logo

SCIENTIA SINICA Informationis, Volume 50 , Issue 4 : 483-495(2020) https://doi.org/10.1360/SSI-2019-0223

A fault detection method for a braking system of high-speed trains

More info
  • ReceivedOct 8, 2019
  • AcceptedNov 18, 2019
  • PublishedApr 9, 2020

Abstract


Funded by

国家自然科学基金(61490701,61751307,61473164)

山东省泰山学者项目


Supplement

Appendix

高速列车制动系统故障诊断架构的离线建模流程

原有的故障诊断离线建模流程主要基于参考系统和专家经验进行模型设计. 以针对电空制动回路的监控为例, 通过参考系统和专家经验确定的故障偏差幅值为20 kPa, 虽然能够保障基本的行车安全, 若发生故障并触发报警, 就会引起列车降速.

基于高速列车制动系统故障诊断架构的离线建模流程如图A1所示, 主要涉及8个部分, 分别是: 数据库, 远程故障诊断系统, 故障诊断模型, 快速开发中心, 在线故障诊断系统, 制动系统, 通讯单元和数据获取单元.

本文提出的架构在建模部分着重考虑了基于数据的建模流程, 弱化了参考系统和专家经验的作用. 该架构可基于多源数据进行统计分析并提取故障特征, 实现数据驱动的离线建模. 然后, 经由快速开发中心对故障诊断模型进行快速部署及在线测试与验证. 如果参数整定存在偏差, 可以基于快速开发中心进行快速修正并重新部署. 这种开发流程克服了原有架构以逆向设计为主及开发迭代周期较长的缺点, 实现了故障诊断模型的正向设计与快速迭代开发. 并将支持“被动式"故障诊断模型向“主动式"的转变.

高速列车制动系统故障诊断架构的在线故障诊断流程

原有的故障诊断系统架构能够实时进行故障诊断, 可以满足基本的行车安全. 但由于在线计算资源受限等原因, 尚无法实现先进故障诊断算法的部署与实现. 另外, 当发生无对应故障代码的故障时, 无相关流程对该故障进行有效检测. 考虑到以上两点, 本文提出的高铁制动系统故障诊断架构完善了在线故障诊断流程, 如图2所示.

在图2中, 将远程故障诊断系统囊括到在线故障检测流程中. 随着通讯技术的快速发展, 高铁在线运行数据可通过网络实时传输至地面远程故障诊断系统中. 从而, 通过在远程故障诊断系统中部署Spark Streaming实现对复杂故障的实时诊断. 同时, 解决了在线故障诊断系统计算资源受限的问题.


References

[1] Zhang W, Wu P, Wu X. An investigation into structural failures of Chinese high-speed trains. Eng Failure Anal, 2006, 13: 427-441 CrossRef Google Scholar

[2] Ai B, Briso-Rodriguez C, Cheng X. Challenges Toward Wireless Communications for High-Speed Railway. IEEE Trans Intell Transp Syst, 2014, 15: 2143-2158 CrossRef Google Scholar

[3] Hodge V J, O'Keefe S, Weeks M. Wireless Sensor Networks for Condition Monitoring in the Railway Industry: A Survey. IEEE Trans Intell Transp Syst, 2015, 16: 1088-1106 CrossRef Google Scholar

[4] Chen H, Jiang B, Lu N. Real-time incipient fault detection for electrical traction systems of CRH2. Neurocomputing, 2018, 306: 119-129 CrossRef Google Scholar

[5] Xiao Y Q, He Y G. A linear ridgelet network approach for fault diagnosis of analog circuit. Sci China Inf Sci, 2010, 53: 2251-2264 CrossRef Google Scholar

[6] Zhang J, Christofides P D, He X. Intermittent sensor fault detection for stochastic LTV systems with parameter uncertainty and limited resolution. Int J Control, 2018, 2010: 1-9 CrossRef Google Scholar

[7] Wan Y M, Wang W, Ye H. Discrete time-varying fault detection filter for non-uniformly sampled-data systems. Sci China Inf Sci, 2014, 57: 1-11 CrossRef Google Scholar

[8] Alauddin M, Khan F, Imtiaz S. A Bibliometric Review and Analysis of Data-Driven Fault Detection and Diagnosis Methods for Process Systems. Ind Eng Chem Res, 2018, 57: 10719-10735 CrossRef Google Scholar

[9] Guo T X. Some data-driven based fault detection methods with applications in brake system of high-speed trains. Dissertation for Ph.D. Degree. Beijing: Tsinghua University, 2019. Google Scholar

