Multi-scale convolution intra-class transfer learning for train bearing fault diagnosis
-
摘要: 考虑变工况下列车轴承振动数据分布不一致情况下, 传统深度学习诊断模型的泛化能力下降, 提出了一种多尺度卷积类内自适应的深度迁移学习模型; 模型利用改进的ResNet-50网络分析振动数据的频谱, 得到了中间层次特征, 构造了多尺度特征提取器, 从不同尺度处理中间层次特征得到高层次特征; 将高层次特征作为分类器的输入, 同时计算了伪标签以缩短在不同工作条件下收集的振动信号的条件分布距离来进行类内匹配; 为了验证模型的通用性和优越性, 将提出的模型分别用于列车轮对轴承数据集和凯斯西储数据集的多个工况进行试验验证和分析。研究结果表明: 通过对齐不同域中同一类样本的高层次特征作为分类器的输入, 提出的模型获得了更为理想的故障诊断精度; 在列车轴承6个变工况诊断实例中, 平均诊断精度为90.75%, 与传统深度学习模型相比, 模型诊断精度平均提高了约10%, 召回率为0.927;在凯斯西储数据集的12个变工况诊断实例中, 模型平均诊断精度达99.97%, 比传统模型提高约10%。可见, 利用伪标签减小了不同域之间的条件分布差异, 很好地处理了源域和目标域数据分布不一致的问题; 多尺度特征提取器能从不同尺度对齐样本的高层次特征, 增强了模型的泛化性与鲁棒性, 是解决变工况列车轴承故障诊断问题的一种有效模型。Abstract: Considering the inconsistent distribution of bearing vibration data collected under different working conditions, the generalization ability of traditional deep learning model decreases. A multi-scale convolution intra-class adaptive deep transfer learning model was proposed. The spectrum of vibration data was analyzed using the modified ResNet-50. The middle-level features were obtained. A multi-scale feature extractor was developed, the middle-level features were processed, and the high-level features were generated. The high-level features were used as the inputs of classifier. The pseudo-labels were computed, and then the conditional distribution distances of vibration data collected under variable working conditions reduced for the intra-class adaptation. To verify the generality and superiority of model, the proposed method was employed to analyze a train wheelset bearing dataset and the Case Western Reserve University dataset under variable working conditions. Analysis result indicates that the high-level features of samples with the same label in different domains are properly aligned. More satisfactory fault diagnosis accuracies are obtained by the proposed model. In six fault diagnosis cases of train bearing under variable working conditions, the average diagnosis accuracy of the proposed model is 90.75%, approximately 10% higher than those of traditional deep learning models, while the recall rate is 0.927. In twelve fault diagnosis cases of Case Western Reserve University dataset under variable working conditions, the average accuracy obtained by the proposed model is 99.97%, approximately 10% higher than those of traditional models. The conditional distribution discrepancy between different domains reduces by using the pseudo-labels. The inconsistency problem of data distribution of source domain and target domain is properly addressed. The high-level features of samples from different scales can be aligned by the multi-scale feature learner. The generalization and robustness of the model largely increase. In conclusion, the proposed model has a high potential for the train bearing fault diagnosis under variable working conditions.
