CN103926092A

CN103926092A - Rail transit train suspension system key component pulse impact detection method

Info

Publication number: CN103926092A
Application number: CN201410064853.1A
Authority: CN
Inventors: 魏秀琨; 贾利民; 郭昆; 柳海; 杨帅; 管岭; 王晓全; 王群
Original assignee: Beijing Jiaotong University
Current assignee: Beijing Jiaotong University
Priority date: 2014-02-25
Filing date: 2014-02-25
Publication date: 2014-07-16

Abstract

The invention discloses a rail transit train suspension system key component pulse impact detection method. According to the detection method, dynamics modeling is carried out on rail vehicle suspension system components, vehicle operating state information and sensor data are obtained, fault detection is carried out through a PCA method, step signals are added to a vehicle model, and fault separation is carried out on a train suspension system by the adoption of a D-S evidence theory. The detection method can be combined with a wireless sensing technology, a set of equipment is adopted to detect many vehicles, it is needless that each vehicle is provided with a fixed detection device, and investment and maintenance cost of the detection devices is greatly reduced.

Description

Pulse impact detection method for key components of rail transit train suspension system

技术领域 technical field

本发明涉及铁路运输技术领域，更具体地涉及一种轨道交通列车悬挂系统关键元部件脉冲冲击检测方法。 The invention relates to the technical field of railway transportation, in particular to a pulse impact detection method for key components of a rail transit train suspension system. the

背景技术 Background technique

随着轨道交通的快速发展，轨道列车的安全性和舒适性得到了社会的广泛关注，而悬挂系统作为轨道车辆走行部分的关键部件，其状态好坏对车辆运行安全与舒适度有极为重要的影响。对列车悬挂系统进行状态监控与故障诊断，不仅可以提高城轨车辆运行的安全性与可靠性，而且能够大幅减少由于定期检修而产生的费用。因此，对列车悬挂系统各个部件进行性能监测和故障诊断是十分必要且具有重要意义的。 With the rapid development of rail transportation, the safety and comfort of rail trains have received widespread attention from the society. As a key component of the running part of rail vehicles, the suspension system is of great importance to the safety and comfort of vehicle operation. Influence. The status monitoring and fault diagnosis of the train suspension system can not only improve the safety and reliability of urban rail vehicle operation, but also greatly reduce the cost of regular maintenance. Therefore, it is very necessary and significant to perform performance monitoring and fault diagnosis of each component of the train suspension system. the

近年来，轨道交通列车悬挂系统健康状态的在线监测技术得到了国内外学者的关注，并取得了一些成果。有研究是采用卡尔曼滤波器估计车辆横向系统的状态值，从而得到理想观测值，该理想观测值与实际观测值的残差用于判断系统何时发生故障。该方法能监测到系统的突发性故障，对于系统的渐变性故障（一般是系统老化引起）则无能为力。还有文献是通过建立轨道车辆系统垂向状态空间模型，提出再次均匀采样策略，解决了监测参数突变的问题但是却需要实时采集数据，每套设备只能针对一辆列车，费用较大。因此，需要提供一种轨道交通列车悬挂系统监测方法。 In recent years, the online monitoring technology of the health status of rail transit train suspension system has attracted the attention of scholars at home and abroad, and some achievements have been made. Some studies use the Kalman filter to estimate the state value of the vehicle lateral system, so as to obtain the ideal observation value, and the residual error between the ideal observation value and the actual observation value is used to judge when the system fails. This method can monitor the sudden failure of the system, but it can't do anything about the gradual failure of the system (usually caused by system aging). There are also literatures that establish a vertical state space model of the rail vehicle system and propose a uniform sampling strategy again, which solves the problem of sudden changes in monitoring parameters, but requires real-time data collection. Each set of equipment can only be used for one train, and the cost is relatively high. Therefore, it is necessary to provide a method for monitoring the suspension system of rail transit trains. the

发明内容 Contents of the invention

为解决以上现有技术的不足，本发明提供一种轨道交通列车悬挂系统关键元部件脉冲冲击检测方法。 In order to solve the above deficiencies in the prior art, the present invention provides a pulse impact detection method for key components of a rail transit train suspension system. the

本发明的技术方案如下： Technical scheme of the present invention is as follows:

轨道交通列车悬挂系统关键元部件脉冲冲击检测方法，该检测方法包括如下步骤： A pulse impact detection method for key components of a rail transit train suspension system, the detection method includes the following steps:

1）对轨道车辆悬挂系统部件进行动力学建模； 1) Dynamic modeling of rail vehicle suspension system components;

2）根据模型，在车辆悬挂系统的部件相应处布设传感器，并通过传感器获取车辆的运动状态信息； 2) According to the model, sensors are arranged at the corresponding parts of the vehicle suspension system, and the vehicle's motion state information is obtained through the sensors;

3）根据获取的车辆运动状态信息，结合列车运行故障，对轨道车辆模型进行仿真分析，得到列车运行时传感器的数据； 3) According to the obtained vehicle motion state information, combined with the train operation fault, the simulation analysis of the rail vehicle model is carried out to obtain the sensor data during the train operation;

4）通过PCA方法对步骤3中得到的的所有传感器数据进行故障检测，判断系统的工作状态，是否有故障； 4) Perform fault detection on all sensor data obtained in step 3 by PCA method, and judge the working status of the system and whether there is a fault;

5）在上述轨道车辆模型的轨道激励中加入阶跃信号，进行仿真分析，得到列车运行时传感器的数据，为故障分离提供可靠的数据基础； 5) Add a step signal to the track excitation of the above-mentioned rail vehicle model, conduct simulation analysis, and obtain the data of the sensor when the train is running, providing a reliable data basis for fault isolation;

6）对步骤5中获得的传感器数据，分别提取仿真得到的所有故障类型的特征值； 6) For the sensor data obtained in step 5, extract the eigenvalues of all fault types obtained by simulation;

7）在得到故障特征值的基础上，运用D-S证据理论的信息融合方法对城轨列车悬挂系统进行故障分离，准确判定故障的部件和衰减程度。 7) On the basis of obtaining the fault eigenvalues, use the information fusion method of the D-S evidence theory to separate the faults of the urban rail train suspension system, and accurately determine the faulty components and attenuation degree. the

本发明的有益效果如下： The beneficial effects of the present invention are as follows:

本发明提出的脉冲冲击检测法，为列车悬挂系统的故障检测提供了一种新的策略，是对现有列车悬挂系统检测方法的补充和完善。本发明中提出的脉冲检测法是在试验线对列车的悬挂系统的性能进行检测，与无线传感技术相结合，采用一套设备就可以对很多辆车进行检测，不需要每辆车安装固定的检测设备。因而，极大地减少了检测设备的投资和维护费用。 The pulse impact detection method proposed by the invention provides a new strategy for the fault detection of the train suspension system, and is a supplement and improvement to the existing detection method of the train suspension system. The pulse detection method proposed in the present invention is to detect the performance of the suspension system of the train on the test line. Combining with wireless sensing technology, a set of equipment can be used to detect many vehicles without the need for each vehicle to be installed and fixed. testing equipment. Therefore, the investment and maintenance costs of detection equipment are greatly reduced. the

附图说明 Description of drawings

下面结合附图对本发明的具体实施方式作进一步详细的说明。 The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings. the

