Fault Diagnosis of Multiple Railway High Speed Train Bogies Based on Federated Learning
-
摘要:
单一线路高速列车转向架缺少足量故障数据特征,导致故障诊断模型泛化能力有限,为实现诊断多条线路高速列车的转向架故障,提出一种基于联邦学习的转向架全局故障诊断方法. 针对每条线路各自的转向架振动信号,在本地使用多尺度卷积融合算法,提取不同尺度下的故障特征并融合,在本地建立局部转向架故障诊断模型;在不泄露数据隐私的前提下,所有线路的故障诊断模型通过第三方聚合,调整模型参数权重,对故障诊断模型进行优化,最终实现多方联合训练转向架全局故障诊断模型. 实验表明:在联邦学习框架下,转向架全局故障诊断模型不仅对参与联邦建模的线路转向架故障诊断准确率达到93%以上,而且对于未参与联邦建模的线路转向架故障诊断率也可达到75%以上,给轨道交通中的“数据孤岛”问题提供了一种切实可行的方案.
Abstract:To solve the problem of limited generalization ability of fault diagnosis model caused by the lack of sufficient fault data characteristics of single railway high-speed train bogie, and to realize the diagnosis of bogie faults of multiple railway high-speed trains, a global bogie fault diagnosis method based on federated learning is proposed in this work. Firstly, according to the bogie vibration signals of each railway, the multi-scale convolution fusion algorithm is conducted locally to extract and fuse the fault features at different scales, and the bogie fault diagnosis model is established locally. On the premise of not divulging data privacy, the fault diagnosis models of all railways are aggregated by the third party, the weights of model parameter are adjusted, the fault diagnosis models are optimized, and finally the global fault diagnosis model of bogie is jointly trained by multiple railways. The experiments show that under the federated learning framework, the fault diagnosis accuracy of the global bogie fault diagnosis model is reach more than 93% for the railway participating in federated modeling, and more than 75% for the railway not participating in, which provides a practical scheme for the ‘data island’ problem in railway transportation.
-
Key words:
- federated learning /
- fault diagnosis /
- bogie /
- high speed train
-
表 1 转向架本地故障诊断模型结构及参数
Table 1. Structure and parameters of local bogie fault diagnosis model
结构类型 输入尺寸 卷积核尺寸 步长/步 通道数/个 卷积层 40000 × 1 6 1 16 卷积层 40000 × 16 6 1 16 最大池化 40000 × 16 2 卷积层 20000 × 16 6 1 16 多尺度卷积 20000 × 16 3/4/5 1 64 卷积层 20000 × 64 3/3/3 1 128 卷积拼接 20000 × 128 卷积层 60000 × 128 2 1 128 Dropout 60000 × 128 卷积层 60000 × 128 2 1 256 全局平均池化 60000 × 256 60000 1 1 全连接层 256 × 1 Softmax 7 表 2 转向架工况
Table 2. Operation status of bogie
运行工况 标签 正常运行 0 LD 故障 1 AD 故障 2 AS 故障 3 AD + LD 故障 4 AS + LD 故障 5 AS + AD 故障 6 表 3 转向架故障诊断准确率
Table 3. Accuracy of bogie fault diagnosis
% 实验线路 训练模型 标签 平均 0 1 2 3 4 5 6 武广线 1D-CNN 87.7 85.4 90.4 77.8 83.3 82.2 87.6 84.9 SecureBoost 95.3 95.7 96.6 87.5 92.4 86.9 88.5 91.8 Multi-1D-CNN 98.2 96.8 98.4 94.5 96.2 94.9 99.1 96.9 郑西线 1D-CNN 81.2 83.3 84.5 73.2 76.3 75.4 81.7 79.4 SecureBoost 92.7 90.4 90.5 74.6 93.2 84.3 82.4 86.9 Multi-1D-CNN 94.3 95.5 96.3 87.4 94.8 91.3 95.5 93.6 京津线 1D-CNN 81.7 83.4 84.4 75.8 73.3 74.2 81.6 79.2 SecureBoost 93.3 93.7 95.6 76.5 89.4 84.9 81.5 87.8 Multi-1D-CNN 98.5 97.2 97.3 93.4 97.6 95.2 97.7 96.7 表 4 泛化能力检测
Table 4. Generalization ability test
泛化实验 参与线路 检测线路 1 武广 胶济 2 武广 + 郑西 胶济 3 武广 + 郑西 + 京津 胶济 4 武广 + 郑西 + 京津 + 金山 胶济 -
[1] WEI X C, CHEN Y, LU C, et al. Acoustic emission source localization method for high-speed train bogie[J]. Multimedia Tools and Applications, 2020, 79(21/22): 14933-14949. [2] 池毓敢,林建辉,李艳萍,等. 二系横向减振器阻尼系数对车辆横向振动影响的仿真研究[J]. 铁道车辆,2014,52(4): 15-16. doi: 10.3969/j.issn.1002-7602.2014.04.005CHI Yugan, LIN Jianhui, LI Yanping, et al. Simulation research on the effect of damping coefficient of the secondary lateral dampers on lateral vibration of vehicles[J]. Rolling Stock, 2014, 52(4): 15-16. doi: 10.3969/j.issn.1002-7602.2014.04.005 [3] 谢树强,王斌杰,王文静,等. 基于动应力的地铁构架疲劳损伤与疲劳寿命计算[J]. 机械工程学报,2022,58(4): 183-190.XIE Shuqiang, WANG Binjie, WANG Wenjing, et al. Calculation for fatigue damage and fatigue life of metro bogie based on dynamic stress[J]. Journal of Mechanical Engineering, 2022, 58(4): 183-190. [4] 张卫华,李艳,宋冬利. 