-
图 5 模型M4对NJ(P)3226X1轴承诊断结果混淆矩阵
Figure 5. Confusion matrixes of NJ(P)3226X1 bearing diagnosis results by model M4
图 6 迁移任务1 t→2 t高层次特征可视化结果
Figure 6. High-level feature visualization result through t-SNE from transfer task 1 t→2 t
图 7 迁移任务1t→2t特征可视化结果
Figure 7. Feature visualization result through t-SNE from transfer task 1 t→2 t
图 11 迁移任务0→735 W高层次特征可视化结果
Figure 11. High-level feature visualization result through t-SNE from transfer task 0→735 W
图 12 迁移任务0→735 W特征可视化结果
Figure 12. Feature visualization result through t-SNE from transfer task 0→735 W
图 13 任务735 W→0中M1和M4的测试精度和训练误差曲线
Figure 13. Testing accuracy and training loss curves of M1 and M4 from task 735 W→0
表 1 每个域中NJ(P)3226X1轴承状态描述
Table 1. Description of states for NJ(P)3226X1 bearing in each domain
状态 标签 符号表示 样本个数 滚子故障 0 BF 500 内圈故障 1 IF 500 正常 2 NO 500 外圈故障 3 OF 500 
下载: 导出CSV
表 2 M1、M2的模型结构
Table 2. Model structures of M1 and M2
层名 激活函数 参数结构 输入 卷积1 ReLU 5×5×16 池化 步长2 卷积2 ReLU 7×7×32 池化 步长2 一维化 完全连接层1 ReLU 288×100 完全连接层2 ReLU 100×100 完全连接层3 Softmax 100×4
下载: 导出CSV
表 3 M3、M4中ResNet-50的模型结构
Table 3. Model structures of ResNet-50 in M3 and M4
层名 输出尺寸 通道数、核尺寸 输入 3×224×224 卷积1 64×112×112 64×7×7, 步长2 批归一化、ReLU 64×112×112 最大池化 64×56×56 64×3×3, 步长2 残差块1 256×56×56 残差块2 512×28×28 残差块3 1 024×14×14 残差块4 2 048×7×7 ReLU 2 048×7×7 全局平均池化 2 048×1 2 048×7×7 分类层 1 000
下载: 导出CSV
表 4 四种模型的NJ(P)3226X1轴承诊断精度
Table 4. Diagnosis accuracies of four models forNJ(P)3226X1 bearing
迁移任务/t M1 M2 M3 M4 1→2 66.70 77.85 81.35 92.85 1→3 63.50 74.15 85.30 89.65 2→1 48.85 55.25 73.45 93.25 2→3 73.50 79.80 79.95 92.70 3→1 48.50 53.45 67.85 88.35 3→2 68.65 78.45 77.10 87.70
下载: 导出CSV
表 5 迁移任务2 t→3 t中M4的分类结果
Table 5. Classification result of M4 in transfer task 2 t →3 t
状态 精度/% 召回率 f1 样本数 滚子故障 89.69 0.800 0.845 7 500 内圈故障 82.85 0.908 0.866 4 500 正常 100.00 1.000 1.000 0 500 外圈故障 98.81 1.000 0.994 0 500 均值 92.84 0.927 0.926 5
下载: 导出CSV
表 6 每个域中CWRU轴承7种状态描述
Table 6. Description of 7 states for CWRU bearing in each domain
故障尺寸/mm 状态 标签 样本数 符号 正常 0 200 NO 0.18 内圈故障 1 200 IF07 0.18 滚子故障 2 200 BF07 0.18 外圈故障 3 200 OF07 0.36 内圈故障 4 200 IF14 0.36 滚子故障 5 200 BF14 0.36 外圈故障 6 200 OF14
下载: 导出CSV
表 7 四种模型CWRU轴承的诊断精度
Table 7. Diagnosis accuracies of four models for CWRU bearing
迁移任务/W M1 M2 M3 M4 0→735 75.86 81.21 89.57 100.00 0→1 470 71.93 84.50 88.36 100.00 0→2 206 76.36 75.65 91.57 99.93 735→0 71.43 85.71 89.79 99.93 735→1 470 81.71 87.07 91.29 100.00 735→2 206 71.29 75.86 82.64 99.86 1 470→0 83.93 85.07 87.35 99.93 1 470→735 81.93 85.79 85.71 100.00 1 470→2 206 71.36 79.50 91.00 100.00 2 206→0 88.64 91.21 84.36 99.93 2 206→735 83.36 84.71 81.00 100.00 2 206→1 470 84.50 86.57 92.36 100.00
下载: 导出CSV
-