图1是检测流程； Figure 1 is the detection process;

图2是车体重心偏移对各车轮垂向偏移的频域响应； Figure 2 is the frequency domain response of the center of gravity offset of the vehicle to the vertical offset of each wheel;

图3是转向架重心偏移对相应车轮垂向偏移的频域响应； Fig. 3 is the frequency domain response of bogie center of gravity offset to corresponding wheel vertical offset;

图4是轨道车辆悬挂系统模型和传感器布置方案； Fig. 4 is rail vehicle suspension system model and sensor arrangement scheme;

图5(a)-5(b)是轨道激励图像； Figures 5(a)-5(b) are orbital excitation images;

图6是PCA故障检测结果。 Figure 6 is the PCA fault detection results. the

具体实施方式 Detailed ways

为了更清楚地说明本发明，下面结合优选实施例和附图对本发明做进一步的说明。附图中相似的部件以相同的附图标记进行表示。本领域技术人员应当理解，下面所具体描述的内容是说明性的而非限制性的，不应以此限制本发明的保护范围。 In order to illustrate the present invention more clearly, the present invention will be further described below in conjunction with preferred embodiments and accompanying drawings. Similar parts in the figures are denoted by the same reference numerals. Those skilled in the art should understand that the content specifically described below is illustrative rather than restrictive, and should not limit the protection scope of the present invention. the

本发明针对上述缺陷公开了一种脉冲冲击检测方法。本发明的主要特点在于可应用到轨道交通列车悬挂系统关键元部件的故障诊断中，主要过程如图1所示，在列车的每一节车厢及相应转向架上布设共计12个加速度传感器，当列车经过试验线路时通过无线传输技术将传感器数据发送至地面故障诊断系统，首先运用PCA算法检测悬挂系统是否发生故障。在确认悬挂系统存在故障的情况下，让列车通过台阶，获取这一过程的传感器数据，并做特征提取。最后利用D-S证据理论完成故障分离。其具体的方法及步骤如下：（表1中都有说明） The invention discloses a pulse impact detection method aiming at the above defects. The main feature of the present invention is that it can be applied to the fault diagnosis of the key components of the suspension system of rail transit trains. When the train passes the test line, the sensor data is sent to the ground fault diagnosis system through wireless transmission technology. First, the PCA algorithm is used to detect whether the suspension system is faulty. In the case of confirming that there is a fault in the suspension system, let the train pass through the steps, obtain the sensor data of this process, and perform feature extraction. Finally, the fault isolation is completed by using the D-S evidence theory. The specific methods and steps are as follows: (All are described in Table 1)

表1 Table 1

1、城轨车辆垂向悬挂系统动力学建模 1. Dynamic modeling of urban rail vehicle vertical suspension system

由城轨车辆悬挂系统部件所在位置力的作用情况，以及车体与前后转向架在各个自由度上的运动状态，依据牛顿定律，可以分别写出车体与前后转向架的沉浮运动微分方程、点头运动微分方程以及侧滚运动微分方程。 According to the action of the force at the position of the suspension system components of the urban rail vehicle, and the motion states of the car body and the front and rear bogies in each degree of freedom, according to Newton's law, the differential equations for the ups and downs of the car body and the front and rear bogies can be written, respectively. Differential equations of nodding motion and differential equations of roll motion. the

车体子系统模型 Body Subsystem Model

沉浮运动： Ups and downs:

$M m \overset{\cdot &Center Dot; \cdot &Center Dot;}{z z} + + {44 C C}_{22 z z} \overset{\cdot &Center Dot;}{z z} - - {22 C C}_{22 z z} {\overset{\cdot &Center Dot;}{z z}}_{FB Facebook} - - {22 C C}_{22 z z} {\overset{\cdot &Center Dot;}{z z}}_{RB RB} + + {44 K K}_{22 z z} z z - - {22 K K}_{22 z z} {z z}_{FB Facebook} - - {22 K K}_{22 z z} {z z}_{RB RB} = = 00$

点头运动： nodding movement:

${J J}_{φ φ} \overset{\cdot &Center Dot; \cdot &Center Dot;}{φ φ} + + {44 C C}_{22 z z} {l l}^{22} \overset{\cdot &Center Dot;}{φ φ} - - {22 C C}_{22 z z} l l {\overset{\cdot &Center Dot;}{z z}}_{FB Facebook} + + {22 C C}_{22 z z} l l {\overset{\cdot &Center Dot;}{z z}}_{RB RB} + + {44 K K}_{22 z z} {l l}^{22} φ φ - - {22 K K}_{22 z z} {lz lz}_{FB Facebook} + + {22 K K}_{22 z z} {lz lz}_{RB RB} = = 00$

侧滚运动： Rolling movement:

${J J}_{θ θ} \overset{\cdot \cdot \cdot \cdot}{θ θ} + + {44 C C}_{22 z z} {b b}^{22} \overset{\cdot \cdot}{θ θ} - - {22 C C}_{22 z z} {b b}^{22} {\overset{\cdot &Center Dot;}{θ θ}}_{FB Facebook} - - {22 C C}_{22 z z} {b b}^{22} {θ θ}_{RB RB} + + (({44 K K}_{22 z z} {b b}^{22} + + {22 K K}_{θ θ})) θ θ - - (({22 K K}_{22 z z} {b b}^{22} + + {K K}_{θ θ})) {θ θ}_{FB Facebook} - - (({22 K K}_{22 z z} {b b}^{22} + + {K K}_{θ θ})) {θ θ}_{FB Facebook} = = 00$

其中，z，z_FB与z_RB分别表示车体与前后转向架质心的垂向位移，φ表示每个质量块围绕质心的点头角，θ则表示侧滚角，它们的脚标与z的脚标意义相同。方程中的其余符号为车辆参数，由南车株洲电力机车厂提供，详见表1。 Among them, z, z _FB and z _RB represent the vertical displacement of the car body and the center of mass of the front and rear bogies respectively, φ represents the nodding angle of each mass block around the center of mass, and θ represents the roll angle, and their subscripts are the same as the subscripts of z The sign has the same meaning. The remaining symbols in the equation are vehicle parameters, provided by CSR Zhuzhou Electric Locomotive Works, see Table 1 for details.