高速列车运动稳定性设计方法研究[J]. 西南交通大学学报,2013,48(1): 1-9.ZHANG Weihua, LI Yan, SONG Dongli. Design methods for motion stability of high-speed trains[J]. Journal of Southwest Jiaotong University, 2013, 48(1): 1-9. [5] MAO Z H, WANG Y, JIANG B, et al. Fault diagnosis for a class of active suspension systems with dynamic actuators’ faults[J]. International Journal of Control, Automation and Systems, 2016, 14(5): 1160-1172. doi: 10.1007/s12555-014-0552-z [6] LI P, GOODALL R, WESTON P, et al. Estimation of railway vehicle suspension parameters for condition monitoring[J]. Control Engineering Practice, 2007, 15(1): 43-55. doi: 10.1016/j.conengprac.2006.02.021 [7] QIN N, LIANG K W, HUANG D Q, et al. Multiple convolutional recurrent neural networks for fault identification and performance degradation evaluation of high-speed train bogie[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 31(12): 5363-5376. doi: 10.1109/TNNLS.2020.2966744 [8] WEI X K, JIA L M, GUO K, et al. On fault isolation for rail vehicle suspension systems[J]. Vehicle System Dynamics, 2014, 52(6): 847-873. doi: 10.1080/00423114.2014.904904 [9] KOU L L, QIN Y, ZHAO X J, et al. A multi-dimension end-to-end CNN model for rotating devices fault diagnosis on high-speed train bogie[J]. IEEE Transactions on Vehicular Technology, 2020, 69(3): 2513-2524. doi: 10.1109/TVT.2019.2955221 [10] 刘建伟,刘媛,罗雄麟. 深度学习研究进展[J]. 计算机应用研究,2014,31(7): 1921-1930,1942.LIU Jianwei, LIU Yuan, LUO Xionglin. Research and development on deep learning[J]. Application Research of Computers, 2014, 31(7): 1921-1930,1942. [11] HATCHER W G, YU W. A survey of deep learning: platforms, applications and emerging research trends[J]. IEEE Access, 2018, 6: 24411-24432. doi: 10.1109/ACCESS.2018.2830661 [12] YU L K, ALBELAIHI R, SUN X, et al. Jointly optimizing client selection and resource management in wireless federated learning for Internet of Things[J]. IEEE Internet of Things Journal, 2022, 9(6): 4385-4395. doi: 10.1109/JIOT.2021.3103715 [13] LI X, ZHANG W. Deep learning-based partial domain adaptation method on intelligent machinery fault diagnostics[J]. IEEE Transactions on Industrial Electronics, 2021, 68(5): 4351-4361. doi: 10.1109/TIE.2020.2984968 [14] LIU Y, KANG Y, XING C P, et al. A secure federated transfer learning framework[J]. IEEE Intelligent Systems, 2020, 35(4): 70-82. doi: 10.1109/MIS.2020.2988525 [15] SEMMA A, HANNAD Y, EL KETTANI M E Y. Impact of the CNN patch size in the writer identification[C]//Proceedings of Networking, Intelligent Systems and Security. Singapore: Springer, 2022: 103-114. [16] ZHANG L B, CAI J, PENG F, et al. MSA-CNN: face morphing detection via a multiple scales attention convolutional neural network[C]//The 20th International Workshop on Digital Forensics and Watermarking (IWDW). Cham: Springer, 2022: 17-31. [17] XU S Z, ADELI E, CHENG J Z, et al. Mammographic mass segmentation using multichannel and multiscale fully convolutional networks[J]. International Journal of Imaging Systems and Technology, 2020, 30(4): 1095-1107. doi: 10.1002/ima.22423 [18] ZHAO Z C, XIA J J, FAN L S, et al. System optimization of federated learning networks with a constrained latency[J]. IEEE Transactions on Vehicular Technology, 2022, 71(1): 1095-1100. doi: 10.1109/TVT.2021.3128559 [19] LIU A, YU Q Y, XIA B M, et al. Privacy-preserving design of smart products through federated learning[J]. CIRP Annals: Manufacturing Technology, 2021, 70(1): 103-106. doi: 10.1016/j.cirp.2021.04.022 [20] HU Y Q, HUA Y, LIU W Y, et al. Reward shaping based federated reinforcement learning[J]. IEEE Access, 2021, 9: 67259-67267. doi: 10.1109/ACCESS.2021.3074221