[1] 谢国, 王竹欣, 黑新宏, 等. 面向热轴故障的高速列车轴温阈值预测模型[J]. 交通运输工程学报, 2018, 18(3): 129-137. doi: 10.3969/j.issn.1671-1637.2018.03.014XIE Guo, WANG Zhu-xin, HEI Xin-hong, et al. Axle temperature threshold prediction model of high-speed train for hot axle fault[J]. Journal of Traffic and Transportation Engineering, 2018, 18(3): 129-137. (in Chinese). doi: 10.3969/j.issn.1671-1637.2018.03.014 [2] YAN Ru-qiang, SHEN Fei, SUN Chuang, et al. Knowledge transfer for rotary machine fault diagnosis[J]. IEEE Sensors Journal, 2020, 20(15): 8374-8393. doi: 10.1109/JSEN.2019.2949057 [3] WU Zheng-hong, JIANG Hong-kai, ZHAO Ke, et al. An adaptive deep transfer learning method for bearing fault diagnosis[J]. Measurement, 2019, DOI: 10.1016/j.measurement.2019.107227. [4] SHAO Si-yu, MCALEER S, YAN Ru-qiang, et al. Highly accurate machine fault diagnosis using deep transfer learning[J]. IEEE Transactions on Industrial Informatics, 2018, 15: 2446-2455. [5] FENG Zhi-peng, CHEN Xiao-wang. Adaptive iterative generalized demodulation for nonstationary complex signal analysis: principle and application in rotating machinery fault diagnosis[J]. Mechanical Systems and Signal Processing, 2018, 110: 1-27. doi: 10.1016/j.ymssp.2018.03.004 [6] JIANG Xing-xing, WANG Jun, SHI Juan-juan, et al. A coarse-to-fine decomposing strategy of VMD for extraction of weak repetitive transients in fault diagnosis of rotating machines[J]. Mechanical Systems and Signal Processing, 2019, 116: 668-692. doi: 10.1016/j.ymssp.2018.07.014 [7] ISLAM M M M, KIM J M. Reliable multiple combined fault diagnosis of bearings using heterogeneous feature models and multiclass support vector machines[J]. Reliability Engineering and System Safety, 2019, 184: 55-66. doi: 10.1016/j.ress.2018.02.012 [8] 刘强, 詹志强, 王硕, 等. 数据驱动的高速列车轴承多模态运行监控与故障诊断[J]. 中国科学: 信息科学, 2020, 50(4): 527-539. https://www.cnki.com.cn/Article/CJFDTOTAL-PZKX202004006.htmLIU Qiang, ZHAN Zhi-qiang, WANG Shuo, et al. Data-driven multimodal operation monitoring and fault diagnosis of high-speed train bearings[J]. Scientia Sinica Informationis, 2020, 50(4): 527-539. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-PZKX202004006.htm [9] KUNCAN M, KAPLAN K, MINAZ M R, et al. A novel feature extraction method for bearing fault classification with one dimensional ternary patterns[J]. ISA Transactions, 2019, 100: 346-357. [10] 周生通, 朱经纬, 周新建, 等. 组合载荷作用下动车牵引电机转子系统弯扭耦合振动特性[J]. 交通运输工程学报, 2020, 20(1): 159-170. doi: 10.19818/j.cnki.1671-1637.2020.01.013ZHOU Sheng-tong, ZHU Jing-wei, ZHOU Xin-jian, et al. Bending-torsional coupling vibration characteristics of EMU traction motor rotor system under combined loads[J]. Journal of Traffic and Transportation Engineering, 2020, 20(1): 159-170. (in Chinese). doi: 10.19818/j.cnki.1671-1637.2020.01.013 [11] 郭俊超, 甄冬, 孟召宗, 等. 基于WAEEMD和MSB的滚动轴承故障特征提取[J]. 中国机械工程, https://kns.cnki.net/kcms/detail/42.1294. TH. 20200810.1640. 024. html. GUO Jun-chao, ZHEN Dong, MENG Zhao-zong, et al. Feature extraction of rolling element bearing based on weighted average ensemble empirical mode decomposition and modulation bispectrum analysis[J]. China Mechanical Engineering, https://kns.cnki.net/kcms/detail/42.1294.TH.20200810.1640.024.html. (inChinese). [12] 雷亚国, 杨彬, 杜兆钧, 等. 