前转向架子系统模型 Front steering rack system model

沉浮运动： Ups and downs:

$\begin{matrix} {M m}_{B B} {\overset{\cdot &Center Dot; \cdot &Center Dot;}{z z}}_{FB Facebook} - - {22 C C}_{22 z z} \overset{\cdot &Center Dot;}{z z} - - {22 C C}_{22 z z} l l \overset{\cdot &Center Dot;}{φ φ} + + (({44 C C}_{11 z z} + + {22 C C}_{22 z z})) {\overset{\cdot &Center Dot;}{z z}}_{FB Facebook} - - {C C}_{11 z z} {\overset{\cdot &Center Dot;}{z z}}_{W W 11 R R} - - {C C}_{11 z z} {\overset{\cdot &Center Dot;}{z z}}_{W W 11 L L} - - {C C}_{11 z z} {\overset{\cdot &Center Dot;}{z z}}_{W W 22 R R} - - {C C}_{11 z z} {\overset{\cdot &Center Dot;}{z z}}_{W W 22 L L} - - {22 K K}_{22 z z} z z - - {22 K K}_{22 z z} lφ lφ + + (({44 K K}_{11 z z} + + {22 K K}_{22 z z})) {z z}_{FB Facebook} - - {K K}_{11 z z} {z z}_{W W 11 R R} \\ - - {K K}_{11 z z} {z z}_{W W 11 L L} - - {K K}_{11 z z} {z z}_{W W 22 R R} - - {K K}_{11 z z} {z z}_{W W 22 L L} = = 00 \end{matrix}$

点头运动： nodding movement:

$\begin{matrix} {J J}_{Bφ Bφ} {\overset{\cdot \cdot \cdot \cdot}{φ φ}}_{FB Facebook} + + {44 C C}_{11 z z} {l l}_{11}^{22} {\overset{\cdot \cdot}{φ φ}}_{FB Facebook} - - {C C}_{11 z z} {l l}_{11} {\overset{\cdot \cdot}{z z}}_{W W 11 R R} - - {C C}_{11 z z} {l l}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 11 L L} + + {C C}_{11 z z} {l l}_{11} {\overset{\cdot \cdot}{z z}}_{W W 22 R R} + + {C C}_{11 z z} {l l}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 22 L L} + + {44 K K}_{11 z z} {l l}_{11}^{22} {\overset{\cdot &Center Dot;}{φ φ}}_{FB Facebook} - - {K K}_{11 z z} {l l}_{11} {z z}_{W W 11 R R} - - {K K}_{11 z z} {l l}_{11} {z z}_{W W 11 L L} + + {K K}_{11 z z} {l l}_{11} {z z}_{W W 22 R R} \\ + + {K K}_{11 z z} {l l}_{11} {z z}_{W W 22 L L} = = 00 \end{matrix}$

侧滚运动： Rolling movement:

$\begin{matrix} {J J}_{Bθ Bθ} {\overset{\cdot &Center Dot; \cdot &Center Dot;}{θ θ}}_{FB Facebook} - - {22 C C}_{22 z z} {b b}^{22} \overset{\cdot &Center Dot;}{θ θ} + + (({22 C C}_{22 z z} {b b}^{22} + + {44 C C}_{11 z z} {b b}_{11}^{22})) {\overset{\cdot &Center Dot;}{θ θ}}_{FB Facebook} + + {C C}_{11 z z} {b b}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 11 R R} - - {C C}_{11 z z} {b b}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 11 L L} + + {C C}_{11 z z} {b b}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 22 R R} - - {C C}_{11 z z} {b b}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 22 L L} - - (({22 K K}_{22 z z} {b b}^{22} \\ + + {K K}_{θ θ})) θ θ + + (({22 K K}_{22 z z} {b b}^{22} + + {44 K K}_{11 z z} {b b}_{11}^{22} + + {K K}_{θ θ})) {θ θ}_{FB Facebook} + + {K K}_{11 z z} {b b}_{11} {z z}_{W W 11 R R} - - {K K}_{11 z z} {b b}_{11} {z z}_{W W 11 R R} + + {K K}_{11 z z} {b b}_{11} {z z}_{W W 22 R R} - - {K K}_{11 z z} {b b}_{11} {z z}_{W W 22 L L} = = 00 \end{matrix}$

其中，z_W1R表示前转向架前方轮对右侧车轮的垂向位移，z_W2L则表示前转向架后方轮对左侧车轮的垂向位移。依照这个逻辑就可以理解公式中其余符号的意义。 Among them, z _W1R represents the vertical displacement of the right wheel of the front wheel of the front bogie, and z _W2L represents the vertical displacement of the left wheel of the rear wheel of the front bogie. According to this logic, the meaning of the remaining symbols in the formula can be understood.

后转向架子系统模型 Rear steering rack system model

沉浮运动： Ups and downs:

$\begin{matrix} {M m}_{B B} {\overset{\cdot \cdot \cdot \cdot}{z z}}_{RB RB} - - {22 C C}_{22 z z} \overset{\cdot \cdot}{z z} + + {22 C C}_{22 z z} l l \overset{\cdot \cdot}{φ φ} + + (({44 C C}_{11 z z} + + {22 C C}_{22 z z})) {\overset{\cdot &Center Dot;}{z z}}_{RB RB} - - {C C}_{11 z z} {\overset{\cdot &Center Dot;}{z z}}_{W W 33 R R} - - {C C}_{11 z z} {\overset{\cdot &Center Dot;}{z z}}_{W W 33 L L} - - {C C}_{11 z z} {\overset{\cdot &Center Dot;}{z z}}_{W W 44 R R} - - {C C}_{11 z z} {\overset{\cdot \cdot}{z z}}_{W W 44 L L} - - {22 K K}_{22 z z} z z \\ + + {22 K K}_{22 z z} lφ lφ + + (({44 K K}_{11 z z} + + {22 K K}_{22 z z})) {z z}_{RB RB} - - {K K}_{11 z z} {z z}_{E E. 33 R R} - - {K K}_{11 z z} {z z}_{W W 33 L L} - - {K K}_{11 z z} {z z}_{W W 44 R R} - - {L L}_{11 z z} {z z}_{W W 44 L L} = = 00 \end{matrix}$

点头运动： nodding movement:

$\begin{matrix} {J J}_{Bφ Bφ} {\overset{\cdot &Center Dot; \cdot &Center Dot;}{φ φ}}_{RB RB} + + {44 C C}_{11 z z} {l l}_{11}^{22} {\overset{\cdot \cdot}{φ φ}}_{RB RB} - - {C C}_{11 z z} {l l}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 33 R R} - - {C C}_{11 z z} {l l}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 33 L L} + + {C C}_{11 z z} {l l}_{11} {\overset{\cdot \cdot}{z z}}_{W W 44 R R} + + {C C}_{11 z z} {l l}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 44 L L} + + {44 K K}_{11 z z} {l l}_{11}^{22} {\overset{\cdot &Center Dot;}{φ φ}}_{RB RB} - - {K K}_{11 z z} {l l}_{11} {z z}_{W W 33 R R} - - {K K}_{11 z z} {l l}_{11} {z z}_{W W 33 L L} \\ + + {K K}_{11 z z} {l l}_{11} {z z}_{W W 44 R R} + + = = {K K}_{11 z z} {l l}_{11} {z z}_{W W 44 L L} 00 \end{matrix}$

侧滚运动： Rolling movement:

$\begin{matrix} {J J}_{Bθ Bθ} {\overset{\cdot &Center Dot; \cdot &Center Dot;}{θ θ}}_{RB RB} - - {22 C C}_{22 z z} {b b}^{22} \overset{\cdot &Center Dot;}{θ θ} + + (({22 C C}_{22 z z} {b b}^{22} + + {44 C C}_{11 z z})) {b b}_{11}^{22} {\overset{\cdot \cdot}{θ θ}}_{RB RB} + + {C C}_{11 z z} {b b}_{11} {\overset{\cdot \cdot}{z z}}_{W W 33 R R} - - {C C}_{11 z z} {b b}_{11} {\overset{\cdot \cdot}{z z}}_{W W 33 L L} + + {C C}_{11 z z} {b b}_{11} {\overset{\cdot &Center Dot;}{z z}}_{W W 44 R R} - - {C C}_{11 z z} {b b}_{11} {\overset{\cdot \cdot}{z z}}_{W W 44 L L} - - (({22 K K}_{22 z z} {b b}^{22} \\ + + {K K}_{θ θ})) θ θ + + (({22 K K}_{22 z z} {b b}^{22} + + {44 K K}_{11 z z} {b b}_{11}^{22} + + {K K}_{θ θ})) {θ θ}_{RB RB} + + {K K}_{11 z z} {b b}_{11} {z z}_{W W 33 R R} - - {K K}_{11 z z} {b b}_{11} {z z}_{W W 33 L L} + + {K K}_{11 z z} {b b}_{11} {z z}_{W W 44 R R} - - {K K}_{11 z z} {b b}_{11} {z z}_{W W 44 L L} = = 00 \end{matrix}$