[10] Ji H, He X, Shang J. Incipient fault detection with smoothing techniques in statistical process monitoring. Control Eng Practice, 2017, 62: 11-21 CrossRef Google Scholar

[11] Zhou D, Ji H, He X. Fault Detection and Isolation of the Brake Cylinder System for Electric Multiple Units. IEEE Trans Contr Syst Technol, 2018, 26: 1744-1757 CrossRef Google Scholar

[12] Ma T, Wu M, Tian C. A general method of braking process simulation for flexible marshalling EMUs. Int J TDI, 2018, 2: 353-361 CrossRef Google Scholar

[13] Ji H Q, He X, Sai H S, et al. Fault detection of EMU brake cylinder. In: Proceedings of the 35th Chinese Control Conference (CCC), 2016. 6668--6672. Google Scholar

[14] Zhou D H, Ji H Q, He X. Fault diagnosis techniques for the information control system of high-speed trains. Act Autom Sin, 2018, 44: 1153--1164. Google Scholar

[15] Zhou D H, Liu Y, He X. Review on Fault Diagnosis Techniques for Closed-loop Systems. Act Autom Sin, 2013, 39: 1933--1943. Google Scholar

[16] Hwang I, Kim S, Kim Y. A Survey of Fault Detection, Isolation, and Reconfiguration Methods. IEEE Trans Contr Syst Technol, 2010, 18: 636-653 CrossRef Google Scholar

[17] Zhang J, He X, Zhou D. Robust optimal filtering for linear time-varying systems with stochastic uncertainties. IET Contr Theor Appl, 2017, 60: 3297-3304 CrossRef Google Scholar

[18] Marino A, Pierri F, Arrichiello F. Distributed Fault Detection Isolation and Accommodation for Homogeneous Networked Discrete-Time Linear Systems. IEEE Trans Automat Contr, 2017, 62: 4840-4847 CrossRef Google Scholar

[19] Mnassri B, El Adel E M, Ouladsine M. Reconstruction-based contribution approaches for improved fault diagnosis using principal component analysis. J Process Control, 2015, 33: 60-76 CrossRef Google Scholar

[20] Huang J, Yan X. Gaussian and non-Gaussian Double Subspace Statistical Process Monitoring Based on Principal Component Analysis and Independent Component Analysis. Ind Eng Chem Res, 2015, 54: 1015-1027 CrossRef Google Scholar

[21] Zhou B, Ye H, Zhang H. Process monitoring of iron-making process in a blast furnace with PCA-based methods. Control Eng Practice, 2016, 47: 1-14 CrossRef Google Scholar

[22] Yin S, Ding S X, Xie X. A Review on Basic Data-Driven Approaches for Industrial Process Monitoring. IEEE Trans Ind Electron, 2014, 61: 6418-6428 CrossRef Google Scholar

[23] Ge Z. Review on data-driven modeling and monitoring for plant-wide industrial processes. Chemometrics Intelligent Laboratory Syst, 2017, 171: 16-25 CrossRef Google Scholar

[24] Shang J, Chen M, Ji H. Recursive transformed component statistical analysis for incipient fault detection. Automatica, 2017, 80: 313-327 CrossRef Google Scholar

[25] Peng X, Tang Y, Du W. Multimode Process Monitoring and Fault Detection: A Sparse Modeling and Dictionary Learning Method. IEEE Trans Ind Electron, 2017, 64: 4866-4875 CrossRef Google Scholar

[26] Tong C, Lan T, Shi X. Statistical process monitoring based on nonlocal and multiple neighborhoods preserving embedding model. J Process Control, 2018, 65: 34-40 CrossRef Google Scholar

[27] Tong C, Lan T, Shi X. Ensemble modified independent component analysis for enhanced non-Gaussian process monitoring. Control Eng Practice, 2017, 58: 34-41 CrossRef Google Scholar

[28] Chou M, Xia X, Kayser C. Modelling and model validation of heavy-haul trains equipped with electronically controlled pneumatic brake systems. Control Eng Practice, 2007, 15: 501-509 CrossRef Google Scholar

[29] Wang S, Li P, Wang F, et al. Control of brake cylinder pressure on metro trains for Beijing No. 5 line. Rolling Stock, 2014, 52: 5--8. Google Scholar

[30] Zhang S G. CRH2 EMU. Beijing: China Railway Publishing House, 2008. Google Scholar

[31] Zhang X D. Matrix Analysis and Applications. Cambridge: Cambridge University Press, 2017. Google Scholar

