Early fault detection of rolling bearings based on self-supervised deep one classification
-
摘要:
针对当前因滚动轴承故障数据难以获取，而导致智能故障诊断模型难以训练的问题，提出一种仅依靠正常类样本进行训练的自监督深度一类分类方法用于滚动轴承早期故障预警。该方法在深度一类分类模型的基础上引入了多任务和自监督机制。采用深度残差网络提取输入信号的深层特征，将所提特征分别作为多个子任务的输入，其中支持向量描述（SVDD）分类子任务的输出结果，作为其余子任务的监督标签，通过所建立的联合损失函数，仅依靠正常类样本即可完成模型的自监督学习。在将所提方法用于滚动轴承故障预警时，先对全寿命周期的振动加速度信号进行频带分解和包络分析，将所得不同频段内的信号进行二维编码后作为网络的输入。在两组实际的滚动轴承故障数据集上对所提的方法进行试验验证。验证结果表明：所提深度一类分类方法在进行故障预警时准确率达到99%以上，充分表明该方法具有很高的故障预警和异常检测能力。
Abstract:At present the intelligent fault diagnosis model is difficult to train due to the difficulty in obtaining the fault data of rolling bearings. A self-supervised deep one-class classification method was proposed for early fault warning of rolling bearings based on the training of normal class samples. Based on the deep one-class classification model, multi-task and self-supervision mechanisms were introduced. Input signal was extracted with the use of the depth of the deep residual network characteristics, and the proposed features were taken respectively as input of more child tasks, including support vector description (SVDD) as the output results of the classification subtasks supervision and label as the rest of the subtasks; through the joint loss function, only relying on normal class samples can complete the model of supervised learning. When the proposed method was applied to the fault warning of rolling bearings, the vibration acceleration signals of the whole life cycle were decomposed by frequency band, and the signals in different frequency bands were encoded as the input of the network. The proposed method was validated on two actual rolling bearing fault data sets. The verification results showed that the accuracy of the proposed depth classification method reached more than 99% in fault warning, which fully indicated that the proposed method has a high ability of fault warning and anomaly detection.
-
Key words:
- deep one classification /
- multitasking /
- self-supervised learning /
- rolling bearing /
- fault diagnosis
-
图 2 自监督深度一类分类
Figure 2. Self supervised deep support vector data description （SSDSVDD）
图 6 滚动轴承振动数据预处理流程图
Figure 6. Flow chart of rolling bearing vibration data preprocessing
表 1 混淆矩阵
Table 1. Confusion matrix
混淆矩阵 真实类别 0 1 预测
类别0
（负类）TN （真反例） FN （假反例） 1
（正类）FP （假正例） TP （真正例） 
下载: 导出CSV
表 2 在Mnist和Cifar-10两种数据集上的对比结果
Table 2. Comparison results of Mnist and CIFAR-10 data sets
正类标签 DSVDD SVDD DCAE KDE ANOGAN OC-NN SSDSVDD 0 98.0±0.7 98.6±0 97.6±0.7 97.1±0 96.6±1.3 97.6±1.7 98.5±1.2 1 99.7±0.1 99.5±0 98.3±0.6 98.9±0 99.6±0.6 99.5±0 99.7±0.1 2 91.7±0.8 82.5±0.1 85.4±2.4 79.0±0 85.0±2.9 97.3±2.1 97.4±1.6 3 91.9±1.5 88.0±0 86.7±0.9 86.2±0 88.7±2.1 86.5±3.9 92.3±0.8 4 94.9±0.8 94.9±0 86.5±2.0 87.9±0 89.4±1.3 93.2±2.4 93.8±0.5 5 88.5±0.9 77.1±0 78.2±2.7 73.8±0 88.3±2.9 86.5±3.3 91.1±1.3 6 98.3±0.5 96.5±0 94.6±0.5 87.6±0 94.7±2.7 97.1±1.4 98.6±1.3 7 94.6±0.9 93.7±0 92.3±1.0 91.4±0 93.5±1.8 93.6±2.1 95.2±2.1 8 93.9±1.6 88.9±0 86.5±1.6 79.2±0 84.9±2.1 88.5±4.7 92.4±0.9 9 96.5±0.3 93.1±0 90.4±1.8 88.2±0 92.4±1.1 93.5±3.3 96.1±1.9 飞机 61.7±4.1 61.6±0.9 59.1±5.1 61.2±0 67.1±2.5 60.4±1.9 73.5±0.3 汽车 65.9±2.1 63.8±0.6 57.4±2.9 64.0±0 54.7±3.4 62.0±2.0 72.1±0.6 鸟 50.8±0.8 50.0±0.5 48.9±2.4 50.0±0 52.9±3.0 63.7±1.4 62.1±0.1 猫 59.1±1.4 55.9±1.3 58.4±1.2 56.4±0 54.5±1.9 53.6±2.1 70.0±0.3 鹿 60.9±1.1 66.0±0.7 54.0±1.3 66.2±0 65.1±3.2 67.4±1.7 67.7±0.2 狗 65.7±2.5 62.4±0.8 62.2±1.8 62.4±0 60.3±2.6 56.1±2.1 69.5±0.1 青蛙 67.7±2.6 74.7±0.3 51.2±5.2 74.9±0 58.5±1.4 63.3±3.0 77.6±0.1 马 67.3±0.9 62.6±0.6 58.6±2.9 62.6±0 62.5±0.8 60.1±2.7 70.2±0.3 船舶 75.9±1.2 74.9±0.4 76.8±1.4 75.1±0 75.8±4.1 64.7±1.6 82.1±0.4 卡车 73.1±1.2 75.9±0.3 67.3±3.0 76.0±0 66.5±2.8 60.3±4.9 79.5±0.3