基于上述公式，车辆悬挂系统空间状态方程如下： Based on the above formula, the spatial state equation of the vehicle suspension system is as follows:

$\overset{\cdot &Center Dot;}{x x} = = Ax Ax + + {B B}_{d d} d d$

y＝Cx+D_dd y=Cx+D _d d

其中 in

$x x = = {[\begin{matrix} \overset{\cdot &Center Dot;}{z z} & \overset{\cdot &Center Dot;}{φ φ} & \overset{\cdot &Center Dot;}{θ θ} & z z & φ φ & θ θ & {\overset{\cdot &Center Dot;}{z z}}_{FB Facebook} & {\overset{\cdot &Center Dot;}{φ φ}}_{FB Facebook} & {\overset{\cdot \cdot}{θ θ}}_{FB Facebook} & {z z}_{FB Facebook} & {φ φ}_{FB Facebook} & {θ θ}_{FB Facebook} & {\overset{\cdot &Center Dot;}{z z}}_{RB RB} & {\overset{\cdot &Center Dot;}{φ φ}}_{B B} & {\overset{\cdot \cdot}{θ θ}}_{RB RB} & {z z}_{RB RB} & {φ φ}_{RB RB} & {θ θ}_{RB RB} \end{matrix}]}^{T T}$

$d d = = {[\begin{matrix} {\overset{\cdot &Center Dot;}{z z}}_{W W 11 R R} & {\overset{\cdot &Center Dot;}{z z}}_{W W 11 L L} & {\overset{\cdot &Center Dot;}{z z}}_{W W 22 R R} & {\overset{\cdot &Center Dot;}{z z}}_{W W 22 L L} & {z z}_{W W 11 R R} & {z z}_{W W 11 L L} & {z z}_{W W 22 R R} & {z z}_{W W 22 L L} & {\overset{\cdot &Center Dot;}{z z}}_{W W 33 R R} & {\overset{\cdot &Center Dot;}{z z}}_{W W 33 L L} & {\overset{\cdot &Center Dot;}{z z}}_{W W 44 R R} & {\overset{\cdot &Center Dot;}{z z}}_{W W 44 L L} & {z z}_{W W 33 R R} & {z z}_{W W 33 L L} & {z z}_{W W 44 R R} & {z z}_{W W 44 L L} \end{matrix}]}^{T T}$

y＝[z φ θ z_FB φ_FB θ_FB z_RB φ_RB θ_RB]^T y＝[z φ θ z _FB φ _FB θ _FB z _RB φ _RB θ _RB ] ^T

注意到此时系统模型输出与实际车载故障诊断系统获得的输出并不一致，系统模型输出为车体与转向架质心位置出的垂向位移与角位移量，而实际车载故障诊断系统获得的输出为个传感器所在位置的垂向位移量，因此需要对车辆垂向悬挂系统模型做适当的变换,变换后的车辆悬挂系统空间状态方程如下： Note that the output of the system model is not consistent with the output obtained by the actual on-board fault diagnosis system. The output of the system model is the vertical displacement and angular displacement of the center of mass of the car body and bogie, while the output obtained by the actual on-board fault diagnosis system is The vertical displacement of each sensor position, so it is necessary to make an appropriate transformation to the vehicle vertical suspension system model, the transformed vehicle suspension system space state equation is as follows:

$\overset{\cdot \cdot}{x x} = = Ax Ax + + {B B}_{d d} d d$

y_cor＝TCx+TD_dd y _cor =TCx+TD _d d

系统输出： System output:

y_cor＝[z_FR z_FL z_RR z_RL z_{FB_FR} z_{FB_FL} z_{FB_RR} z_{FB_RL} z_{RB_FR} z_{RB_FL} z_{RB_RR} z_{RB_RL}]^T y _cor =[z _FR z _FL z _RR z _RL z _{FB_FR} z _{FB_FL} z _{FB_RR} z _{FB_RL} z _{RB_FR} z _{RB_FL} z _{RB_RR} z _{RB_RL} ] ^T

公式中符号l_e，w，l₁以及b₁可以从车辆参数表中查阅它们的含义以及取值。z_FR与z_FL分别代表车体右前角与左前角的垂向位移，z_{FB_FR}代表前转向架右前角垂向位移，其余符号可以按照相同的逻辑得知其含义。 Symbols l _e , w, l ₁ and b ₁ in the formula can refer to their meanings and values from the vehicle parameter table. z _FR and z _FL represent the vertical displacement of the right front corner and left front corner of the car body respectively, z _{FB_FR} represents the vertical displacement of the right front corner of the front bogie, and the meanings of other symbols can be obtained according to the same logic.

依据车辆悬挂系统的状态方程，利用MATLAB对所建立的城轨车辆垂向悬挂系统模型进行频域响应分析，获得伯德图如图2和图3所示。 According to the state equation of the vehicle suspension system, MATLAB is used to analyze the frequency domain response of the urban rail vehicle vertical suspension system model, and the Bode diagram is obtained as shown in Figure 2 and Figure 3. the

从图2可以看出，低频区域输入到输出的对数幅值约为-18分贝，即系统输出增益为0.125，4轮对共计8项垂向位置输入，与车辆运动实际情况相符。在输入频率约为5rad/秒，即0.80Hz时，系统输出达到谐振峰值，之后输出开始衰减，相位滞后，系统带宽为0Hz至1.59Hz。 It can be seen from Figure 2 that the logarithmic amplitude from input to output in the low-frequency region is about -18 decibels, that is, the system output gain is 0.125, and there are 8 vertical position inputs in total for the 4 wheelsets, which is consistent with the actual situation of vehicle movement. When the input frequency is about 5rad/second, that is, 0.80Hz, the system output reaches the resonance peak, and then the output begins to attenuate, the phase lags, and the system bandwidth is 0Hz to 1.59Hz. the

图3显示低频区域输入到输出的对数幅值约为-12分贝，即系统输出增益为0.25，对于转向架垂向位置输出，相应轮对每一车轮的系统输出增益皆为0.25，与车辆运动实际情况相符。在输入频率约为10rad/秒，即1.59Hz时，系统输出达到谐振峰值，之后输出开始衰减，相位滞后，系统带宽为0Hz至6.37Hz。 Figure 3 shows that the logarithmic amplitude from input to output in the low-frequency region is about -12 dB, that is, the system output gain is 0.25. For the output of the vertical position of the bogie, the system output gain of each wheel of the corresponding wheel pair is 0.25, which is consistent with the vehicle Match the reality of the movement. When the input frequency is about 10rad/second, that is, 1.59Hz, the system output reaches a resonance peak, and then the output begins to attenuate, the phase lags, and the system bandwidth is 0Hz to 6.37Hz. the

