WO2022141623A1 - 列车车室振动监测方法、振动信号特征库建立及应用方法 - Google Patents

列车车室振动监测方法、振动信号特征库建立及应用方法 Download PDF

Info

Publication number
WO2022141623A1
WO2022141623A1 PCT/CN2021/070117 CN2021070117W WO2022141623A1 WO 2022141623 A1 WO2022141623 A1 WO 2022141623A1 CN 2021070117 W CN2021070117 W CN 2021070117W WO 2022141623 A1 WO2022141623 A1 WO 2022141623A1
Authority
WO
WIPO (PCT)
Prior art keywords
vibration
train
signal
train compartment
vibration signal
Prior art date
Application number
PCT/CN2021/070117
Other languages
English (en)
French (fr)
Inventor
刘辉
于程名
李燕飞
秦进
张雷
尹诗
段铸
Original Assignee
中南大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中南大学 filed Critical 中南大学
Priority to US17/760,937 priority Critical patent/US11988547B2/en
Priority to AU2021413008A priority patent/AU2021413008B2/en
Priority to PCT/CN2021/070117 priority patent/WO2022141623A1/zh
Priority to JP2023513270A priority patent/JP7417342B2/ja
Publication of WO2022141623A1 publication Critical patent/WO2022141623A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H1/00—Measuring characteristics of vibrations in solids by using direct conduction to the detector
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B61—RAILWAYS
    • B61L—GUIDING RAILWAY TRAFFIC; ENSURING THE SAFETY OF RAILWAY TRAFFIC
    • B61L15/00—Indicators provided on the vehicle or train for signalling purposes
    • B61L15/0081—On-board diagnosis or maintenance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the invention belongs to the technical field of train compartment vibration identification, and particularly relates to a train compartment vibration monitoring method, and a vibration signal feature library establishment and application method.
  • the existing real-time monitoring method of train cabin vibration has the following shortcomings:
  • the patent with the publication number CN111044303A proposes a method for diagnosing abnormal vibration of a maglev train passenger compartment.
  • the method installs a large number of sensors in the train passenger compartment to realize the abnormal vibration monitoring of the train passenger compartment, but the patent does not involve the detection of unknown vibration sources. monitor.
  • the current monitoring method is mainly to directly monitor a large number of intrusive sensors installed on the train. Although the accuracy is high, it will inevitably lead to redundant and waste of sensors.
  • the patent with publication number CN110879102A proposes a vibration monitoring system for rail trains. The system sets up multiple vibration monitoring terminals in each car, and transmits data to the general control center in real time, so as to determine whether the vibration of the car is abnormal, but the The method does not involve monitoring the vibration source of the train, and requires many monitoring devices such as sensors, and the cost is high.
  • the purpose of the present invention is to provide a non-invasive vibration monitoring method for a train compartment, a method for establishing a vibration signal feature library and an application method in view of the shortcomings of the above-mentioned prior art, which can realize the known vibration source and unknown vibration source of the train without a large number of sensors Abnormal vibration monitoring of the source.
  • the technical scheme adopted in the present invention is:
  • Step 1 Under the conditions of normal vibration and abnormal vibration of the known vibration source outside the train compartment, pre-collect the horizontal, vertical and vertical vibration data M of multiple sub-measurement points in the train compartment, and pre-collect the total measurement points in the train compartment. Vibration data C in the lateral, longitudinal and vertical directions, pre-collect the vibration data S of the known vibration sources outside the train compartment in the lateral, longitudinal and vertical directions;
  • Step 2 Extract the phase information and amplitude information of the first 1-J harmonic signals of the vibration data M of each sub-measurement point, and extract the phase information and the first 1-J harmonic signals of the vibration data C of the total measuring point. Amplitude information, extract the phase information and amplitude information of the first 1 ⁇ J harmonic signals of each known vibration source;
  • Step 3 take the first 1 to J harmonic phase difference and harmonic amplitude ratio between the vibration data of each non-total measurement point and the vibration data of the total measurement point as input, and use the relative value of each non-total measurement point to the total measurement point.
  • the position information of the point in the horizontal, vertical and vertical directions is used as the output, and the model 1 is obtained by training the machine learning algorithm;
  • the non-total measurement points are composed of sub-measurement points and known vibration sources;
  • Step 4 collect the real-time vibration data CR of the total measuring points in the horizontal, vertical and vertical directions;
  • Step 5 Set the optimization objective with the phase offset of the vibration source relative to the total measurement point, the combined working state of each known vibration source, and the abnormal vibration type as independent variables, and execute the multi-objective optimization algorithm based on model 1 and model 2, and output Vibration monitoring result information of the train compartment, the vibration monitoring result information includes the position of the vibration source and/or the abnormal vibration type of the vibration source and/or the vibration source relative to the total measuring point.
  • the normal vibration of the known vibration sources outside the train includes two situations: the first is that all known vibration sources outside the train are vibrating normally, and the second is that any known vibration sources outside the train are vibrating normally.
  • the abnormal vibration of the known vibration source outside the train compartment includes two situations: the first is that any known vibration source outside the train compartment vibrates abnormally alone and the other known vibration sources do not vibrate , and the second type is that any known vibration source outside the train compartment vibrates abnormally alone and the other known vibration sources vibrate normally.
  • m( t) is the vibration signal value corresponding to the sub-measurement point,
  • c(t) is the vibration signal value corresponding to the total measurement point,
  • s(t) is the vibration signal value corresponding to the known vibration source;
  • L C is the label value of the vibration data corresponding to the total measurement point and
  • L C [l s ,l t ,l g , l d ]
  • L S is the label value of the vibration data corresponding to the known vibration source and
  • L S [l S , l t , l g , l d , l o2 ],
  • l M is the number
  • the weighted difference between the sum of the frequency domain characteristics of the vibration signals of each known vibration source and the frequency domain characteristics of each harmonic order of the vibration signal of the total measuring point is minimized as One of the optimization goals; the second optimization goal is to minimize the weighted variance of the difference between the harmonic orders of each known vibration source vibration signal.
  • step 6 is also included, converting and restoring the vibration monitoring result information obtained in step 5 into a real-time vibration source vibration signal in the time domain.
  • the present invention also provides a method for establishing a vibration signal feature library of a train compartment, which is characterized by including the non-invasive vibration monitoring method for a train compartment, and further comprising:
  • Step 7 extract the frequency domain feature, time domain feature and image feature of the real-time vibration source vibration signal, and perform feature fusion or feature reconstruction on the frequency domain feature and time domain feature of the real-time vibration source vibration signal;
  • Step 8 based on the frequency domain features, time domain features, image features, feature fusion or feature reconstruction results obtained in step 7, establish a train cabin vibration signal feature library.
  • it also includes using the information in the train compartment vibration signal feature library to perform online training on the model 1 and/or the model 2.
  • the present invention also provides an application method of the train compartment vibration signal feature library established by the described train compartment vibration signal feature library establishment method, which is characterized by including:
  • the relationship model between the vibration signal features in the train compartment vibration signal feature library and the train service performance characterization parameters is obtained by training. During the train operation, the vibration signal features in the train compartment vibration signal feature library obtained in real time are used as the relationship model. Input and output to obtain real-time train service performance characterization parameters.
  • the present invention also provides an application method of the train compartment vibration signal feature library established by the described train compartment vibration signal feature library establishment method, which is characterized by including:
  • the abnormal vibration of the vibration source at a certain moment in the future is predicted.
  • the present invention has the following beneficial effects:
  • a non-intrusive train compartment vibration monitoring and vibration source position calculation method is proposed.
  • the multi-objective optimization method is used to determine the optimal phase shift and the optimal vibration signal combination, and then realize the vibration through the relationship between the phase shift and the position information.