大数据下机械装备故障的深度迁移诊断方法[J]. 机械工程学报, 2019, 55(7): 1-8. https://www.cnki.com.cn/Article/CJFDTOTAL-JXXB201907001.htmLEI Ya-guo, YANG Bin, DU Zhao-jun, et al. Deep transfer diagnosis method for machinery in big data era[J]. Journal of Mechanical Engineering, 2019, 55(7): 1-8. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-JXXB201907001.htm [13] 严文超, 王伟奇, 黄蓉. 基于RSSD和小波变换的滚动轴承故障诊断[J]. 武汉工程大学学报, 2019, 41(4): 399-404. https://www.cnki.com.cn/Article/CJFDTOTAL-WHHG201904018.htmYAN Wen-chao, WANG Wei-qi, HUANG Rong. Rolling bearing fault diagnosis method based on resonance-based sparse signal decomposition and wavelet transform[J]. Journal of Wuhan Institute of Technology, 2019, 41(4): 399-404. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-WHHG201904018.htm [14] MAQSOOD A, OSLEBO D, CORZINE K, et al. STFT cluster analysis for DC pulsed load monitoring and fault detection on naval shipboard power systems[J]. IEEE Transactions on Transportation Electrification, 2020, 6(2): 821-831. doi: 10.1109/TTE.2020.2981880 [15] MAURYA S, SINGH V, VERMA N K. Condition monitoring of machines using fused features from EMD-based local energy with DNN[J]. IEEE Sensors Journal, 2020, 20(15): 8316-8327. doi: 10.1109/JSEN.2019.2927754 [16] 王海龙, 夏筱筠, 孙维堂. 基于EMD与卷积神经网络的滚动轴承故障诊断[J]. 组合机床与自动化加工技术, 2019(10): 46-48, 52. https://www.cnki.com.cn/Article/CJFDTOTAL-ZHJC201910012.htmWANG Hai-long, XIA Xiao-jun, SUN Wei-tang. Rolling bearing fault diagnosis based on EMD and convolutional neural network[J]. Modular Machine Tool and Automatic Manufacturing Technique, 2019(10): 46-48, 52. (in Chinese). https://www.cnki.com.cn/Article/CJFDTOTAL-ZHJC201910012.htm [17] LI Ning, HUANG Wei-guo, GUO Wen-jun, et al. Multiple enhanced sparse decomposition for gearbox compound fault diagnosis[J]. IEEE Transactions on Instrumentation Measurement, 2020, 69(3): 770-781. doi: 10.1109/TIM.2019.2905043 [18] QIN Yi. A new family of model-based impulsive wavelets and their sparse representation for rolling bearing fault diagnosis[J]. IEEE Transactions on Industrial Electronics, 2018, 65(3): 2716-2726. doi: 10.1109/TIE.2017.2736510 [19] HE Qing-bo, WU En-hao, PAN Yuan-yuan. Multi-scale stochastic resonance spectrogram for fault diagnosis of rolling element bearings[J]. Journal of Sound and Vibration, 2018, 420: 174-184. doi: 10.1016/j.jsv.2018.01.001 [20] JIA Feng, LEI Ya-guo, GUO Liang, et al. A neural network constructed by deep learning technique and its application to intelligent fault diagnosis of machines[J]. Neurocomputing, 2018, 272: 619-628. doi: 10.1016/j.neucom.2017.07.032 [21] LI Ke, XIONG Meng, LI Fu-cai, et al. A novel fault diagnosis algorithm for rotating machinery based on a sparsity and neighborhood preserving deep extreme learning machine[J]. Neurocomputing, 2019, 350: 261-270. doi: 10.1016/j.neucom.2019.03.084 [22] ZHU Jun, CHEN Nan, PENG Wei-wen. Estimation of bearing remaining useful life based on multiscale convolutional neural