从得出的伯德图可以得出，所建立的城轨车辆垂向悬挂系统模型是合理可靠的，可以作为进行故障检测研究的基础。 From the obtained Bode diagram, it can be concluded that the established model of the vertical suspension system of urban rail vehicles is reasonable and reliable, and can be used as the basis for fault detection research. the

数据驱动的故障诊断方法要求所获取的数据能够充分的反映系统的动力学性能，并在系统发生故障时，承载足够的信息用于故障诊断，因此对系统输出的选择十分关键。从建模过程中我们可以看到，所提出的传感器布设方案能够充分地反映车辆运动状态，从而获得足够的数据用于故障检测与分离。 The data-driven fault diagnosis method requires that the acquired data can fully reflect the dynamic performance of the system, and carry enough information for fault diagnosis when the system fails, so the selection of the system output is very critical. From the modeling process, we can see that the proposed sensor layout scheme can fully reflect the vehicle motion state, so as to obtain enough data for fault detection and separation. the

2、传感器的布设方案 2. The layout of the sensor

传感器类型有三种，分别为车体加速度传感器、转向架加速度传感器与角加速度传感器，分别用于获取车体垂向加速度信号，转向架垂向加速度信号，以及车体与转向架的角加速度信号。 There are three types of sensors, namely car body acceleration sensor, bogie acceleration sensor and angular acceleration sensor, which are respectively used to obtain car body vertical acceleration signals, bogie vertical acceleration signals, and car body and bogie angular acceleration signals. the

如图4所示的车辆垂向悬挂系统模型，包括4个二系空气弹簧、4个二系阻尼器、8个一系弹簧以及8个一系阻尼器。在车辆进站时，要获得车辆悬挂系统的各种动力学特性来检测车辆悬挂系统的状态，12个无线传输加速度传感器被布设于车体与转向架以获取车辆运动状态信息。车体传感器布设在车底板的四个角，与车底板边缘距离均为400mm。转向架传感器布设在转向架四个轴箱的正上方。传感器旁边圈内数字代表传感器编号。如图4所示为传感器布置方案。在对车辆悬挂系统进行故障检测时，利用加速度传感器获得的加速度数据经过滤波和如下公式积分处理后就可获得速度位移等参数的大小。速度： $V = {&Integral;}_{0}^{t} adt + C,$ 位移： $X = {&Integral; &Integral;}_{0}^{t} adt + C .$ The vehicle vertical suspension system model shown in Figure 4 includes 4 secondary air springs, 4 secondary dampers, 8 primary springs and 8 primary dampers. When the vehicle enters the station, to obtain various dynamic characteristics of the vehicle suspension system to detect the state of the vehicle suspension system, 12 wireless transmission acceleration sensors are arranged on the vehicle body and bogie to obtain vehicle motion state information. The car body sensors are arranged at the four corners of the car floor, and the distance from the edge of the car floor is 400mm. The bogie sensors are arranged directly above the four axle boxes of the bogie. The number in the circle next to the sensor represents the sensor number. Figure 4 shows the layout of the sensors. When performing fault detection on the vehicle suspension system, the acceleration data obtained by the acceleration sensor can be obtained after filtering and integral processing of the following formula to obtain the magnitude of parameters such as velocity and displacement. speed: $V = {&Integral;}_{0}^{t} adt + C,$ Displacement: $x = {&Integral; &Integral;}_{0}^{t} adt + C .$

3、确定故障类型及幅值 3. Determine the fault type and magnitude

仿真实验中，论文分别考虑了二系空簧故障、阻尼器故障与一系弹簧故障。故障发生位置分别为左前方二系空簧、二系阻尼器，右后方二系空簧、二系阻尼器，与前、后转向架的向架上的左前方一系弹簧和右后方一系弹簧。对每一个发生故障的悬挂系部件又考虑了两种不同程度的性能衰减情况，便于建立故障库和检测库，即75%幅值性能衰减与80%的幅值性能衰减。详细故障信息见表2。 In the simulation experiment, the paper considers the secondary air spring fault, the damper fault and the primary spring fault respectively. The location of the fault is the secondary air spring and the secondary damper in the left front, the secondary air spring and the secondary damper in the right rear, and the left front primary spring and the right rear primary spring on the steering frame of the front and rear bogies. spring. For each faulty suspension component, two different degrees of performance attenuation are considered to facilitate the establishment of a fault library and a detection library, that is, 75% amplitude performance attenuation and 80% amplitude performance attenuation. See Table 2 for detailed fault information. the

表2 Table 2

确定故障类型及幅值以后，在已经建立的列车模型和Simulink仿真的基础上进行建库，分别建立故障库(包括悬挂系统正常和八个75%衰减情况)和检测库（包括八个80%衰减情况）。 After determining the fault type and magnitude, build a library on the basis of the established train model and Simulink simulation, respectively establish a fault library (including suspension system normal and eight 75% attenuation conditions) and a detection library (including eight 80% attenuation conditions) attenuation). the

4、Simulink仿真 4. Simulink simulation

在进行轨道车辆模型仿真时，作为系统输入的轨道激励的数值模拟方法得到了广泛的研究。在轨道不平顺中随机不平顺是使车辆产生响应的最主要输入函数，功率谱密度函数则是表述作为平稳随机过程的轨道不平顺的最重要和最常用的统计函数。由于仿真实验的需要，文中首先采用美国轨道五级谱模拟轨道的高低不平顺。高低不平顺功率谱密度函数为： Numerical simulation methods for rail excitations as system inputs have been extensively studied when performing rail vehicle model simulations. In the track irregularity, the random irregularity is the most important input function to make the vehicle respond, and the power spectral density function is the most important and commonly used statistical function to express the track irregularity as a stationary random process. Due to the needs of the simulation experiment, the paper first uses the American orbital pentatonic spectrum to simulate the unevenness of the track. The power spectral density function of high and low irregularities is:

${S S}_{v v} ((Ω Ω)) \frac{{kA kA}_{v v} {Ω Ω}_{c c}^{22}}{{Ω Ω}^{22} (({Ω Ω}^{22} + + {Ω Ω}_{c c}^{22}))},, {cm cm}^{22} / / ((rad rad / / m m))$

式中S(Ω)表示功率谱密度，Ω表示空间频率，A_v表示粗糙度常数，Ω_c代表截断频率，k为系数，一般取0.25。同时，为模拟轨道的水平不平顺（即轨道左右轨之间存在高度差，使车体产生侧滚运动），仿真时从同一轨道功率谱函数生成的轨道垂向不平顺序列中截取了不相同的两段分别作为左右轨的高低不平顺。 In the formula, S(Ω) represents the power spectral density, Ω represents the spatial frequency, A _v represents the roughness constant, Ω _c represents the cut-off frequency, and k is the coefficient, generally 0.25. At the same time, in order to simulate the horizontal irregularity of the track (that is, there is a height difference between the left and right rails of the track, which causes the vehicle body to roll), different tracks were extracted from the sequence of vertical irregularities generated by the power spectrum function of the same track during the simulation. The two sections of are respectively used as the height irregularity of the left and right rails.