  • a non-intrusive vibration source signal acquisition and real-time update method of the multivariate vibration signal feature library is proposed, which can reconstruct the vibration source signal on the basis of only measuring the vibration signal of the total measuring point, so as to realize its multivariate vibration signal feature database. It is updated in real time, and the multi-dimensional vibration signal feature library can provide a data basis for the optimal decomposition of vibration signals, image recognition of vibration signals, and identification of abnormal vibration signals.
  • FIG. 1 is a flowchart of an embodiment of the present invention.
  • the interior of the train compartment will be affected by external vibration sources, resulting in vibration.
  • the vibration of the train compartment can reflect the operation of the train components in contact with the compartment, and on the other hand, it will affect the comfort and passenger comfort of passengers.
  • the non-invasive monitoring of the vibration of the train room can provide important support for determining the service performance of the train and the source of abnormal vibration, which is of great significance to ensure the safety of train operation and improve the comfort of passengers.
  • the present invention comprises the following steps:
  • Step 1 Intrusive pre-collection of multiple collection points of train cabin vibration data
  • the present invention needs to collect part of the initial training data in advance before the non-invasive vibration monitoring of the train compartment.
  • the initial training data is collected invasively with multiple measurement points:
  • Step 101 firstly, a plurality of sub-measurement points and a total measurement point are arranged in the train compartment.
  • the arrangement rule of the sub-measurement points is: for multiple transverse sections of the train, at least two vibration sensors for measuring the lateral vibration signal of the train compartment are arranged on each transverse section; for multiple longitudinal sections of the train, in At least two vibration sensors are arranged on each longitudinal section for measuring the longitudinal vibration signal of the train compartment; for multiple vertical sections of the train, at least two vibration sensors are arranged on each vertical section for measuring the vertical vibration of the train compartment Vibration sensor for vibration signal.
  • a total of 3 vibration sensors are arranged at the total measuring point, which are respectively used to measure the horizontal, vertical and vertical vibration signals at the total measuring point.
  • three vibration sensors are arranged at each known vibration source, which are respectively used to measure the horizontal, vertical and vertical vibration signals at the corresponding known vibration source.
  • Step 102 During the normal operation of the experimental train (that is, all known vibration sources outside the train compartment vibrate normally), the vibration signals of each sub-measurement point, the total measurement point vibration signal and the external known vibration source vibration signal of the train compartment are collected.
  • Step 103 Design a separate vibration experiment for each vibration source (that is, any known vibration source outside the train compartment vibrates normally and the other known vibration sources do not vibrate) and an abnormal vibration experiment for each vibration source. The latter is divided into abnormal vibration of each vibration source.
  • Single vibration experiment that is, any known vibration source outside the train compartment vibrates abnormally alone and the other known vibration sources do not vibrate
  • single vibration source abnormal vibration multi-vibration source mixed vibration experiment that is, any known vibration source outside the train compartment alone Abnormal vibration and normal vibration of other known vibration sources
  • the above label values are automatically assigned according to the vibration sensor ID of the returned signal.
  • the abnormal type label needs to be manually marked, but it will be assigned automatically after non-invasive feature collection is realized.
  • Step 2 Extracting the frequency domain features of the cabin vibration signal, extracting the phase information and amplitude information of the first 1-J harmonic signals of the vibration data M of each sub-measurement point, and extracting the first 1-J harmonic signals of the vibration data C of the total measurement point.
  • the phase information and amplitude information of the J harmonic signal are extracted, and the phase information and amplitude information of the first 1 to J harmonic signals of each known vibration source are extracted. details as follows:
  • the time-frequency domain characteristics of the vibration signals generated by different vibration sources are distinguishable from the harmonic phase shift, signal time shift and amplitude attenuation of the total vibration signal measuring points; the same vibration source corresponds to different abnormal vibration conditions.
  • Signal characteristics are also unique, and time-frequency domain analysis of the signal can improve the accuracy of non-invasive monitoring and feature acquisition.
  • Step 3 the data modeling of the relationship between the harmonic phase shift, amplitude attenuation and measuring point position of the cabin vibration signal.
  • step 3 the first 1 to J harmonic phase difference and harmonic amplitude ratio between the vibration data of each non-total measurement point and the vibration data of the total measurement point are used as input, and the relative value of each non-total measurement point to the total measurement point is used as the input.
  • the position information of the measuring points in the horizontal, vertical and vertical directions is used as the output, and the model 1 is obtained by training with the machine learning algorithm;
  • the phase difference is used as the input, and the amplitude ratio of the first 1 to J harmonics between the vibration data of each non-total measurement point and the vibration data of the total measurement point is used as the output, and the model 2 is obtained by training the machine learning algorithm;
  • the measuring points are composed of sub-measuring points and known vibration sources.
  • the invention adopts the data modeling method to analyze the relationship between the phase shift and amplitude attenuation of the vibration signals of different measuring points and the position of the measuring point. Under ideal conditions, there is a definite relationship between the phase offset and amplitude attenuation of the vibration signal and the signal propagation distance. However, due to the complex material structure of the train compartment, traditional methods often cannot reflect the real signal offset, attenuation and measurement points. Therefore, the present invention adopts the method of data modeling to analyze it.
  • Step 301 calculate the harmonic phase difference between the non-total measuring point signal (including the vibration signal of the vehicle compartment sub-measurement point and the known vibration source signal) and the vibration signal of the total measuring point Harmonic amplitude ratio u ki , positional relationship l o of non-measurement point signal and total measurement point signal, where k represents signal k, i represents harmonic order, and averages the harmonic phase differences of measurement points measured multiple times value, to obtain the harmonic phase difference and harmonic amplitude ratio of each harmonic order of the vibration signal of the sub-measurement point or the known vibration source relative to the vibration signal of the total measurement point
  • Step 302 data modeling of harmonic phase difference, harmonic amplitude ratio-total measurement point relative position relationship
  • the deep echo state network is used to model and describe the harmonic phase difference and harmonic amplitude ratio of each non-total measuring point signal relative to the total measuring point signal, and the relative position information of each non-total measuring point relative to the total measuring point.
  • the parameters of the network are set as follows, the number of nodes in the pool is set to 10, and the number of layers of the pool and the radius of the matrix spectrum are adaptively determined by 10-fold cross-validation, and the selection range is [1, 2, 3, ..., 10] and [ 0.1,0.2,...,0.9], select the parameters that can accurately reflect the above relationship, and finally construct the nonlinear relationship between the harmonic phase difference and harmonic amplitude ratio of each non-total measuring point and the relative position information of each non-total measuring point.
  • the relational models f 1z (I), f 1h (I), and f 1c (I) represent the relational models of longitudinal, lateral and vertical vibration signals, respectively.
  • Step 303 modeling the relationship between the harmonic phase difference of each non-total measuring point signal and the harmonic amplitude ratio
  • the deep echo state network is used to model and describe the relationship between the harmonic phase difference and the harmonic amplitude ratio of each non-total measuring point signal relative to the total measuring point signal.
  • the parameters of the deep echo state network are set as follows, and the number of storage pool nodes is set is 10, and 10-fold cross-validation is used to adaptively determine the number of layers and the matrix spectral radius of the pool.
  • step 304 the above steps 302 to 303 are repeated J times according to the harmonic order to obtain a corresponding relationship model of each harmonic order of the vibration signal.
  • Step 4 non-invasively collect the real-time vibration data CR of the total measuring points in the train compartment in the lateral, longitudinal and vertical directions
  • the total measuring point vibration sensors in the horizontal, vertical and vertical directions are installed at the total measuring points inside the train compartment, and the vibration signals of the total measuring points in the three directions are collected in real time, and the time stamp is recorded.