[32] Friedman J, Hastie T, Tibshirani R. The Elements of Statistical Learning. Berlin: Springer, 2001. Google Scholar

[33] MacGregor J F, Kourti T. Statistical process control of multivariate processes. Control Eng Practice, 1995, 3: 403-414 CrossRef Google Scholar

[34] Zhou D H, Ye Y Z. Modern Fault Diagnosis and Fault-Tolerant Control. Beijing: Tsinghua University Press, 2000. Google Scholar

[35] Lee J M, Yoo C K, Lee I B. Statistical process monitoring with independent component analysis. J Process Control, 2004, 14: 467-485 CrossRef Google Scholar

[36] Ji H Q. Data-driven incipient fault diagnosis methods and applications. Dissertation for Ph.D. Degree. Beijing: Tsinghua University, 2017. Google Scholar

  • Figure 1

    (Color online) Control schematic diagram of electro-pneumatic control loop of high-speed train braking system

  • Figure 2

    (Color online) Braking test platform provided by Sifang Rolling Stock Research Institute

  • Figure 3

    (Color online) Fault diagnosis system architecture of high-speed train braking system

  • Figure 4

    (Color online) Detection results of KNORR logic-based fault detection method

  • Figure 5

    (Color online) Detection results of GMDA-based fault detection method. (a) $l=20$, $p=8$; (b) $l=20$, $p=16$; (c) $l=40$, $p=8$; (d) $l=40$, $p=16$; (e) $l=60$, $p=8$; (f) $l=60$, $p=16$

  •   

    Algorithm 1 基于GMDA的故障检测算法

    Require:正常工况训练数据矩阵$X~\in~\mathbb{R}^{d~\times~n}$, $l$类异常工况训练数据矩阵${Y_k}~\in~\mathbb{R}^{d~\times~{m_k}}$, $k=1,2,3,\ldots,l$, 参数: 子空间维数$p$,显著性水平$\alpha$.

    Output: 张成故障检测模型子空间的矩阵$W$, $p$个最大特征值对应的特征值对角阵$\Lambda$, $T^2$统计量所对应的控制限$T_{\lim~}^2$.

    通过式(1)计算 $S_{W}$;

    通过式(3)计算 $S_{G_i}$;

    通过式(2)计算 $S_{G}$;

    通过公式(6)的广义特征值问题,得到闭式解$W^*$和对应的特征值对角阵${\Lambda}^*$,取出$p$个最大特征值对应的特征向量组成矩阵$W$并构成最大特征值对角阵$\Lambda$;

    求解$T_{\lim~}^2$控制限.(在线故障检测部分)

    Require:待检测的数据样本$x_i~\in~\mathbb{R}^{d}$, 建模过程求得的矩阵$W$和特征值矩阵$\Lambda$, 统计量$T^2$对应求得的控制限$T_{\lim~}^2$.

    Output: 待检测的数据样本是否为故障数据样本.

    通过式(10)求解$x_i$对应的$T^2$统计量 $T_i^2$;

    判断$x_i$是否为故障数据样本;

    如果 $T_i^2~\leq~T_{\lim~}^2$, $x_i$ 为正常样本;否则 $x_i$ 为故障样本.

  • Table 1   Fault detection rate (FDR) and false alarm rate (FAR) of GMDA-based fault detection method
    Augment parameter $l=$ 20 $l=$ 40 $l=$ 60
    Leakage (SLPM) $1.0\sim2.0$ $2.0\sim4.0$ Average $1.0\sim2.0$ $2.0\sim4.0$ Average $1.0\sim2.0$ $2.0\sim4.0$ Average
    Dimension $p$ FDR (%) FDR (%) FAR (%) FDR (%) FDR (%) FAR (%) FDR (%) FDR (%) FAR (%)
    2 55.33 60.00 0.00 62.00 79.33 0.00 60.67 76.00 0.00
    4 69.33 56.67 0.00 69.33 76.67 0.00 70.67 84.67 0.00
    6 72.67 58.67 0.00 75.33 85.33 0.60 80.00 92.67 0.00
    8 72.67 62.67 1.00 78.67 88.67 0.60 93.33 97.33 0.20
    10 74.00 69.33 0.40 82.67 91.33 0.60 99.33 98.00 0.20
    12 74.00 67.33 0.00 86.00 92.00 0.60 98.67 98.00 1.60
    14 78.00 68.67 0.20 90.00 92.67 0.20 100.00 98.67 2.00
    16 81.33 76.67 0.00 90.00 92.67 0.00 100.00 100.00 2.00