5、基于PCA的故障检测方法及结果 5. Fault detection method and results based on PCA

基于PCA（主元分析）的故障诊断方法的原理是将多变量样本空间分解成由主元变量张成的较低维的投影子空间和一个相应的残差子空间，并分别在这两个空间中构造能够反映空间变化的统计量，然后将观测向量分别向两个子空间进行投影，并计算相应的统计量指标用于过程监控。 The principle of the fault diagnosis method based on PCA (Principal Component Analysis) is to decompose the multivariate sample space into a lower-dimensional projection subspace spanned by the principal component variables and a corresponding residual subspace, and respectively in these two The statistics that can reflect the spatial changes are constructed in the space, and then the observation vectors are projected to the two subspaces respectively, and the corresponding statistics indicators are calculated for process monitoring. the

基于标准PCA的故障检测算法一般包括以下三个步骤： The fault detection algorithm based on standard PCA generally includes the following three steps:

4)数据的获取与预处理 4) Data acquisition and preprocessing

无故障状态下，m维传感器经过N次采样之后获得数据矩阵X∈R^N×m。对X进行标准化处理，化为均值为0，方差为1的数据为多元数据序列，标准化处理的方式为： In the fault-free state, the m-dimensional sensor obtains the data matrix X∈R ^N×m after N times of sampling. Standardize X to make the data with mean value 0 and variance 1 a multivariate data sequence, and the standardization method is as follows:

${y the y}_{i i} = = \frac{{x x}_{i i} - - {x x}_{mean mean}}{{x x}_{std std}}$

其中x_i为原始数据的第i次采样值，x_mean为原始数据的均值，x_std为原始数据的方差，y_i为标准化后的数据。标准化处理后的数据矩阵表示为： Where x _i is the i-th sampling value of the original data, x _mean is the mean of the original data, x _std is the variance of the original data, and y _i is the standardized data. The standardized data matrix is expressed as:

$Y Y = = [\begin{matrix} {y the y}_{11}^{Y Y} \\ \cdot \cdot \\ \cdot &Center Dot; \\ \cdot &Center Dot; \\ {y the y}_{N N}^{T T} \end{matrix}] &Element; &Element; {R R}^{N N \times \times m m}$

5)协方差矩阵的分解 5) Decomposition of covariance matrix

协方差矩阵为： The covariance matrix is:

${Σ Σ}_{00} \approx \approx \frac{11}{N N} {Y Y}^{Y Y} Y Y$

采用SVD（奇异值分解）或者EVD（特征值分解），协方差矩阵∑₀被分解为如下形式： Using SVD (singular value decomposition) or EVD (eigenvalue decomposition), the covariance matrix ∑ ₀ is decomposed into the following form:

$\frac{11}{N N} {Y Y}^{T T} Y Y = = {PΛP PΛP}^{T T}$

$Λ Λ = = [\begin{matrix} {Λ Λ}_{pc pc} & 00 \\ 00 & {Λ Λ}_{res res} \end{matrix}]$

用表示协方差矩阵的第i个奇异值，有 use Represents the ith singular value of the covariance matrix, with

${Λ Λ}_{pc pc} = = diag diag \begin{matrix} (\begin{matrix} {σ σ}_{11}^{22} & \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot; & {σ σ}_{l l}^{22} \end{matrix}) \end{matrix}$

${Λ Λ}_{res res} = = diag diag \begin{matrix} (\begin{matrix} {σ σ}_{l l + + 11}^{22} & \cdot \cdot \cdot \cdot \cdot \cdot & {σ σ}_{m m}^{22} \end{matrix}) \end{matrix}$

${σ σ}_{11}^{22} &GreaterEqual; &Greater Equal; \cdot \cdot \cdot \cdot \cdot \cdot &GreaterEqual; &Greater Equal; {σ σ}_{l l}^{22} > > > > {σ σ}_{l l + + 11}^{22} \cdot \cdot \cdot \cdot \cdot \cdot &GreaterEqual; &Greater Equal; {σ σ}_{m m}^{22},, {PP PP}^{T T} = = {I I}_{m m \times \times m m}$

$[\begin{matrix} {P P}_{pc pc}^{T T} \\ {P P}_{res res}^{T T} \end{matrix}] [\begin{matrix} {P P}_{pc pc} & {p p}_{res res} \end{matrix}] = = [\begin{matrix} {I I}_{l l \times \times l l} & 00 \\ 00 & {I I}_{((m m - - l l)) \times \times ((m m - - l l))} \end{matrix}]$

6)在线故障检测 6) Online fault detection

在获得新的传感器采集数据x∈R^m之后，首先也将其标准化，得到数据y∈R^m。这样，故障检测指标SPE与T²就可以通过如下公式计算得到。 After obtaining the new sensor acquisition data x∈R ^m , it is also standardized at first to obtain the data y∈R ^m . In this way, the fault detection index SPE and ^T2 can be calculated by the following formula.

$SPE SPE = = {y the y}^{T T} {P P}_{res res} {P P}_{res res}^{T T} y the y$

${T T}^{22} = = {y the y}^{T T} {P P}_{pc pc} {ΛP ΛP}_{pc pc}^{T T} y the y$

检测指标SPE与T²都有与之相对应的报警阈值，分别为J_th,SPE与一般可由历史数据计算获得。于是故障报警逻辑为： The detection indicators SPE and T ² have corresponding alarm thresholds, which are respectively J _{th, SPE} and Generally, it can be calculated from historical data. So the fault alarm logic is:

本发明利用PCA故障检测法对检测库里的八种故障进行故障检测，效果明显，如图6列举了一系弹簧衰减80%和二系阻尼衰减80%的故障检测结果。 The present invention uses the PCA fault detection method to detect eight types of faults in the detection library, and the effect is obvious. Figure 6 lists the fault detection results of the primary spring attenuation of 80% and the secondary damping attenuation of 80%. the

6、脉冲冲击 6. Pulse impact

台阶的模拟是在上述轨道激励中加入阶跃信号，因此在以上轨道激励的基础上选取10s—11s区间作为脉冲区间。脉冲的幅值设置为0.1m,这个幅值使得列车悬挂系统的垂向运动影响比较明显，易于故障检测且较符合实际情况。最终轨道激励如图5（a）-5(b)所示。 The simulation of the step is to add a step signal to the above orbital excitation, so the 10s-11s interval is selected as the pulse interval on the basis of the above orbital excitation. The amplitude of the pulse is set to 0.1m, which makes the influence of the vertical motion of the train suspension system more obvious, easy to detect faults and more in line with the actual situation. The final orbital excitation is shown in Fig. 5(a)-5(b). the

7、特征提取 7. Feature extraction

本发明依据simulink中的车体模型，对行驶中的车辆给予脉冲冲击，获得12路传感器数据，分别提取仿真得到的所有故障类型的特征值。 According to the car body model in simulink, the invention gives pulse impact to the running vehicle, obtains 12 sensor data, and extracts the characteristic values of all fault types obtained by simulation respectively. the

（1）时域特征 (1) Time domain characteristics

平均值：C₁＝m₁；标准差：偏斜度： Mean: C ₁ =m ₁ ; Standard deviation: Skewness:

峰度： $C_{4} = m_{4} - {3 m}_{2}^{2} - {4 m}_{3} m_{1} + 12 m_{2} m_{1}^{2} - {6 m}_{1}^{4} .$ Kurtosis: $C_{4} = m_{4} - {3 m}_{2}^{2} - {4 m}_{3} m_{1} + 12 m_{2} m_{1}^{2} - {6 m}_{1}^{4} .$

其中，是力矩系数，x_i是i时刻的数据；N是数据的总数。 in, is the moment coefficient, x _i is the data at time i; N is the total number of data.