  • the real-time collected data is sent to the data storage module in 4G mode, and the data transmission interval is 1 minute.
  • Step 5 Non-invasive vibration source signal monitoring and phase shift determination based on multi-objective optimal phasor decomposition
  • Step 5 specifically includes:
  • Step 501 select the multi-objective gray wolf optimization method to construct the vibration source signal and its phase shift determination model, select the multi-objective optimization method and reasonably select the corresponding hyperparameters: adopt the multi-objective particle swarm optimization model, determine the corresponding parameters such as the maximum number of iterations is: 50, the number of gray wolves is 100, and the number of archives is 50.
  • the model feature library only adopts the feature library composed of the frequency domain features of the vibration signal of the vibration source.
  • Step 502 the optimization variables are different vibration sources and abnormal vibration types.
  • the independent variables mainly include: the phase offset of the vibration source relative to the total measurement point, the combined working state of each known vibration source, and the type of abnormal vibration.
  • the phase offset of the vibration source relative to the total measurement point is a continuous independent variable , which is used to optimize the appropriate phase.
  • the combined working state and abnormal vibration type of each known vibration source are discrete independent variables, which are used to optimize the appropriate vibration source signal source and abnormal vibration type.
  • the independent variables can be expressed as
  • D is used to limit the number of vibration sources, represents the phase shift
  • Step 503 according to the phase offset in the independent variable, use the relationship between the harmonic phase shift of each non-total measuring point vibration signal relative to the total measuring point vibration signal obtained in step 303 and the harmonic amplitude ratio in the vertical, horizontal and vertical directions.
  • Model f 2z (I), f 2h (I), f 2c (I) calculate the harmonic amplitude ratio u i of the total measuring point signal corresponding to each vibration source signal, and then obtain the harmonic amplitude corresponding to each vibration source signal wave phasor
  • Step 504 setting an optimization objective, to minimize the weighted difference between the sum of the frequency domain characteristics of the vibration signals of each known vibration source and the frequency domain characteristics of each harmonic order of the vibration signal of the total measuring point as one of the optimization objectives.
  • the second optimization goal is to minimize the weighted variance of the difference between the harmonic orders of each known vibration source vibration signal.
  • the optimization objective function is as follows:
  • F 1 and F 2 represent two objective functions
  • w j is the weight of each order harmonic
  • a is the total number of known vibration sources.
  • Step 505 start to perform multi-objective gray wolf optimization on the independent variables, calculate the optimization function values of all search results, and select non-dominated solutions and store them in the file.
  • Step 506 update the search path to generate a new argument scheme.
  • Step 508 take the independent variable that minimizes the objective function in the non-dominated solution set as the final solution.
  • Step 509 according to the selected optimal vibration source combination and the optimal phase offset, perform step 503 to obtain the corresponding harmonic amplitude ratio; take the optimal phase offset and the corresponding harmonic amplitude ratio as input, apply step 402
  • the harmonic phase difference, harmonic amplitude ratio and the positional relationship model between the non-total measuring point signal and the total measuring point signal trained in the training are f 1z (I), f 1h (I), f 1c (I) obtains the position of the vibration source that is currently vibrating.
  • Step 5010 output the vibration monitoring result information of the train compartment, the vibration monitoring result information includes the position of the vibration source that vibrates and/or the abnormal vibration type of the vibration source and/or the vibration source that vibrates relative to the total.
  • Step 5011 take corresponding vibration reduction measures according to the obtained abnormal vibration position to meet the comfort requirements of passengers; take corresponding fault early warning measures to avoid the occurrence of faults.
  • Step 6 time domain restoration of the real-time non-invasive vibration source signal, that is, converting and restoring the vibration monitoring result information obtained in step 5 into a real-time vibration source vibration signal in the time domain.
  • Step 601 based on the optimization result of step 5, restore the signal of each vibration source by using the harmonic phase shift and the harmonic amplitude ratio.
  • Step 5 restore the signal of each vibration source by using the harmonic phase shift and the harmonic amplitude ratio.
  • step 5 extract the frequency domain signal f of the corresponding vibration source of the pre-collected data feature library, and modify the pre-collected vibration source signal according to the following formula to obtain real-time
  • the frequency domain signal of the vibration source :
  • a sr and represents the frequency domain amplitude and harmonic phase of the real-time signal A so and Indicates the frequency domain amplitude and harmonic phase of the pre-collected vibration source signal, u so and Represents the harmonic amplitude ratio and harmonic phase shift of the pre-collected vibration source signal relative to the total measurement point, u sr and Indicates the optimal harmonic phase shift and harmonic amplitude ratio obtained by optimization.
  • the amplitude correction near each harmonic adopts the harmonic amplitude ratio corresponding to the harmonic.
  • Step 7 real-time non-invasive vibration source signal feature extraction and feature fusion or reconstruction
  • Step 701 Real-time vibration signal feature extraction
  • Signal image feature extraction Feature extraction is performed on the time domain and frequency domain images of the real-time vibration source signal to obtain image features.
  • Step 702 Real-time vibration signal feature fusion
  • the KPCA method is used to perform feature fusion or feature reconstruction on the time domain and frequency domain features.
  • the kernel function adopts a Gaussian kernel.
  • the cumulative contribution rate of the features is calculated, and the features with a cumulative contribution rate greater than 95% are taken as the feature fusion result. 10-fold cross-validation was performed on all feature data to determine the optimal number of features.
  • Step 8 Establish a non-invasive real-time updated multi-feature library of train cabin vibration signals
  • a multivariate feature library is established according to the applicable method of the feature and the characteristics of the feature:
  • the frequency domain characteristics include characteristics such as harmonic phasors, harmonic amplitudes, harmonic phase shifts, harmonic amplitude ratios, etc., which can further enrich the frequency domain feature library of vibration source signals. , and store it in the data storage module according to the data format of step 2 to complete the real-time update and supplement of the non-invasive vibration signal multi-objective optimization decomposition feature library. This signature library can be reapplied to non-invasive vibration monitoring.
  • a real-time update non-invasive vibration signal identification feature library can be established, which can be applied to artificial intelligence algorithms to identify vibration signals from different sources. identification.
  • the abnormal vibration signal characteristics in the abnormal vibration signal characteristics are designated and collected, and a real-time update non-invasive abnormal vibration signal identification characteristic library can be established.
  • the deep learning algorithm based on image learning can more accurately identify the vibration characteristics.
  • a huge image feature library of the vibration source of the train compartment can be constructed, which lays a solid foundation for the application of artificial intelligence image recognition methods. Base.
  • the fusion features of each vibration source signal can be obtained, and the fusion features are more effective in the process of vibration signal identification. Promote the application of artificial intelligence algorithms in this field.
  • Step 9 Multi-feature database application of non-intrusive real-time train vibration source signals
  • the relationship model between the vibration signal features in the train compartment vibration signal feature database and the parameters of the train service performance is established by using the experimental data.
  • the vibration signal features in the train compartment vibration signal feature library obtained in step 8 are used as the input of the relational model, and the real-time train service performance characterization parameters are obtained as the output, which can realize the real-time evaluation of the service performance of the train components.
  • the common time series prediction method is used to predict the abnormal vibration of the vibration source at a certain moment in the future, which can realize minute-level detection during the train operation. Prediction of train component failures is helpful for real-time control of train safety performance.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Mechanical Engineering (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
  • Burglar Alarm Systems (AREA)
  • Electric Propulsion And Braking For Vehicles (AREA)

