Unsupervised domain-adaptive bearing fault diagnosis method based on simulation data
-
摘要: 针对目前实际诊断任务中难以获得有效的滚动轴承故障数据以及目前诊断模型泛化能力差的问题,文章提出一种基于动力学仿真与无监督领域自适应的故障诊断方法。首先建立滚动轴承动力学仿真模型,获得大量的仿真数据充当源域;然后使用无监督领域自适应的迁移学习故障诊断方法,在全局领域适配的基础上,引入最大最小化分类器差异的对抗学习策略,进一步减小了源域和目标域特征的条件分布差异;最后通过与其他迁移学习方法对比验证所提方法的可行性与优异性。Abstract: Aiming at the current problem that it is difficult to obtain effective rolling bearing fault data in practical diagnostic tasks and the poor generalization ability of the current diagnostic model, a fault diagnosis method based on dynamics simulation and unsupervised domain adaptation is proposed. Firstly, a rolling bearing dynamics simulation model is established, and a large amount of simulation data is obtained to serve as the source domain. Then, an unsupervised domain-adaptive transfer learning fault diagnosis approach is used, which introduces an adversarial learning strategy that maximizes and minimizes classifier differences on basis of global domain adaptation to further reduce the conditional distribution differences between source and target domain features. Finally, the feasibility and excellence of the proposed method is verified by comparing it with other transfer learning methods.
-
Key words:
- fault diagnosis /
- rolling bearing /
- dynamic simulation /
- unsupervised domain adaptation
-
表 1 特征提取网络参数
网络层 核大小、填充、步长 输出尺寸 Input (1024,1) Layer1 [9, 4, 2] (512,16) Layer2 [7, 2, 3] (256,32) Layer3 [5, 2, 2] (128,64) Layer4 [3, 2, 1] (64,128) Layer5 [3, 2, 1] (32,256) AdaptiveAvgPool (4,256) Flatten (1024,) 表 2 分类器参数
分类器 输入维度 输出维度 激活函数 分类器C1 256 256 Tanh 256 4 Softmax 分类器C2 256 256 ReLu 256 4 Softmax 表 3 数据样本
样本 故障模式 标签号 样本长度 样本数量 仿真/数据集 正常 0 1 024 500 内圈故障 1 1 024 500 外圈故障 2 1 024 500 滚动体故障 3 1 024 500 表 4 不同方法平均诊断精度
方法 平均诊断准确率/(%) 准确率标准差/(%) 无迁移 36.68 2.86 DeepCoral 56.42 7.34 DAN 75.17 4.67 DANN 72.67 8.18 DSAN 84.53 6.03 所提方法 87.53 1.62 -
[1] Zhao Z,Li T,Wu J,et al. Deep learning algorithms for rotating machinery intelligent diagnosis:An open source benchmark study[J]. ISA Transactions,2020,107:224-255. doi: 10.1016/j.isatra.2020.08.010 [2] 赵志宏,吴冬冬,窦广鉴,等. 一种基于CNN-BiGRU孪生网络的轴承故障诊断方法[J]. 振动与冲击,2023,42(6):166-171,211. doi: 10.13465/j.cnki.jvs.2023.06.020 [3] 郭俊锋,王淼生,孙磊,等. 基于生成对抗网络的滚动轴承不平衡数据集故障诊断新方法[J]. 计算机集成制造系统,2022,28(9):2825-2835. doi: 10.13196/j.cims.2022.09.015 [4] 张龙,张号,王朝兵,等. 显式动力学驱动的轴承变工况故障诊断方法[J]. 中国机械工程,2023,34(8):982-992. doi: 10.3969/j.issn.1004-132X.2023.08.013 [5] Yang B,Lei Y,Jia F,et al. An intelligent fault diagnosis approach based on transfer learning from laboratory bearings to locomotive bearings[J]. Mechanical Systems and Signal Processing,2019,122:692-706. doi: 10.1016/j.ymssp.2018.12.051 [6] 王业统,吴海威,李美,等. 基于特征的迁移神经网络轴承智能故障诊断[J]. 机械设计与制造,2022(8):80-87. [7] 董韵佳. 基于动力学仿真和迁移学习的滚动轴承故障诊断方法研究[D]. 哈尔滨:哈尔滨工业大学,2019. [8] 罗茂林,郭瑜,伍星. 考虑冲击力的球轴承外圈剥落缺陷双冲击现象动力学建模[J]. 振动与冲击,2019,38(14):48-54. doi: 10.13465/j.cnki.jvs.2019.14.007 [9] 张韬,贾倩,辛月杰. 基于无监督特征对齐的滚动轴承故障诊断[J]. 机械强度,2022,44(3):547-553. [10] Gretton A,Borgwardt K,Rasch M,et al. A kernel method for the two-sample-problem[J]. Advances in Neural Information Processing Systems,2006,19. [11] Hu R,Zhang M,Xiang Z,et al. Guided deep subdomain adaptation network for fault diagnosis of different types of rolling bearings[J]. Journal of Intelligent Manufacturing,2023,34(5):2225-2240. doi: 10.1007/s10845-022-01910-7 [12] Sun B C,Saenko K. Deep coral:correlation alignment for deep domain adaptation[C].Computer Vision–ECCV 2016 Workshops:Amsterdam,2016:443-450. [13] Long M S,Cao Y,Wang J M,et al. Learning transferable features with deep adaptation networks[C].International Conference on Machine Learning,2015,37:97-105. [14] Ganin Y,Lempitsky V. Unsupervised domain adaptation by backpropagation [C].International Conference on Machine Learning,2015,37:1180-1189. [15] Zhu Y C,Zhuang F Z,Wang J D,et al. Deep subdomain adaptation network for image classification[J]. IEEE Transactions on Neral Networks and Learning Systems,2020,32(4):1713-1722.