下载: 导出CSV
表 3 IMS数据集上的异常点样本
Table 3. Outlier samples on the IMS dataset
信号 SSDSVDD DSVDD ANOGAN OC-NN d1 530 537 533 533 d2 533 535 533 533 d3 700 698 701 700 d4 829 841 835 830 d5 900 903 900 901 a5 981 980 980 980 AUC 100.0 99.6 99.8 99.8 Acc 100.0 99.6 100.0 100.0
下载: 导出CSV
表 4 IDES数据集上的异常点样本
Table 4. Outlier samples on the IDES dataset
信号 SSDSVDD DSVDD ANOGAN OC-NN d1 590 596 599 593 d2 590 595 597 595 d3 590 595 598 594 d4 590 595 598 594 d5 643 650 645 652 a5 690 693 689 690 AUC 100.0 99.3 98.9 99.6 Acc 100.0 99.3 98.9 99.6
下载: 导出CSV
-
[1] 陈果. 滚动轴承早期故障的特征提取与智能诊断[J]. 航空学报,2009,30(2): 362-367. CHEN Guo. Feature extraction and intelligent diagnosis for ball bearing early faults[J]. Acta Aeronautica et Astronautica Sinica,2009,30(2): 362-367. (in Chinese doi: 10.3321/j.issn:1000-6893.2009.02.028CHEN Guo. Feature extraction and intelligent diagnosis for ball bearing early faults[J]. Acta Aeronautica et Astronautica Sinica, 2009, 30(2): 362-367. (in Chinese) doi: 10.3321/j.issn:1000-6893.2009.02.028 [2] LEI Yaguo,YANG Bin,JIANG Xinwei,et al. Applications of machine learning to machine fault diagnosis: a review and roadmap[J]. Mechanical Systems and Signal Processing,2020,138: 106587. doi: 10.1016/j.ymssp.2019.106587 [3] RAI A,UPADHYAY S H. A review on signal processing techniques utilized in the fault diagnosis of rolling element bearings[J]. Tribology International,2016,96: 289-306. doi: 10.1016/j.triboint.2015.12.037 [4] XU Gaowei,LIU Min,JIANG Zhuofu,et al. Online fault diagnosis method based on transfer convolutional neural networks[J]. IEEE Transactions on Instrumentation and Measurement,2020,69(2): 509-520. doi: 10.1109/TIM.2019.2902003 [5] 李巍华,单外平,曾雪琼. 基于深度信念网络的轴承故障分类识别[J]. 振动工程学报,2016,29(2): 340-347. LI Weihua,SHAN Waiping,ZENG Xueqiong. Bearing fault identification based on deep belief network[J]. Journal of Vibration Engineering,2016,29(2): 340-347. (in ChineseLI Weihua, SHAN Waiping, ZENG Xueqiong. Bearing fault identification based on deep belief network[J]. Journal of Vibration Engineering, 2016, 29(2): 340-347. (in Chinese) [6] 刘飞,陈仁文,邢凯玲,等. 基于迁移学习与深度残差网络的滚动轴承快速故障诊断算法[J]. 振动与冲击,2022,41(3): 154-164. LIU Fei,CHEN Renwen,XING Kailing,et al. Fast fault diagnosis algorithm for rolling bearing based on transfer learning and deep residual network[J]. Journal of Vibration and Shock,2022,41(3): 154-164. (in ChineseLIU Fei, CHEN Renwen, XING Kailing, et al. Fast fault diagnosis algorithm for rolling bearing based on transfer learning and deep residual network[J]. Journal of Vibration and Shock, 2022, 41(3): 154-164. （in Chinese） [7] PANG,Guansong,SHEN,Chunhua,CAO,Longbin,et al. Deep learning for anomaly detection: a survey[J]. Association for Computing Machinery,2021,54(2): 1-38. [8] HUANG Xin,WEN Guangrui,DONG Shuzhi,et al. Memory residual regression autoencoder for bearing fault detection[J]. IEEE Transactions on Instrumentation and Measurement,2021,70: 3515512. [9] 戴俊,王俊,朱忠奎,等. 基于生成对抗网络和自动编码器的机械系统异常检测[J]. 仪器仪表学报,2019,40(9): 16-26. DAI Jun,WANG Jun,ZHU Zhongkui,et al. Anomaly detection of mechanical systems based on generative adversarial network and auto-encoder[J]. Chinese Journal of Scientific Instrument,2019,40(9): 16-26. (in ChineseDAI Jun, WANG Jun, ZHU Zhongkui, et al. Anomaly detection of mechanical systems based on generative adversarial network and auto-encoder[J]. Chinese Journal of Scientific Instrument, 2019, 40(9): 16-26. (in Chinese) [10] WU Jingyao,ZHAO Zhibin,SUN Chuang,et al. Fault-attention generative probabilistic adversarial autoencoder for machine anomaly detection[J]. IEEE Transactions on Industrial Informatics,2020,16(12): 7479-7488. doi: 10.1109/TII.2020.2976752 [11] ZHANG