Abstract

本发明公开了一种列车车室振动监测方法、振动信号特征库建立及应用方法，采用非侵入式列车车室振动监测方法，实际监测时仅需获取列车车室内总测点的振动信号，避免传感器冗余，节省成本，有利于列车轻量化，准确真实地反映出振源位置与相位偏移、幅值衰减之间的关系，能够监测未知振源的振动情况，能够识别振源的异常振动，从而为异常振动处理提供依据；能够在仅测量总测点振动信号的基础上实现振源信号的重建，从而实现其多元振动信号特征库的实时更新，且多元振动信号特征库能够为振动信号优化分解、振动信号图像辨识、异常振动信号辨识等提供数据基础。

Description

列车车室振动监测方法、振动信号特征库建立及应用方法 技术领域
本发明属于列车车室振动识别技术领域，特别涉及一种列车车室振动监测方法、振动信号特征库建立及应用方法。
背景技术
近年来，我国高速列车不断发展，在列车高速运行过程中，列车车室内部会受到来自外部振源的影响，产生振动，车室振动一方面能够反映与车室接触的列车部件的运行情况，另一方面会影响旅客舒适性和乘车体验。
现有的列车车室振动实时监测方法具有下述缺点：
第一，仅在关键部位安装传感器，不能够实现未知振源的监测。如，公开号为CN111044303A的专利提出了一种磁悬浮列车客室异常振动诊断方法，该方法在列车客室安装大量传感器，实现了对列车客室的异常振动监测，但该专利未涉及到对未知振源的监测。
第二，目前的监测方式主要为对列车侵入式安装大量传感器进行直接监测，虽然准确性高，但不可避免地会产生传感器冗余浪费的情况。如，公开号为CN110879102A的专利提出了一种轨道列车振动监控系统，该系统在每节车厢内设置多个振动监控端，并实时向总控制中心传输数据，能够判断车厢振动是否异常，但该方法没有涉及到对列车振源的监控，且所需传感器等监控设备较多，成本较高。
基于上述理由，针对列车车室进行非侵入式振动监测，并确定异常振动来源，对降低列车使用维护成本，保障列车运行安全和提高旅客舒适性具有重要意义。
发明内容
本发明的目的在于，针对上述现有技术的不足，提供一种列车车室非侵入式振动监测方法、振动信号特征库建立及应用方法，无需大量传感器，能够实现列车已知振源和未知振源的异常振动监测。
为解决上述技术问题，本发明所采用的技术方案是：
一种列车车室非侵入式振动监测方法，其特点是包括以下步骤：
步骤1，在列车车室外已知振源正常振动和异常振动条件下，分别预采集列车车室内多个分测点在横向、纵向、垂向的振动数据M，预采集列车车室内总测点在横向、纵向、垂向的振动数据C，预采集列车车室外各已知振源在横向、纵向、垂向的振动数据S；
步骤2，提取各分测点的振动数据M的前1～J次谐波信号的相位信息和幅值信息，提 取总测点的振动数据C的前1～J次谐波信号的相位信息和幅值信息，提取各已知振源的前1～J次谐波信号的相位信息和幅值信息；
步骤3，以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波相位差和谐波幅值比作为输入，以各非总测点相对于总测点在横向、纵向、垂向的位置信息作为输出，运用机器学习算法训练获得模型一；
以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波相位差作为输入，以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波幅值比作为输出，运用机器学习算法训练获得模型二；
其中，非总测点由分测点和已知振源组成；
步骤4，采集总测点在横向、纵向、垂向的实时振动数据CR；
步骤5，以振源相对于总测点的相位偏移、各已知振源组合工作状态、异常振动类型作为自变量，设定优化目标，基于模型一和模型二执行多目标优化算法，输出列车车室的振动监测结果信息，所述振动监测结果信息包括发生振动的振源的位置和/或发生振动的振源的异常振动类型和/或发生振动的振源相对于总测点的前1～J次谐波相位偏移和/或发生振动的振源相对于总测点的前1～J次谐波幅值比。
作为一种优选方式，所述步骤1中，列车车室外已知振源正常振动包括两种情形：第一种为列车车室外各已知振源均正常振动，第二种为列车车室外任一已知振源单独正常振动而其余已知振源不振动。
作为一种优选方式，所述步骤1中，列车车室外已知振源异常振动包括两种情形：第一种为列车车室外任一已知振源单独异常振动而其余已知振源不振动，第二种为列车车室外任一已知振源单独异常振动而其余已知振源正常振动。
作为一种优选方式，所述步骤1中，M＝[L M,m(t)]，C＝[L C,c(t)]，S＝[L S,s(t)]；m(t)为分测点对应的振动信号值，c(t)为总测点对应的振动信号值，s(t)为已知振源对应的振动信号值；L M为分测点对应振动数据的标签值且L M＝[l M,l t,l g,l d,l o1]，L C为总测点对应振动数据的标签值且L C＝[l s,l t,l g,l d]，L S为已知振源对应振动数据的标签值且L S＝[l S,l t,l g,l d,l o2]，l M为分测点的编号标签，l t为时间戳标签，l g为异常振动类型标签，l o1为分测点相对于总测点在横向、纵向或垂向的位置信息标签，l s为同时发生振动的已知振 源标签，l d为振动方向标签，l S为已知振源的编号标签，l o2为已知振源相对于总测点在横向、纵向或垂向的位置信息标签。
作为一种优选方式，所述步骤5中，以实现各已知振源振动信号频域特征之和与总测点振动信号各谐波阶次的频域特征之间的加权差值最小化为优化目标之一；以实现各已知振源振动信号各谐波阶次差值的加权方差最小化为优化目标之二。
进一步地，还包括步骤6，将步骤5中获得的振动监测结果信息转换还原为时域内的实时振源振动信号。
基于同一个发明构思，本发明还提供了一种列车车室振动信号特征库建立方法，其特点是包括所述的列车车室非侵入式振动监测方法，还包括：
步骤7，提取实时振源振动信号的频域特征、时域特征和图像特征，并对实时振源振动信号的频域特征和时域特征进行特征融合或特征重建；
步骤8，基于步骤7中获得的频域特征、时域特征、图像特征、特征融合或特征重建结果，建立列车车室振动信号特征库。
进一步地，还包括利用列车车室振动信号特征库中的信息对所述模型一和/或模型二进行在线训练。
基于同一个发明构思，本发明还提供了一种由所述的列车车室振动信号特征库建立方法建立的列车车室振动信号特征库的应用方法，其特点是包括：
训练获得列车车室振动信号特征库中的振动信号特征与列车服役性能表征参数的关系模型，在列车运行过程中，以实时获得的列车车室振动信号特征库中的振动信号特征作为关系模型的输入，输出获得实时列车服役性能表征参数。
基于同一个发明构思，本发明还提供了一种由所述的列车车室振动信号特征库建立方法建立的列车车室振动信号特征库的应用方法，其特点是包括：
利用列车车室振动信号特征库中的振动信号特征的时间序列信息，运用时间序列预测方法，预测未来某一时刻的振源异常振动情况。
与现有技术相比，本发明具有以下有益效果：
1)采用非侵入式列车车室振动监测方法，实际监测时仅需获取列车车室内总测点的振动信号，避免传感器冗余，节省成本，有利于列车轻量化。
2)对各分测点振动信号相对于总测点振动信号谐波相位偏移、谐波幅值衰减二者与相对位置信息之间的关系进行建模，能够实现上述关系的非线性建模，准确真实地反映出振源 位置与相位偏移、幅值衰减之间的关系，为非侵入式振动监测奠定基础，能够监测未知振源的振动情况。
3)提出了一种非侵入式列车车室振动监测和振源位置计算方法，采用多目标优化方法，确定最优相移、最优振动信号组合，进而通过相移与位置信息的关系实现振源位置计算。该方法在实际监测时仅需测量总测点的振动信号，且能够识别振源的异常振动，从而为异常振动处理提供依据。
4)提出了一种非侵入式振源信号采集和多元振动信号特征库实时更新方法，能够在仅测量总测点振动信号的基础上实现振源信号的重建，从而实现其多元振动信号特征库的实时更新，且多元振动信号特征库能够为振动信号优化分解、振动信号图像辨识、异常振动信号辨识等提供数据基础。
附图说明
图1为本发明一实施方式流程图。
具体实施方式
在列车高速运行过程中，列车车室内部会受到来自外部振源的影响，产生振动，车室振动一方面能够反映与车室接触的列车部件的运行情况，另一方面会影响旅客舒适性和乘车体验，针对车室振动进行非侵入式监测，能为确定列车服役性能和异常振动来源提供重要支撑，对保障列车运行安全和提高旅客舒适性都有重要意义。
如图1所示，本发明包括以下步骤：
步骤1，列车车室振动数据多采集点侵入式预采集
本发明在进行列车车室非侵入式振动监测前，需要预先采集部分初始训练数据。该初始训练数据采用多测点侵入式采集：