（2）频域特征 (2) Frequency domain characteristics

频域中心： $FC = \frac{{&Integral;}_{0}^{+ \infty} fs (f) df}{{&Integral;}_{0}^{+ \infty} s (f) df};$ 均方根频率： $RMSF = {[\frac{{&Integral;}_{0}^{+ \infty} f^{2} s (f) df}{{&Integral;}_{0}^{+ \infty} s (f) df}]}^{1 / 2};$ Frequency domain center: $FC = \frac{{&Integral;}_{0}^{+ \infty} fs (f) df}{{&Integral;}_{0}^{+ \infty} the s (f) df};$ RMS frequency: $RMSF = {[\frac{{&Integral;}_{0}^{+ \infty} f^{2} the s (f) df}{{&Integral;}_{0}^{+ \infty} the s (f) df}]}^{1 / 2};$

均方差频率： $RVF = {[\frac{{&Integral;}_{0}^{+ \infty} {(f - FC)}^{2} s (f) df}{{&Integral;}_{0}^{+ \infty} s (f) df}]}^{1 / 2} .$ mean square error frequency: $RVF = {[\frac{{&Integral;}_{0}^{+ \infty} {(f - FC)}^{2} the s (f) df}{{&Integral;}_{0}^{+ \infty} the s (f) df}]}^{1 / 2} .$

其中，s(f)是信号功率谱，f代表频率。 Among them, s(f) is the power spectrum of the signal, and f represents the frequency. the

8、基于D-S证据理论的故障分离方法 8. Fault isolation method based on D-S evidence theory

令 make

A＝[a_i1 a_i2 … a_ij],B＝[b_i1 b_i2 … b_ij] A＝[a _i1 a _i2 ... a _ij ], B＝[b _i1 b _i2 ... b _ij ]

其中A是故障库中某种故障的其中第i种特征值（时域或频域）向量，B是待检故障库中某种故障的相应的特征向量。j是传感器的个数。 Among them, A is the i-th eigenvalue (time domain or frequency domain) vector of a certain fault in the fault database, and B is the corresponding eigenvector of a certain fault in the fault database to be checked. j is the number of sensors. the

距离公式（计算上述故障的特征向量的距离）： Distance formula (calculate the distance of the eigenvectors of the above faults):

$d_{i} = {[\frac{1}{12} Σ_{j = 1}^{12} {(a_{ij} - b_{ij})}^{2}]}^{\frac{1}{2}}$ (i为特征数,j为传感器数) $d_{i} = {[\frac{1}{12} Σ_{j = 1}^{12} {(a_{ij} - b_{ij})}^{2}]}^{\frac{1}{2}}$ (i is the number of features, j is the number of sensors)

式中，d_i是A与B的第i种特征值向量之间的距离，最后得到一个7×1的列向量。 In the formula, d _i is the distance between the i-th eigenvalue vectors of A and B, and finally a 7×1 column vector is obtained.

$D D. = = [\begin{matrix} {d d}_{1111} & {d d}_{1212} & \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot; & {d d}_{11 M m} \\ {d d}_{21 twenty one} & {d d}_{22 twenty two} & \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot; & {d d}_{22 M m} \\ \cdot &Center Dot; & \cdot \cdot & \cdot \cdot & \cdot \cdot \\ \cdot \cdot & \cdot &Center Dot; & \cdot &Center Dot; & \cdot \cdot \\ \cdot &Center Dot; & \cdot &Center Dot; & \cdot \cdot & \cdot &Center Dot; \\ {d d}_{7171} & {d d}_{7272} & \cdot &Center Dot; \cdot &Center Dot; \cdot \cdot & {d d}_{77 M m} \end{matrix}]$

矩阵D就是某种待检故障的7种特征与故障库中的所有故障的特征之间的距离矩阵，M为故障特征库中的故障总数。距离越近就越相似，即矩阵D中d_im值越小，表明该待检故障与故障库中的第m种故障越相似，也就是能诊断出的故障最可能的是故障库的第m种故障。 Matrix D is the distance matrix between the seven features of a fault to be detected and all fault features in the fault database, and M is the total number of faults in the fault feature database. The closer the distance is, the more similar it is, that is, the smaller the value of d _im in matrix D, it indicates that the fault to be detected is more similar to the mth fault in the fault library, that is, the fault that can be diagnosed is most likely to be the mth fault in the fault library kind of failure.

距离相似性： Distance Similarity:

p′_im＝1/d_im p′ _im = 1/d _im

归一化： Normalized:

$P P = = [\begin{matrix} {p p}_{1111} & {p p}_{1212} & \cdot &Center Dot; \cdot &Center Dot; \cdot &Center Dot; & {p p}_{11 M m} \\ {p p}_{21 twenty one} & {p p}_{22 twenty two} & \cdot &Center Dot; \cdot &Center Dot; \cdot \cdot & {p p}_{22 M m} \\ \cdot &Center Dot; & \cdot \cdot & \cdot \cdot & \cdot &Center Dot; \\ \cdot &Center Dot; & \cdot &Center Dot; & \cdot &Center Dot; & \cdot &Center Dot; \\ \cdot \cdot & \cdot &Center Dot; & \cdot &Center Dot; & \cdot &Center Dot; \\ {p p}_{7171} & {p p}_{7272} & \cdot &Center Dot; \cdot &Center Dot; \cdot \cdot & {p p}_{77 M m} \end{matrix}] = = [\begin{matrix} {p p}_{11} \\ {p p}_{22} \\ \cdot \cdot \\ \cdot &Center Dot; \\ \cdot \cdot \\ {p p}_{77} \end{matrix}]$ ${Σ Σ}_{m m = = 11}^{M m} {p p}_{im im} = = 11,, ((i i = = 11,, . . . . . .,, 77,, M m = = 1616))$

最后得到的相似性矩阵P就是D-S证据理论中7组基本概率赋值函数。以下是D-S证据理论方法的步骤。 The resulting similarity matrix P is the seven basic probability assignment functions in D-S evidence theory. Following are the steps of the D-S evidence theory approach. the

D-S证据理论定义： D-S evidence theory definition:

设U为识别框架，则函数m:2^U→[0,1](2^U为U的幂集)满足m(φ)＝0，且时，称m为框架U上的基本可信度分配(basic probability assignment)；m(A)称为A的基本概率分配函数。 Let U be the recognition frame, then the function m: 2 ^U → [0,1] (2 ^U is the power set of U) satisfies m(φ)=0, and , we call m the basic probability assignment on the framework U; m(A) is called the basic probability distribution function of A.