Shen,YE Fei,WANG Bingnan,et al. Semi-supervised learning of bearing anomaly detection via deep variational autoencoders[EB/OL]. (2019-10-09)[2022-06-10]. http://arxiv.org/abs/1912.01096 [12] ZENG Ming,YANG Yu,LUO Songrong,et al. One-class classification based on the convex hull for bearing fault detection[J]. Mechanical Systems and Signal Processing,2016,81: 274-293. doi: 10.1016/j.ymssp.2016.04.001 [13] LIN Tong,CHEN Guo,OUYANG Wenli,et al. Hyper-spherical distance discrimination: a novel data description method for aero-engine rolling bearing fault detection[J]. Mechanical Systems and Signal Processing,2018,109: 330-351. doi: 10.1016/j.ymssp.2018.01.009 [14] PAN Yuna,CHEN Jin,GUO Lei. Robust bearing performance degradation assessment method based on improved wavelet packet-support vector data description[J]. Mechanical Systems and Signal Processing,2009,23(3): 669-681. doi: 10.1016/j.ymssp.2008.05.011 [15] RUFF L ,VANDERMEULEN R A ,GÖRNITZ N,et al. Deep one-class classification[C]//Proceedings of the 35th International Conference on Machine Learning. Stockholm,Sweden: International Machine Learning Societ,2018 : 4393-4402 [16] CHALAPATHY R,MENON A K,CHAWLA S. Anomaly detection using one-class neural networks[EB/OL]. (2019-01-09)[2022-06-10]. http://arxiv.org/abs/1802.06360. [17] ZHANG Hang,WU Chongruo,ZHANG Zhongyue,et al. ResNeSt: split-attention networks[EB/OL]. (2020-12-30)[2022-06-10]. http://arxiv.org/abs/2004.08955. [18] HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al. Delving deep into rectifiers: surpassing human-level performance on ImageNet classification[C]//2015 IEEE International Conference on Computer Vision. Piscataway,US: IEEE,2015: 1026-1034. [19] 康玉祥,陈果,尉询楷,等. 深度残差对冲网络及其在滚动轴承故障诊断中的应用[J]. 航空学报,2022,43(8): 63-74. KANG Yuxiang,CHEN Guo,WEI Xunkai,et al. Deep residual hedging network and its application in fault diagnosis of roll-ing bearings[J]. Acta Aeronautica et Astronautica Sinica,2022,43(8): 63-74. (in ChineseKANG Yuxiang, CHEN Guo, WEI Xunkai, et al. Deep residual hedging network and its application in fault diagnosis of roll-ing bearings[J]. Acta Aeronautica et Astronautica Sinica, 2022, 43(8): 63-74. （in Chinese） [20] ERFANI S M,RAJASEGARAR S,KARUNASEKERA S,et al. High-dimensional and large-scale anomaly detection using a linear one-class SVM with deep learning[J]. Pattern Recognition,2016,58: 121-134. doi: 10.1016/j.patcog.2016.03.028 [21] DENG L. The mnist database of handwritten digit images for machine learning research[J]. IEEE Signal Processing Magazine,2012,29(6): 141-142. doi: 10.1109/MSP.2012.2211477 [22] KRIZHEVSKY A,HINTON G. Learning multiple layers of features from tiny images[J]. Handbook of Systemic Autoimmune Diseases,2009,1(4): 84-90. [23] SCHOLKOPF B,SMOLA A J. Support vector machines,regularization,optimization,and beyond[J]. IEEE Transactions on Neural Networks,2005,16(3): 781-781. doi: 10.1109/TNN.2005.848998 [24] MASCI J,MEIER U,CIRES D,et al. Stacked convolutional auto-encoders for hierarchical feature extraction [C]// International Conference on Artificial Neural Networks. Espoo,Finland: Springer-Verlag,2011 52-59. [25] PARZEN E. On estimation of a probability density function and mode[J]. The Annals of Mathematical Statistics,1962,33(3): 1065-1076. doi: 10.1214/aoms/1177704472 [26] RADFORD A,METZ L,CHINTALA S. Unsupervised representation learning with deep convolutional generative adversarial networks[EB/OL]. (2016-01-07)[2022-06-10]. http://arxiv.org/abs/1511.06434. [27] QIU Hai,LEE J,LIN Jing,et al. Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics[J]. Journal of Sound Vibration,2006,289(4/5): 1066-1090. -
点击查看大图
计量
- 文章访问数: 914
- HTML浏览量: 290
- PDF量: 36
- 被引次数: 0