步骤101，首先在列车车室内布置多个分测点和一个总测点。其中，分测点的布置规则为：针对列车的多个横向断面，在每一个横向断面上布置至少两个用于测量列车车室横向振动信号的振动传感器；针对列车的多个纵向断面，在每一个纵向断面上布置至少两个用于测量列车车室纵向振动信号的振动传感器；针对列车的多个垂向断面，在每一个垂向断面上布置至少两个用于测量列车车室垂向振动信号的振动传感器。在总测点处共布置3个振动传感器，分别用于测量总测点处横向、纵向、垂向的振动信号。同时，针对列车车室外各已知振源，在每个已知振源处都布置3个振动传感器，分别用于测量对应的已知振源处横向、纵向、垂向的振动信号。
步骤102，在实验列车正常运行过程中(即列车车室外各已知振源均正常振动)，采集车室的各分测点振动信号、总测点振动信号和外部已知振源振动信号。
步骤103，设计各振源单独振动实验(即列车车室外任一已知振源单独正常振动而其余已知振源不振动)和各振源异常振动实验，后者分为各振源异常振动单独振动实验(即列车车室外任一已知振源单独异常振动而其余已知振源不振动)和单一振源异常振动多振源混合振动实验(即列车车室外任一已知振源单独异常振动而其余已知振源正常振动)，采集车室的各分测点振动信号、总测点振动信号和外部已知振源振动信号，且相同实验重复五次。
步骤104，将步骤102和步骤103获得的振动信号实时保存至数据存储模块，主要采用如下数据存储格式：列车车室内多个分测点在横向、纵向、垂向的振动数据M＝[L M,m(t)]，列车车室内总测点在横向、纵向、垂向的振动数据C＝[L C,c(t)]，列车车室外各已知振源在横向、纵向、垂向的振动数据S＝[L S,s(t)]；m(t)为分测点对应的振动信号值，c(t)为总测点对应的振动信号值，s(t)为已知振源对应的振动信号值；L M为分测点对应振动数据的标签值且L M＝[l M,l t,l g,l d,l o1]，L C为总测点对应振动数据的标签值且L C＝[l s,l t,l g,l d]，L S为已知振源对应振动数据的标签值且L S＝[l S,l t,l g,l d,l o2]，l M为分测点的编号标签(用于区分来自不同车室测点的信号)，l t为时间戳标签，l g为异常振动类型标签(异常振动类型由现有技术确定)，l o1为分测点相对于总测点在横向、纵向或垂向的位置信息标签，l s为同时发生振动的已知振源标签(用于标记当前正在振动的振源)，l d为振动方向标签(用于区分横向、纵向或垂向振动信号)，l S为已知振源的编号标签(用于区分来自不同振源测点的信号)，l o2为已知振源相对于总测点在横向、纵向或垂向的位置信息标签。
以上标签值根据传回信号的振动传感器ID进行自动赋予，异常类型标签需要手动标记，但实现非侵入特征采集后会自动赋予。
步骤2，车室振动信号频域特征提取，提取各分测点的振动数据M的前1～J次谐波信号的相位信息和幅值信息，提取总测点的振动数据C的前1～J次谐波信号的相位信息和幅值信息，提取各已知振源的前1～J次谐波信号的相位信息和幅值信息。具体如下：
为便于对信号进行进一步分析和实现信号的非侵入式监测，需对初始训练集信号的频 域特征进行分析提取。不同振源产生的振动信号其时频域特征和其与总振动信号测点的谐波相移、信号时移以及幅值衰减等特性具有可区分性；相同振源不同的异常振动情况对应的信号特征也具有独特性，对信号进行时频域分析能够提高非侵入监测和特征采集的准确性。
对步骤1测得的列车车室总测点振动信号、分测点振动信号和振源振动信号的信号值m(t)、c(t)和s(t)进行FFT变换，用于频域分析的振动时序信号为5s，即对每个样本信号以5s为间隔进行频域分析，可以得到各振动信号的1～J阶次谐波信号{m 1(t),m 2(t),…,m J(t)}、{c 1(t),c 2(t),…,c J(t)}、{s 1(t),s 2(t),…,s J(t)}，进而获得振动信号各谐波阶次的幅值信息和相位信息，即mA＝{ma 1(t),ma 2(t),…,ma J(t)}、cA＝{ca 1(t),…,ca 2(t),ca J(t)}、sA＝{sa 1(t),sa 2(t),…,sa J(t)}和
Figure PCTCN2021070117-appb-000001
其中m J表示车室测点信号的第J次谐波信号，s J表示振源信号的第J次谐波信号，c J表示某个振源信号的第J次谐波信号，后缀A和B分别表示幅值集合和相位集合，后缀a和
Figure PCTCN2021070117-appb-000002
分别表示幅值和相位值。
步骤3，车室振动信号谐波相移、幅值衰减与测点位置关系数据建模。
步骤3中，以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波相位差和谐波幅值比作为输入，以各非总测点相对于总测点在横向、纵向、垂向的位置信息作为输出，运用机器学习算法训练获得模型一；以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波相位差作为输入，以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波幅值比作为输出，运用机器学习算法训练获得模型二；其中，非总测点由分测点和已知振源组成。
具体如下：
本发明采用数据建模方式分析不同测点振动信号的相移、幅值衰减与测点位置的关系。在理想状态下，振动信号的相位偏移、幅值衰减与信号传播距离存在确定关系，但由于列车车室材料结构复杂等原因，往往传统方法不能够反映真实的信号偏移、衰减与测点位置的关系，因此本发明采用数据建模的方式对其进行分析。
步骤301，计算非总测点信号(包括车室分测点振动信号和已知振源信号)与总测点振动信号的谐波相位差
Figure PCTCN2021070117-appb-000003
谐波幅值比u ki、非测点信号与总测点信号的位置关系l o，其中k代表信号k，i代表谐波阶次，对多次测得的测点谐波相位差取平均值，得到分测点或已知振源相对于总测点振动信号的各谐波阶次谐波相位差和谐波幅值比
Figure PCTCN2021070117-appb-000004
Figure PCTCN2021070117-appb-000005
步骤302，谐波相位差、谐波幅值比-总测点相对位置关系数据建模
(1)读取横向、纵向、垂向各分测点及振源测点相对总测点的谐波相位差、谐波幅值比数据以及其相对总测点位置关系数据，数据集的80％作为训练集，20％作为测试集。
(2)以各非总测点信号相对于总测点信号的谐波相位差、谐波幅值比作为输入，
Figure PCTCN2021070117-appb-000006
I＝[I 1,I 2,...I k]，以各非总测点相对总测点位置作为输出O＝[l o1,l o2,...l ok]。
(3)采用深度回声状态网络对各非总测点信号相对总测点信号谐波相位差、谐波幅值比与各非总测点相对总测点位置信息进行建模描述，深度回声状态网络的参数设置如下，储蓄池节点数设置为10，采用10折交叉验证对储蓄池的层数和矩阵谱半径进行自适应确定，选择范围为[1,2,3,…,10]和[0.1,0.2,…,0.9]，选择能够准确反映上述关系的参数，最后构建各非总测点谐波相位差和谐波幅值比与各非总测点相对总测点位置信息的非线性关系模型f 1z(I),f 1h(I),f 1c(I)，分别代表纵向、横向、垂向振动信号关系模型。
步骤303，各非总测点信号谐波相位差与谐波幅值比关系数据建模
(1)读取横向、纵向、垂向各分测点及振源测点相对总测点的谐波相位差、谐波幅值比数据，数据集的80％作为训练集，20％作为测试集。
(2)以各非总测点信号相对于总测点信号的谐波相位差作为输入，
Figure PCTCN2021070117-appb-000007
以各非总测点相对总测点的谐波幅值比作为输出O＝[u 1,u 2,...u k]。
(3)采用深度回声状态网络对各非总测点信号相对总测点信号谐波相位差与谐波幅值比关系进行建模描述，深度回声状态网络的参数设置如下，储蓄池节点数设置为10，采用10折交叉验证对储蓄池的层数和矩阵谱半径进行自适应确定，选择范围为[1,2,3,…,10]和[0.1,0.2,…,0.9]，选择能够准确反映上述关系的参数，最后构建各非总测点振动信号相对总测点振动信号谐波相移与谐波幅值比在纵向、横向、垂向的关系模型f 2z(I),f 2h(I),f 2c(I)，分别代表纵向、横向、垂向振动信号关系模型。
步骤304，将上述步骤302～步骤303按照谐波阶次重复J次，得到振动信号各谐波阶次的相应关系模型。
步骤4，非侵入式采集列车车室总测点在横向、纵向、垂向的实时振动数据CR
在列车高速运行过程中，仅在列车车室内部总测点处安装横向、纵向、垂向三个方向的总测点振动传感器，实时采集三个方向总测点振动信号，并记录时间戳，采用4G方式将 实时采集数据发送至数据存储模块，数据传输间隔为1分钟。
