeneration of expressions using auxiliary classifier generative adversarial network (AC-GAN) based model[J]. Journal of Visual Communication and Image Representation, 2021, 77: 103110.
26 LI W, ZHONG X, SHAO H, et al. Multi-mode data augmentation and fault diagnosis of rotating machinery using modified ACGAN designed with new framework[J]. Advanced Engineering Informatics, 2022, 52: 101552.
27 ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein generative adversarial networks[C]//34th International Conference on Machine Learning, Sydney, Australia, Aug. 6-11, 2017.
28 王正龙,张保稳.生成对抗网络研究综述[J].网络与信息安全学报,2021,7(4):68-85. doi:10.11959/j.issn.2096-109x.2021080
WANG Z L, ZHANG B W. A survey of research on generative adversarial network[J]. Chinese Journal of Network and Information Security, 2021, 7(4): 68-85.
doi: 10.11959/j.issn.2096-109x.2021080
29 GULRAJANI I, AHMED F, ARJOVSKY M, et al. Improved training of Wasserstein GANS[EB/OL]. [2023-10-20]. .
30 MIRZA M, OSINDERO S. Conditional generative adversarial nets[J]. Computer Science, 2014: 2672-2680.
31 ODENA A. Semi-supervised learning with generative adversarial networks[EB/OL]. [2023-10-20]. .
32 MORDIDO G, YANG H, MEINEL C. Dropout-GAN: Learning from a dynamic ensemble of discriminators[J]. arXiv preprint arXiv:, 2018.
33 WANG Q, WU B, ZHU P, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks [C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, Washington, USA, Jun. 14-19. 2020.
34 PETZKA H, FISCHER A, LUKOVNICOV D. On the regularization of Wasserstein GANS[J]. arXiv preprint arXiv:, 2017.
35 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.
36 WEN L, LI X, GAO L, et al. A new convolutional neural network-based data-driven fault diagnosis method[J]. IEEE Transactions on Industrial Electronics, 2018, 65(7): 5990-5998.
37 HEUSEL M, RAMSAUER H, UNTERTHINER T, et al. GANS trained by a two time-scale update rule converge to a local Nash equilibrium[C]//NIPS'17: Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, CA, USA, Dec. 4-9, 2017.
38 GRETTON A, BORGWARDT K M, RASCH M J, et al. A kernel two-sample test[J]. The Journal of Machine Learning Research, 2012, 13(1): 723-773.
39 PEZZOTTI N, LELOEVELDT B P, VAN DER MAATEN L, et al. Approximated and user steerable tSNE for progressive visual analytics[J]. IEEE Transactions on Visualization and Computer Graphics, 2016, 23(7): 1739-1752.
40 雷亚国,韩天宇,王彪,等.XJTU-SY滚动轴承加速寿命试验数据集解读[J].机械工程学报,2019,55(16):1-6. doi:10.3901/jme.2019.16.001
LEI Y G, HAN T Y, WANG B, et al. XJTU-SY rolling element bearing accelerated life test datasets: A tutorial [J]. Journal of Mechanical Engineering, 2019, 55(16): 1-6.
doi: 10.3901/jme.2019.16.001
[1] 纪超,王亮,王孝敬,李小兵,曹雯. 基于MSSA-SVM的电缆隧道故障预警系统设计[J]. 工程设计学报, 2023, 30(1): 109-116.
[2] 吴国沛, 余银犬, 涂文兵. 永磁同步电机故障诊断研究综述[J]. 工程设计学报, 2021, 28(5): 548-558.
[3] 倪洪启, 金驰, 冯霏. 波纹补偿器故障诊断系统研制[J]. 工程设计学报, 2019, 26(3): 354-363.
[4] 马天兵, 王孝东, 杜菲, 王鑫泉. 基于GA-SVM的刚性罐道故障诊断[J]. 工程设计学报, 2019, 26(2): 170-176.
[5] 胡艺耀, 朱斌, 张伟, 何畏, 沈平生. 知识库构建工具软件的设计与实现[J]. 工程设计学报, 2018, 25(4): 367-373.
[6] 李晓豁,翁正洋,钱亚森,史尚伟,李 岩. 基于果蝇算法优化模糊RBF网络的液压破碎锤故障诊断[J]. 工程设计学报, 2015, 22(6): 540-545.
[7] 李玲玲, 景丽婷, 马东娟, 李志刚. 改进证据理论及其在电力系统故障诊断中的应用[J]. 工程设计学报, 2012, 19(6): 485-488.
[8] 韩海涛,马红光,韩琨,郑耿乐. 关于Volterra频域核辨识的多音激励信号设计[J]. 工程设计学报, 2012, 19(2): 123-127.
[9] 郝志勇, 刘伟, 夏玮, 闫闯. 基于BP神经网络的吸运风机故障诊断[J]. 工程设计学报, 2012, 19(1): 57-60.
[10] 韩海涛, 马红光, 李飞, 张家良. 基于系统输出频率响应函数的非线性系统研究[J]. 工程设计学报, 2011, 18(5): 373-376.
[11] 郭玉秀, 倪晓红, 王玉田. 基于混沌弱信号检测的轧机故障诊断研究[J]. 工程设计学报, 2011, 18(3): 218-221.
[12] 王成刚, 王学伟, 杨智勇, 曾儒伟. 测试性建模中故障模式概率更新算法研究[J]. 工程设计学报, 2009, 16(4): 252-255.
[13] 王子玲, 许爱强, 牛双诚. 基于多故障假设的诊断策略研究与应用[J]. 工程设计学报, 2009, 16(4): 281-286.
[14] 王 莉, 周新东. 基于Siemens PLC和自校正控制算法的炭素回转窑系统的控制[J]. 工程设计学报, 2008, 15(1): 50-53.
[15] 陈勇, 梁军. 滑动模型MPCA在非线性系统故障监测与诊断中的应用[J]. 工程设计学报, 2003, 10(3): 131-135.