计算机科学

计算机科学 ›› 2024, Vol. 51 ›› Issue (10): 320-329.doi: 10.11896/jsjkx.230900139

• 人工智能 • 上一篇    下一篇

基于锐度感知增强卷积神经网络的变工况机械故障诊断

范家源1, 徐德胜2, 罗灵鲲1, 胡士强1   

  1. 1 上海交通大学航空航天学院 上海 200240
    2 大型客机集成技术与模拟飞行全国重点实验室 上海 201210
  • 收稿日期:2023-09-25 修回日期:2024-03-12 出版日期:2024-10-15 发布日期:2024-10-11
  • 通讯作者: 胡士强(sqhu@sjtu.edu.cn)
  • 作者简介:(jiayuanf@sjtu.edu.cn)
  • 基金资助:
    国家自然科学基金( 61773262,62006152);中国航空科学基金(2022Z071057002,20142057006)

Mechanical Fault Diagnosis Under Variable Working Conditions Based on Sharpness AwarenessReinforced Convolutional Neural Network

FAN Jiayuan1, XU Desheng2, LUO Lingkun1, HU Shiqiang1   

  1. 1 School of Aeronautics and Astronautics,Shanghai Jiao Tong University,Shanghai 200240,China
    2 State Key Laboratory of Airliner Integration Technology and Flight Simulation,Shanghai 201210,China
  • Received:2023-09-25 Revised:2024-03-12 Online:2024-10-15 Published:2024-10-11
  • About author:FAN Jiayuan,born in 1998,postgra-duate.His main research interests include deep learning and mechanical fault diagnosis.
    HU Shiqiang,born in 1969,Ph.D,professor,Ph.D supervisor.His main research interests include pattern recognition and application and optimization of machine learning on aviation tasks.
  • Supported by:
    National Natural Science Foundation of China (61773262,62006152) and China Aviation Science Foundation (2022Z071057002,20142057006).

PDF (PC)

255

摘要： 传统的深度迁移学习网络从有标签的源域故障数据中学习并迁移知识,完成无标签目标域上的诊断任务,有效解决了机械故障诊断中变工况场景面临的数据特征空间非对称问题。然而其知识迁移模块加剧了深度学习网络结构的复杂性,导致其损失函数的地貌特征比浅层网络复杂得多,优化难度更高。传统方法无法感知损失函数的地貌特征,容易使模型参数陷入参数泛化间隙大的局部最小值处,导致模型泛化性差,诊断精度降低。为了应对这一挑战,研究提出锐度感知增强的卷积神经网络(Sharpness Awareness Reinforced Convolutional Neural Network,SA-CNN),通过感知一定范围内模型损失函数的锐度,联合优化损失函数与其地貌特征的平坦程度,约束模型参数向损失函数锐度降低的方向收敛,进而提升模型的泛化性能。经典机械故障诊断数据集上的实验结果表明,相比传统的深度迁移模型,所提方法在变工况场景下进行跨域机械故障诊断时性能提升显著。

关键词: 轴承故障诊断, 损失函数地貌分析, 迁移学习, 卷积神经网络

Abstract: Traditional deep transfer learning networks have effectively addressed the challenges arising from the asymmetry introduced by cross-domain data distributions in variable operational scenarios.It is achieved by leveraging knowledge learned from labeled fault data and applying it to the task of diagnosing unlabeled fault data collected under varying conditions.However,the inclusion of knowledge transfer modules has added complexity to the deep network's structure,resulting in a more intricate loss landscape.This,in turn,presents challenges for optimization.Traditional methods often struggle to navigate the sharpness of this loss landscape,leading to the model's parameters getting stuck in local minima characterized by high sharpness.This hinders model generalization and reduces accuracy.To tackle this challenge,this paper proposes the sharpness awareness reinforced con-volutional neural network(SA-CNN).This approach involves a joint optimization of the loss function and its flatness by assessing sharpness within a specified range.This process steers the fault diagnosis model parameters away from regions of high sharpness,ultimately improving model generalization.Extensive experiments on established mechanical fault diagnosis datasets demonstrate that,compared to traditional deep transfer learning-based fault diagnosis models,the proposed SA-CNN significantly enhances the performance of bearing fault diagnosis under varying working conditions.

Key words: Bearing fault diagnosis, Loss function landscape analysis, Transfer learning, Convolutional neural network

中图分类号: 