步骤5，基于多目标优化相量分解的非入侵式振源信号监测及其相移确定
步骤5具体包括：
步骤501，选取多目标灰狼优化方法构建振源信号及其相移确定模型，选取多目标优化方法并合理选择相应超参数：采用多目标粒子群优化模型，确定相应的参数如最大迭代次数为50，灰狼数100，存档数50。模型特征库仅采用振源振动信号的频域特征所组成的特征库。
步骤502，优化变量为不同的振源与异常振动类型。
(1)自变量主要包括：振源相对于总测点的相位偏移、各已知振源组合工作状态、异常振动类型，其中，振源相对于总测点的相位偏移为连续自变量，用于优化合适的相位，各已知振源组合工作状态和异常振动类型为离散自变量，用于优化合适的振源信号来源和异常振动类型，自变量可表示为
Figure PCTCN2021070117-appb-000008
式中D用于限定振源个数，
Figure PCTCN2021070117-appb-000009
表示相移；
(2)相应的连续自变量为
Figure PCTCN2021070117-appb-000010
式中d∈[0.5,z+0.5),Lc i∈(0,1),Lg i∈(0,1)，
且D＝[d+1/2]，lc i＝[Lc i]，lg i＝[Lg i]，即D为d的向上取整，lc i、lg i分别为Lc i、Lg i的四舍五入取整，d、lc i、lg i无其他特别含义。
步骤503，根据自变量中的相位偏移，利用步骤303中获得的各非总测点振动信号相对总测点振动信号谐波相移与谐波幅值比在纵向、横向、垂向的关系模型f 2z(I),f 2h(I),f 2c(I)，计算每个振源信号对应的总测点信号部分的谐波幅值比u i，进而获得各振源信号对应的谐波相量
Figure PCTCN2021070117-appb-000011
Figure PCTCN2021070117-appb-000012
其中
Figure PCTCN2021070117-appb-000013
表示预采集数据特征库中第i个振源信号在总测点处产生的振动信号的第j次谐波相量，该相量是由振源信号的变比幅值与相位和相位偏移组成的向量。
步骤504，设定优化目标，以实现各已知振源振动信号频域特征之和与总测点振动信号 各谐波阶次的频域特征之间的加权差值最小化为优化目标之一；以实现各已知振源振动信号各谐波阶次差值的加权方差最小化为优化目标之二。优化目标函数如下：
min
Figure PCTCN2021070117-appb-000014
min
Figure PCTCN2021070117-appb-000015
其中，F 1和F 2表示两个目标函数，w j为各阶次谐波的权重，
Figure PCTCN2021070117-appb-000016
表示实时总测点信号的第j次谐波相量值，该相量由其相应的幅值和相位组成；a为各已知振源的总数。
步骤505，开始对自变量进行多目标灰狼优化，计算所有搜索结果的优化函数值，并选取非支配解存入档案中。
步骤506，更新搜索路径生成新的自变量方案。
步骤507，搜索次数It＝It+1，若It小于最大迭代次数，则返回步骤504；否则多目标优化算法结束，输出最终存档中的非支配解集NS。
步骤508，取非支配解集中使目标函数最小的自变量，作为最终方案。
步骤509，根据所选择的最优振源组合以及最优相位偏移，执行步骤503得到对应的谐波幅值比；以最优相位偏移和对应谐波幅值比为输入，应用步骤402中训练好的各非总测点信号与总测点信号谐波相位差、谐波幅值比和非总测点信号与总测点位置关系模型f 1z(I),f 1h(I),f 1c(I)得到目前正在发生振动的振源位置。
步骤5010，输出列车车室的振动监测结果信息，所述振动监测结果信息包括发生振动的振源的位置和/或发生振动的振源的异常振动类型和/或发生振动的振源相对于总测点的前1～J次谐波相位偏移和/或发生振动的振源相对于总测点的前1～J次谐波幅值比。
步骤5011，根据得到的振动异常位置采取相应的振动削减措施，以满足乘客的舒适性要求；采取相应的故障预警措施，避免故障发生。
步骤6，实时非入侵式振源信号时域还原，即将步骤5中获得的振动监测结果信息转换还原为时域内的实时振源振动信号。
步骤601，基于步骤5的优化结果，利用谐波相移、谐波幅值比对各振源信号进行还原。首先根据步骤5中所得最优结果的(l c1,l c2,……)标签，提取预采集数据特征库相应振源的频域信号f，根据如下公式对预采集振源信号进行修正得到实时振源频域信号：
Figure PCTCN2021070117-appb-000017
其中A sr和
Figure PCTCN2021070117-appb-000018
表示实时信号的频域幅值和谐波相位，A so和
Figure PCTCN2021070117-appb-000019
表示预采集的振源信号频域幅值和谐波相位，u so和
Figure PCTCN2021070117-appb-000020
表示预采集的振源信号相对总测点谐波幅值比和谐波相移，u sr和
Figure PCTCN2021070117-appb-000021
表示优化得到的最优谐波相移和谐波幅值比。频域信号修正过程中，每个谐波附近的幅值修正采用该谐波对应的谐波幅值比。最终可得到实时振源信号s(f)。
步骤602，采用IFFT技术对实时振动信号频域值进行傅里叶逆变换，得到时域实时振源信号，s(t)＝IFFT(s(f))。
步骤7，实时非侵入式振源信号特征提取及特征融合或重建
步骤701：实时振动信号特征提取
(1)信号频域特征提取：针对步骤6得到的实时振源信号重复步骤2所提到的频域特征提取方法，提取其频域特征。
(2)信号时域特征提取：针对实时振源信号，提取其时域特征，例如最大振幅、时间偏移、5s均值等。
(3)信号图像特征提取：针对实时振源信号的时域和频域图像进行特征提取，获得图像特征。
步骤702：实时振动信号特征融合
采用KPCA方法对时域、频域特征进行特征融合或特征重建，核函数采用高斯核，在特征融合过程中，计算特征的累计贡献率，取累计贡献率大于95％的特征作为特征融合结果。对所有特征数据进行10折交叉验证以确定最优特征数。
步骤8，建立非侵入实时更新的列车车室振动信号多元特征库
基于步骤7提取的信号特征根据特征的适用方法和特征的特性建立多元特征库：
(1)建立非侵入式振动信号优化分解特征库
上述步骤获得的各实时振源信号特征中，其频域特征包括谐波相量、谐波幅值、谐波相移、谐波幅值比等特征，可进一步丰富振源信号频域特征库，按照步骤2的数据格式存入 数据存储模块，完成对非侵入式振动信号多目标优化分解特征库的实时更新和补充。该特征库可再次应用到非侵入式振动监测中。
(2)建立非侵入式振动信号辨识特征库
根据上述步骤获得的不同位置实时振源信号特征，包括其频域特征、时域特征、融合特征可建立实时更新的非侵入式振动信号辨识特征库，可应用于人工智能算法对不同来源振动信号的辨识。
(3)建立非侵入式异常振动信号辨识特征库
根据上述步骤获得的不同实时振源信号特征，对其中的异常振动信号特征进行指定收集，可建立实时更新的非侵入式异常振动信号辨识特征库。
(4)建立非侵入实时更新的列车车室振动图像识别特征库
基于图像学习的深度学习算法能够更准确的对振动特征进行辨识，根据实时非侵入采集的振源信号可以构建出一个庞大的列车车室振源图像特征库，为人工智能图像识别方法的应用奠定基础。
(5)建立非侵入实时更新的列车车室振动融合特征、重构特征数据库
依据步骤7中的特征融合、特征重构方法可以得到各振源信号的融合特征，融合特征在进行振动信号辨识过程中效果更明显，建立非侵入式列车车室振动信号融合特征库有助于促进人工智能算法在本领域的应用。
步骤9，非侵入实时列车振源信号多元特征库多元应用
(1)列车部件服役性能实时评估
采用实验数据建立列车车室振动信号特征库中的振动信号特征与列车服役性能表征参数的关系模型，模型建立方法可采用支持向量机、极限学习机、人工神经网络、长短时深度神经网络等，在列车运行过程中，以基于步骤8获得的列车车室振动信号特征库中的振动信号特征作为关系模型的输入，输出获得实时列车服役性能表征参数，可实现列车部件服役性能的实时评估。
(2)列车部件故障预测
基于步骤8建立的列车车室振动信号特征库中的振动信号特征的时间序列信息，采用常用时间序列预测方法，预测未来某一时刻的振源异常振动情况，可实现列车运行过程中分钟级的列车部件故障预测，有助于列车安全性能实时管控。
上面结合附图对本发明的实施例进行了描述，但是本发明并不局限于上述的具体实施方式，上述的具体实施方式仅仅是示意性的，而不是局限性的，本领域的普通技术人员在本发明的 启示下，在不脱离本发明宗旨和权利要求所保护的范围情况下，还可做出很多形式，这些均属于本发明的保护范围之内。