network[J]. IEEE Transactions on Industrial Electronics, 2019, 66(4): 3208-3216. doi: 10.1109/TIE.2018.2844856 [23] PAN S J, YANG Qiang. A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2009, 22: 1345-1359. [24] HAN Te, LIU Chao, YANG Wen-guang, et al. Deep transfer network with joint distribution adaptation: a new intelligent fault diagnosis framework for industry application[J]. ISA Transactions, 2020, 97: 269-281. doi: 10.1016/j.isatra.2019.08.012 [25] XU Yong-hui, PAN S J, XIONG Hui, et al. A unified framework for metric transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2017, 29(6): 1158-1171. doi: 10.1109/TKDE.2017.2669193 [26] WEISS K, KHOSHGOFTAAR T M, WANG Ding-ding. A survey of transfer learning[J]. Journal of Big Data, 2016, DOI: 10.1186/s40537-016-0043-6. [27] CAO Pei, ZHANG Sheng-li, TANG Jiong. Preprocessing-free gear fault diagnosis using small datasets with deep convolutional neural network-based transfer learning[J]. IEEE Access, 2018, 6: 26241-26253. doi: 10.1109/ACCESS.2018.2837621 [28] GUO Liang, LEI Ya-guo, XING Sai-bo, et al. Deep convolutional transfer learning network: a new method for intelligent fault diagnosis of machines with unlabeled data[J]. IEEE Transactions on Industrial Electronics, 2018, 66(9): 7316-7325. [29] WEN Long, GAO Liang, LI Xin-yu. A new deep transfer learning based on sparse auto-encoder for fault diagnosis[J]. IEEE Transactions on Systems, Man, and Cybernetics Systems, 2019, 49(1): 136-144. doi: 10.1109/TSMC.2017.2754287 [30] YANG Bin, LEI Ya-guo, JIA Feng, 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 [31] GRETTON A, BORGWARDT K M, RASCH M J, et al. A kernel two-sample test[J]. Journal of Machine Learning Research, 2012, 13: 723-773. [32] HE Kai-ming, ZHANG Xiang-yu, REN Shao-qing, et al. Deep residual learning for image recognition[C]//IEEE. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2016: 770-778. [33] PANS J, TSANG I W, KWOK J T, et al. Domain adaptation via transfer component analysis[J]. IEEE Transactions on Neural Networks, 2010, 22(2): 199-210. [34] WANG Jin-dong, CHEN Yi-qiang, HU Li-sha, et al. Stratified transfer learning for cross-domain activity recognition[C]//IEEE. 2018 IEEE International Conference on Pervasive Computing and Communications (PerCom). New York: IEEE, 2018: 1-10. [35] JIAO Jin-yang, ZHAO Ming, LIN Jing, et al. Residual joint adaptation adversarial network for intelligent transfer fault diagnosis[J]. Mechanical Systems and Signal Processing, 2020, 145: 106962-1-14. doi: 10.1016/j.ymssp.2020.106962 [36] ZHU Jun, CHEN Nan, SHEN Chang-qing. A new deep transfer learning method for bearing fault diagnosis under different working conditions[J]. IEEE Sensors Journal, 2020, 20(15): 8394-8402. doi: 10.1109/JSEN.2019.2936932 [37] VAN DERMAATEN L, HINTON G. Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008, 9: 2579-2605. -
点击查看大图
图(13) / 表(7)
计量
- 文章访问数: 801
- HTML全文浏览量: 161
- PDF下载量: 2341
- 被引次数: 0