  • TP181
[1]ZHANG Y P,LIU B G,LIU G,et al.Equipment fault diagnosis technology based on data-driven [J].Mechanical Engineering and Automation,2022(2):130-132.
[2]JIAO J,ZHAO M,LIN J,et al.Deep Coupled Dense Convolutional Network with Complementary Data for Intelligent Fault Diagnosis [J].IEEE Transactions on Industrial Electronics,2019,66(12):9858-9867.
[3]XIAO Q H.Survey of mechanical fault diagnosis methods based on machine learning theory [J].Modern manufacturing engineering,2021(7):148-161.
[4]XU P.Research on bearing fault diagnosis method based on neural network [D].Yancheng Institute of Technology,2023.
[5]SUN S.Research on real-time fault diagnosis of motor Bearing based on Deep Learning [D].Daqing:Northeast Petroleum University,2023.
[6]MA B T.Abnormal detection of industrial robot joint healthstate based on current signal [D].Shanghai:Donghua University,2023.
[7]ZHAO Z,ZHANG Q,YU X,et al.Applications of Unsupervised Deep Transfer Learning to Intelligent Fault Diagnosis:A Survey and Comparative Study [J].IEEE Transactions on Instrumentation and Measurement,2021,70:1-28.
[8]LYU C H,CHENG J J,HU Y G,et al.Servo Online Fault Diag-nosis Based on Multi-source Domain Deep Transfer Learning [J].Journal of Ordnance and Equipment Engineering,2022,43(9):60-67.
[9]XU S M,LUO L K,HU J L,et al.Semantic driven attentionnetwork with attribute learning for unsupervised person re-identification [J].Knowledge-Based Systems,2022,252(27):10935.1-10935.13.
[10]LUO L K,CHEN L,HU S,et al.Discriminative and Geometry-Aware Unsupervised Domain Adaptation [J].IEEE Transactions on Cybernetics,2020,50(9):3914-3927.
[11]LUO L K.Transfer Learning and Interactive Image Segmentation [D].Shanghai:Shanghai Jiao Tong University,2019.
[12]TAN M,QUOC V L.EfficientNet:Rethinking Model Scaling for Convolutional Neural Networks [C]//Proceedings of International Conference on Machine Learning.PMLR,2019:6105-6114.
[13]ZHANG C,BENGIO S,HARDT M R,et al.Understandingdeep learning requires rethinkinggeneralization [J].Association for Computing Machinery,2021,64(3):107-115.
[14]KESKAR N S,MUDIGERE D,NOCEDAL J,et al.On Large-Batch Training for Deep Learning:Generalization Gap and Sharp Minima[C]//Proceedings of International Conference on Lear-ning Representations.Palais des Congrès Neptune:ICLR,2017:1-16.
[15]FORET P,KLEINER A,MOBAHI H,et al.Sharpness-aware minimization for efficiently improving generalization[C]//Proceedings of International Conference on Learning Representations.2021:1-18.
[16]WANG J.Research on Text Representation and Classificationwith Deep Learning [D].Beijing:Beijing University of Posts and Telecommunications,2019.
[17]KAVIANPOUR M,GHORVEI M,KAVIANPOUR P,et al.An Intelligent Gearbox Fault Diagnosis under Different Operating Conditions using Adversarial Domain Adaptation[C]//Procee-dings of International Conference on Control,Instrumentation and Automation.Tehran:ICCIA,2022:1-6.
[18]WANG Q,TAAL C,FINK O.Integrating Expert Knowledgewith Domain Adaptation for Unsupervised Fault Diagnosis [J].IEEE Transactions on Instrumentation and Measurement,2022,71:1-12.
[19]NESTEROV Y E.A method for solving the convex programming problem with convergence rate O(1/k2) [J].Dokl.Akad.Nauk SSSR,1983,269:543-547.
[20]KINGMA D P,BA J.Adam:AMethod for Stochastic Optimization[C]//Proceedings of 3rd International Conference on Lear-ning Representations.2015:1-15.
[21]DZIUGAITE G,ROY D M.Computing nonvacuous generalization bounds for deep(stochastic) neural networks with many more parameters than training data [C]//Proceedings of the Thirty-Third Conference on Uncertainty in Artificial Intelligence.2017:1-14.
[22]NEYSHABUR B,MCALLESTER D,SREBRO N,et al.Exploring generalization in deep learning[C]//Proceedings of 31st Conference on Neural Information Processing Systems.2019:5947-5956.
[23]CHEN X,HSIEH C,GONG B.When vision transformers out-perform resnets without pretraining or strong data augmentations[C]//Proceedings of International Conference on Learning Representations.2022:1-20.
[24]ABBA S,MOMI N,QUAN X,et al.Sharp-MAML:Sharpness-Aware Model-Agnostic Meta Learning[C]//Proceedings of International Conference on Machine Learning.2022:10-32.
[25]BAHRI D,MOBAHI H,YI T.Sharpness-aware minimizationimproves language model generalization[C]//Annual Meeting of the Association for Computational Linguistics.2022:1-12.
[26]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,2015,64:100-131.
[27]SHAO S Y,STEPHEN M M.Highly accurate machine fault diagnosis using deep transfer learning [J].IEEE Transactions on Industrial Informatics,2019,15(4):2446-2455.
[28]ZHANG Y,REN Z,ZHOU S.A new deep convolutional domain adaptation network for bearing fault diagnosis under different working conditions [J].Shock and Vibration,2020,2020:1-14.
[29]JIAO J,ZHAO M,LIN J,et al.Residual joint adaptation adversarial network for intelligent transfer fault diagnosis [J].Mechanical Systems and Signal Processing.2020,145:106962.
[30]GANIN Y,USTINOVA E,AJAKAN H,et al.Domain-adversarial training of neural networks [J].The Journalof Machine Learning Research,2016,17(1):2096-2030.
[31]SUN B,SAENKO K.Deep coral:Correlation alignment for deep domain adaptation[C]//Proceedings of European Conference on Computer Vision.Switzerland:Springer International Publi-shing,2016:443-445.
[32]LONG M,CAO Z,WANG J,et al.Conditional adversarial do-main adaptation[C]//Proceedings of Advances in Neural Information Processing.Burlington:Morgan Kaufmann,2018:1640-1650.
[33]RAUBER P E,FALCAO A X,TELEA A C.Visualizing time-dependent data using dynamic t-SNE[C]//Proceedings of the Eurographics/IEEE VGTC Conference on Visualization:Short Papers.2016:1-5.
[34]LI H,XU Z,TAYLOR G,et al.Visualizing the loss landscape of neural nets[C]//Advances in Neural Information Processing Systems.Burlington:Morgan Kaufmann,2018.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市渝北区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发