Claims (10)

  1. 一种列车车室非侵入式振动监测方法，其特征在于，包括以下步骤：
    步骤1，在列车车室外已知振源正常振动和异常振动条件下，分别预采集列车车室内多个分测点在横向、纵向、垂向的振动数据M，预采集列车车室内总测点在横向、纵向、垂向的振动数据C，预采集列车车室外各已知振源在横向、纵向、垂向的振动数据S；
    步骤2，提取各分测点的振动数据M的前1～J次谐波信号的相位信息和幅值信息，提取总测点的振动数据C的前1～J次谐波信号的相位信息和幅值信息，提取各已知振源的前1～J次谐波信号的相位信息和幅值信息；
    步骤3，以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波相位差和谐波幅值比作为输入，以各非总测点相对于总测点在横向、纵向、垂向的位置信息作为输出，运用机器学习算法训练获得模型一；
    以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波相位差作为输入，以各非总测点的振动数据与总测点的振动数据之间的前1～J次谐波幅值比作为输出，运用机器学习算法训练获得模型二；
    其中，非总测点由分测点和已知振源组成；
    步骤4，采集总测点在横向、纵向、垂向的实时振动数据CR；
    步骤5，以振源相对于总测点的相位偏移、各已知振源组合工作状态、异常振动类型作为自变量，设定优化目标，基于模型一和模型二执行多目标优化算法，输出列车车室的振动监测结果信息，所述振动监测结果信息包括发生振动的振源的位置和/或发生振动的振源的异常振动类型和/或发生振动的振源相对于总测点的前1～J次谐波相位偏移和/或发生振动的振源相对于总测点的前1～J次谐波幅值比。
  2. 如权利要求1所述的列车车室非侵入式振动监测方法，其特征在于，所述步骤1中，列车车室外已知振源正常振动包括两种情形：第一种为列车车室外各已知振源均正常振动，第二种为列车车室外任一已知振源单独正常振动而其余已知振源不振动。
  3. 如权利要求1所述的列车车室非侵入式振动监测方法，其特征在于，所述步骤1中，列车车室外已知振源异常振动包括两种情形：第一种为列车车室外任一已知振源单独异常振动而其余已知振源不振动，第二种为列车车室外任一已知振源单独异常振动而其余已知振源正常振动。
  4. 如权利要求1所述的列车车室非侵入式振动监测方法，其特征在于，所述步骤1中，M＝[L M,m(t)]，C＝[L C,c(t)]，S＝[L S,s(t)]；m(t)为分测点对应的振动信号值，c(t) 为总测点对应的振动信号值，s(t)为已知振源对应的振动信号值；L M为分测点对应振动数据的标签值且L M＝[l M,l t,l g,l d,l o1]，L C为总测点对应振动数据的标签值且L C＝[l s,l t,l g,l d]，L S为已知振源对应振动数据的标签值且L S＝[l S,l t,l g,l d,l o2]，l M为分测点的编号标签，l t为时间戳标签，l g为异常振动类型标签，l o1为分测点相对于总测点在横向、纵向或垂向的位置信息标签，l s为同时发生振动的已知振源标签，l d为振动方向标签，l S为已知振源的编号标签，l o2为已知振源相对于总测点在横向、纵向或垂向的位置信息标签。
  5. 如权利要求4所述的列车车室非侵入式振动监测方法，其特征在于，所述步骤5中，以实现各已知振源振动信号频域特征之和与总测点振动信号各谐波阶次的频域特征之间的加权差值最小化为优化目标之一；以实现各已知振源振动信号各谐波阶次差值的加权方差最小化为优化目标之二。
  6. 如权利要求1至5任一项所述的列车车室非侵入式振动监测方法，其特征在于，还包括步骤6，将步骤5中获得的振动监测结果信息转换还原为时域内的实时振源振动信号。
  7. 一种列车车室振动信号特征库建立方法，其特征在于，包括如权利要求6所述的列车车室非侵入式振动监测方法，还包括：
    步骤7，提取实时振源振动信号的频域特征、时域特征和图像特征，并对实时振源振动信号的频域特征和时域特征进行特征融合或特征重建；
    步骤8，基于步骤7中获得的频域特征、时域特征、图像特征、特征融合或特征重建结果，建立列车车室振动信号特征库。
  8. 如权利要求7所述的列车车室振动信号特征库建立方法，其特征在于，还包括利用列车车室振动信号特征库中的信息对所述模型一和/或模型二进行在线训练。
  9. 一种由权利要求7所述的列车车室振动信号特征库建立方法建立的列车车室振动信号特征库的应用方法，其特征在于，包括：
    训练获得列车车室振动信号特征库中的振动信号特征与列车服役性能表征参数的关系模型，在列车运行过程中，以实时获得的列车车室振动信号特征库中的振动信号特征作为关系模型的输入，输出获得实时列车服役性能表征参数。
  10. 一种由权利要求7所述的列车车室振动信号特征库建立方法建立的列车车室振动信号特征库的应用方法，其特征在于，包括：
    利用列车车室振动信号特征库中的振动信号特征的时间序列信息，运用时间序列预测方法，预测未来某一时刻的振源异常振动情况。
PCT/CN2021/070117 2021-01-04 2021-01-04 列车车室振动监测方法、振动信号特征库建立及应用方法 WO2022141623A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
US17/760,937 US11988547B2 (en) 2021-01-04 2021-01-04 Train compartment vibration monitoring method and vibration signal feature library establishment and application methods
AU2021413008A AU2021413008B2 (en) 2021-01-04 2021-01-04 Train compartment vibration monitoring method, vibration signal feature library establishment method and vibration signal feature library application method
PCT/CN2021/070117 WO2022141623A1 (zh) 2021-01-04 2021-01-04 列车车室振动监测方法、振动信号特征库建立及应用方法
JP2023513270A JP7417342B2 (ja) 2021-01-04 2021-01-04 列車車室の振動監視方法、振動信号特徴ライブラリの構築及び応用方法