如果m₁,m₂分别是两个定义在同一个辨识框架Θ上的两个不同基本概率赋值函数，定义m＝m₁⊕m₂为组合后的基本概率赋值函数 If m ₁ and m ₂ are two different basic probability assignment functions defined on the same identification frame Θ, define m=m ₁ ⊕m ₂ as the combined basic probability assignment function

其中，符号“⊕”表示Dempster组合规则可作用于两个或多个基本概率赋值函数上；式中 Among them, the symbol "⊕" indicates that the Dempster combination rule can act on two or more basic probability assignment functions; where

称其为规范化因子，用它可以度量两个或多个证据之间的冲突程度大小。k值越大，表示证据间冲突越大，融合后得到的信息量就越少。 It is called normalization factor, and it can be used to measure the degree of conflict between two or more evidences. The larger the k value, the greater the conflict between evidences, and the less information obtained after fusion. the

9、故障分离结果 9. Fault isolation results

基于前面PCA故障检测的基础，本文采用D-S进行故障分离。对于建立的检测库中的八种故障，通过D-S方法可以明显的匹配成功。如表3是对车体前左二系空簧衰减80%的分离结果。表中数据显示，检测库二系空簧故障与故障库故障匹配正确。 Based on the previous PCA fault detection, this paper adopts D-S for fault separation. For the eight types of faults in the established detection library, the D-S method can be clearly matched successfully. As shown in Table 3, the separation results of the attenuation of 80% of the front left secondary air spring of the car body. The data in the table shows that the faults of the secondary air spring in the detection library are matched correctly with the faults in the fault library. the

表3 table 3

显然，本发明的上述实施例仅仅是为清楚地说明本发明所作的举例，而并非是对本发明的实施方式的限定，对于所属领域的普通技术人员来说，在上述说明的基础上还可以做出其它不同形式的变化或变动，这里无法对所有的实施方式予以穷举，凡是属于本发明的技术方案所引伸出的显而易见的变化或变动仍处于本发明的保护范围之列。 Obviously, the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, rather than limiting the implementation of the present invention. For those of ordinary skill in the art, on the basis of the above description, they can also make It is not possible to exhaustively list all the implementation methods here, and all obvious changes or changes derived from the technical solutions of the present invention are still within the scope of protection of the present invention. the

Claims

1. The method for detecting the pulse impact of the key element component of the rail transit train suspension system is characterized by comprising the following steps of:

1) performing dynamic modeling on a rail vehicle suspension system component;

2) according to the model, sensors are arranged at corresponding positions of components of the vehicle suspension system, and motion state information of the vehicle is obtained through the sensors;

3) according to the acquired vehicle motion state information, in combination with train operation faults, carrying out simulation analysis on a rail vehicle model to obtain data of a sensor during train operation;

4) performing fault detection on all the sensor data obtained in the step 3 by using a PCA method, and judging the working state of the system to judge whether a fault exists or not;

5) adding step signals into the track excitation of the rail vehicle model, and performing simulation analysis to obtain data of a sensor when the train runs, so as to provide a reliable data base for fault separation;

6) respectively extracting characteristic values of all fault types obtained by simulation from the sensor data obtained in the step 5;

7) on the basis of obtaining the fault characteristic value, fault separation is carried out on the urban rail train suspension system by using an information fusion method of a D-S evidence theory, and the components and attenuation degree of the fault are accurately judged.

2. The method for detecting the impulse shock of the key element component of the rail transit train suspension system as claimed in claim 1, wherein the fault detection algorithm of the PCA generally comprises the following three steps:

1) data acquisition and preprocessing

Under the fault-free state, the m-dimensional sensor obtains a data matrix X belonging to R after N times of sampling^N×m. Normalizing X to obtain a mean value of 0 and data with a variance of 1 as a multivariate data sequence, wherein the normalization mode is as follows:

y_{i} = \frac{x_{i} - x_{mean}}{x_{std}}

wherein x_iIs the ith sample value, x, of the original data_meanIs the mean of the raw data, x_stdIs the variance of the original data, y_iIs normalized data. The normalized data matrix is represented as:

<math> <mrow> <mi>Y</mi> <mo>=</mo> <mfenced open='[' close=']'> <mtable> <mtr> <mtd> <msubsup> <mi>y</mi> <mn>1</mn> <mi>Y</mi> </msubsup> </mtd> </mtr> <mtr> <mtd> <mo>·</mo> </mtd> </mtr> <mtr> <mtd> <mo>·</mo> </mtd> </mtr> <mtr> <mtd> <mo>·</mo> </mtd> </mtr> <mtr> <mtd> <msubsup> <mi>y</mi> <mi>N</mi> <mi>T</mi> </msubsup> </mtd> </mtr> </mtable> </mfenced> <mo>&Element;</mo> <msup> <mi>R</mi> <mrow> <mi>N</mi> <mo>×</mo> <mi>m</mi> </mrow> </msup> </mrow> </math>

2) decomposition of covariance matrix

The covariance matrix is:

using SVD (singular value decomposition) or EVD (eigenvalue decomposition), covariance matrix sigma₀Is decomposed into the following forms:

by usingThe ith singular value representing the covariance matrix, has

<math> <mrow> <msubsup> <mi>σ</mi> <mn>1</mn> <mn>2</mn> </msubsup> <mo>&GreaterEqual;</mo> <mo>·</mo> <mo>·</mo> <mo>·</mo> <mo>&GreaterEqual;</mo> <msubsup> <mi>σ</mi> <mi>l</mi> <mn>2</mn> </msubsup> <mo>></mo> <mo>></mo> <msubsup> <mi>σ</mi> <mrow> <mi>l</mi> <mo>+</mo> <mn>1</mn> </mrow> <mn>2</mn> </msubsup> <mo>·</mo> <mo>·</mo> <mo>·</mo> <mo>&GreaterEqual;</mo> <msubsup> <mi>σ</mi> <mi>m</mi> <mn>2</mn> </msubsup> <mo>,</mo> <msup> <mi>PP</mi> <mi>T</mi> </msup> <mo>=</mo> <msub> <mi>I</mi> <mrow> <mi>m</mi> <mo>×</mo> <mi>m</mi> </mrow> </msub> </mrow> </math>

3) Online fault detection

When new sensor acquisition data x belongs to R^mThen, it is first normalized to obtain data y ∈ R^mThus, the fault detection indicators SPE and T²It can be calculated by the following formula

SPE = y^{T} P_{res} P_{res}^{T} y

The detection indexes SPE and T2 have corresponding alarm thresholds J respectively_th,SPEAndtypically obtained from historical data calculations, the fault alarm logic is then:

3. the method for detecting the impulse shock of the key element component of the rail transit train suspension system according to claim 1, wherein the fault separation method of the D-S evidence theory is characterized in that the following contents are carried out:

order to

A＝[a_i1 a_i2 … a_ij],B＝[b_i1 b_i2 … b_ij]

Wherein A is the ith time domain or frequency domain characteristic value vector of a certain fault in a fault library, B is the corresponding characteristic vector of the certain fault in the fault library to be detected, j is the number of sensors,

calculating the distance of the feature vector of the fault:

i is the number of features, j is the number of sensors),

in the formula (d)_iIs the distance between the ith eigenvalue vectors of a and B, resulting in a 7 x 1 column vector.

The matrix D is the distance matrix between 7 features of a certain fault to be detected and all the faulty features in the fault library, M is the total number of faults in the fault feature library, and the closer the distance is, the more similar the distance is, namely D in the matrix D_imThe smaller the value is, the more similar the fault to be detected to the mth fault in the fault library, that is, the closest diagnosed fault is the mth fault of the fault library.