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2021/070117 WO2022141623A1 (zh) 2021-01-04 2021-01-04 列车车室振动监测方法、振动信号特征库建立及应用方法

Publications (1)

Publication Number Publication Date
WO2022141623A1 true WO2022141623A1 (zh) 2022-07-07

Family

ID=82258831

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/070117 WO2022141623A1 (zh) 2021-01-04 2021-01-04 列车车室振动监测方法、振动信号特征库建立及应用方法

Country Status (4)

Country Link
US (1) US11988547B2 (zh)
JP (1) JP7417342B2 (zh)
AU (1) AU2021413008B2 (zh)
WO (1) WO2022141623A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116256054A (zh) * 2023-05-15 2023-06-13 广东电网有限责任公司阳江供电局 一种桥臂电抗器的故障监测方法、系统、设备和介质
CN117272022A (zh) * 2023-09-19 2023-12-22 小谷粒(广州)母婴用品有限公司 一种mems振荡器的检测方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103250107A (zh) * 2010-12-31 2013-08-14 中国科学院自动化研究所 用于设备故障检测的智能检测系统及检测方法
CN107192444A (zh) * 2017-07-11 2017-09-22 国电联合动力技术有限公司 智能输出式振动监测装置和包括该装置的系统及监测方法
CN108750856A (zh) * 2017-12-13 2018-11-06 浙江新再灵科技股份有限公司 一种基于振动分析的电梯平台检测系统及方法
CN109353376A (zh) * 2018-10-24 2019-02-19 西安英特迈思信息科技有限公司 轨道车辆监测系统及其监测方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5306947B2 (ja) 2009-09-03 2013-10-02 公益財団法人鉄道総合技術研究所 鉄道車両の状態監視システム
WO2016115443A1 (en) * 2015-01-16 2016-07-21 International Electronic Machines Corp. Abnormal vehicle dynamics detection
JP6557110B2 (ja) 2015-10-13 2019-08-07 公益財団法人鉄道総合技術研究所 状態診断装置及びプログラム
JP6830028B2 (ja) 2017-04-24 2021-02-17 日本車輌製造株式会社 鉄道車両の振動表示システム
EP3667337B1 (en) 2018-12-14 2024-02-28 Metro de Madrid, S.A. Monitoring device for monitoring catenary-pantograph interaction in railway vehicles
US11429900B1 (en) * 2021-10-26 2022-08-30 Tractian Limited Systems and methods for automatic detection of error conditions in mechanical machines

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103250107A (zh) * 2010-12-31 2013-08-14 中国科学院自动化研究所 用于设备故障检测的智能检测系统及检测方法
CN107192444A (zh) * 2017-07-11 2017-09-22 国电联合动力技术有限公司 智能输出式振动监测装置和包括该装置的系统及监测方法
CN108750856A (zh) * 2017-12-13 2018-11-06 浙江新再灵科技股份有限公司 一种基于振动分析的电梯平台检测系统及方法
CN109353376A (zh) * 2018-10-24 2019-02-19 西安英特迈思信息科技有限公司 轨道车辆监测系统及其监测方法

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116256054A (zh) * 2023-05-15 2023-06-13 广东电网有限责任公司阳江供电局 一种桥臂电抗器的故障监测方法、系统、设备和介质
CN116256054B (zh) * 2023-05-15 2023-08-04 广东电网有限责任公司阳江供电局 一种桥臂电抗器的故障监测方法、系统、设备和介质
CN117272022A (zh) * 2023-09-19 2023-12-22 小谷粒(广州)母婴用品有限公司 一种mems振荡器的检测方法

Also Published As

Publication number Publication date
AU2021413008A1 (en) 2023-03-09
AU2021413008B2 (en) 2024-05-16
US11988547B2 (en) 2024-05-21
JP2023538686A (ja) 2023-09-08
JP7417342B2 (ja) 2024-01-18
US20240044695A1 (en) 2024-02-08

Similar Documents

Publication Publication Date Title
CN112395959B (zh) 基于音频特征的电力变压器故障预测与诊断方法及系统
Wang et al. Remaining life prediction method for rolling bearing based on the long short-term memory network
Chen et al. A novel deep learning method based on attention mechanism for bearing remaining useful life prediction
WO2021232655A1 (zh) 一种基于振动特征的高压并联电抗器机械状态评估方法
Guo et al. Remaining useful life prediction for rolling bearings using EMD-RISI-LSTM
CN105241660B (zh) 基于健康监测数据的高铁大型桥梁性能测试方法
An et al. Wind farm power prediction based on wavelet decomposition and chaotic time series
WO2022141623A1 (zh) 列车车室振动监测方法、振动信号特征库建立及应用方法
CN102053016B (zh) 旋转机械滚动轴承的无线振动监测系统
CN102520697B (zh) 一种远程协同诊断的现场信息预处理方法
CN110308002B (zh) 一种基于地面检测的城轨列车悬挂系统故障诊断方法
CN106644162B (zh) 基于邻域保持嵌入回归算法的环网柜线芯温度软测量方法
CN102062832B (zh) 基于微扰动信号低频振荡模式辨识的电力系统在线预警方法
CN108256556A (zh) 基于深度信念网络的风力发电机组齿轮箱故障诊断方法
CN110457786B (zh) 基于深度置信网络的卸船机关联规则故障预测模型方法
CN112800855B (zh) 一种列车转向架非侵入式实时故障监测方法
CN108053110A (zh) 一种基于pmu数据的变压器状态在线诊断方法
CN116380445B (zh) 基于振动波形的设备状态诊断方法及相关装置
CN114021932A (zh) 风电机组的能效评价与诊断方法、系统及介质
CN112816052B (zh) 列车车室振动监测方法、振动信号特征库建立及应用方法
Niu et al. Operation performance evaluation of elevators based on condition monitoring and combination weighting method
CN117371207A (zh) 一种特高压换流阀状态评价方法、介质及系统
Li et al. Non-Gaussian non-stationary wind pressure forecasting based on the improved empirical wavelet transform
Li et al. Transformer-based meta learning method for bearing fault identification under multiple small sample conditions
Bhardwaj et al. Estimation of lifespan of diesel locomotive engine

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 17760937

Country of ref document: US

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21912406

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2023513270

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2021413008

Country of ref document: AU

Date of ref document: 20210104

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21912406

Country of ref document: EP

Kind code of ref document: A1