Pantograph running state monitoring and analyzing method and system

Abstract

The invention discloses a pantograph running state monitoring and analyzing method and system, and relates to the field of rail train monitoring. According to the main scheme, vibration data, contact force data, contact point temperature data and abrasion data are comprehensively analyzed; whether a pantograph pressure control instruction is triggered or not, whether a train speed limiting instruction is triggered or not and whether the pantograph has an operation fault or not are judged, a passive early warning strategy is changed into active regulation and control, the operation state of the pantograph can be evaluated more comprehensively, the frequency and cost of manual detection are reduced through the intelligent monitoring and prediction technology, and the working efficiency is improved. The evaluation accuracy is improved, and meanwhile, the maintenance cost and the safety risk caused by fault expansion are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of rail train monitoring, and in particular to a pantograph operation status monitoring and analysis method and system. Background Art

[0002] In the rail transit system, the pantograph has the key task of obtaining electrical energy from the contact network and transmitting it to the train. The stability of its operating state is directly related to the reliable operation of the train and the safety and operational efficiency of the entire rail transit system. Therefore, developing an effective pantograph operating state monitoring and analysis method is of great significance to ensure the safety of train operation and improve transportation efficiency.

[0003] At present, the monitoring of the pantograph operating status mainly relies on traditional detection technology. For example, Chinese patent document CN204495300U discloses a pantograph operating status monitoring device. By collecting image data information of the pantograph, the image data information of the pantograph is obtained, and the pantograph size information such as the thickness of the pantograph slide plate is detected in a timely, convenient and rapid manner, thereby ensuring that the pantograph is in normal operating condition.

[0004] Although the existing technology can reflect the operating status of the pantograph to a certain extent, it has the following defects: 1. It only relies on a single type of data, and the data evaluation mode is single, which makes it difficult to fully reflect the operating status of the pantograph. 2. It adopts simple statistical analysis or threshold judgment, lacks in-depth mining and comprehensive analysis of data, resulting in limited accuracy and reliability of the diagnosis results. 3. Traditional methods have time errors in multi-data processing and analysis, making it difficult to achieve accurate evaluation of the operating status of the pantograph. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the deficiencies in the prior art, the present invention provides a method for monitoring and analyzing the operating status of a pantograph, which at least solves the problem in the prior art that the reference parameter type is single and the accuracy of the pantograph operating status assessment is insufficient.

[0007] (II) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: A pantograph operation status monitoring and analysis method, comprising:

[0009] Step 1: synchronously collect the pantograph vibration data, contact force data, contact point temperature data and wear data through a multi-modal sensor system;

[0010] Step 2: Time alignment and data preprocessing of the collected data;

[0011] Step 3: Perform comprehensive calculations on the processed vibration data, contact force data, and contact point temperature data to generate the dynamic contact force of the pantograph, and determine whether to trigger the pantograph pressure control instruction based on the dynamic contact force;

[0012] Step 4: Analyze the vibration data and dynamic contact force, and calculate the hard point impact probability; determine whether to trigger the train speed limit instruction according to the hard point impact probability;

[0013] Step 5: The wear data, contact point temperature data and contact force data are comprehensively analyzed to obtain the predicted damage depth; the wear area value of the pantograph friction area is collected through the quantum dot fluorescence monitoring system; the dynamic contact force, hard point impact probability, predicted wear depth and dynamic weights of the wear area value are calculated respectively through the adaptive model;

[0014] Step 6: Extract features of the preprocessed dynamic contact force, hard point impact probability, predicted wear depth and wear area values ​​respectively, and then input the extracted different features and corresponding dynamic weights into the multimodal data fusion neural network to output the multimodal fusion feature vector;

[0015] Step 7: By inputting the multimodal fusion feature vector into the operation status assessment model, the operation status assessment model is used to assess whether the pantograph has an operation fault.

[0016] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for time alignment of the collected data is:

[0017] A unified time scale is added to each sensor of the multimodal sensor system, and then the data transmission time difference of each sensor is calculated by the optical fiber transmission delay compensation algorithm. The compensation formula is as follows:

[0018]

[0019] Among them, t 补偿 The transmission time difference that needs to be compensated for the sensor's data transmission process; L is the length of the optical fiber; n 纤 is the core refractive index; c is the speed of light; t 处理 Processing delay time of the demodulator during the data transmission process of the sensor;

[0020] Calculate t for each sensor 补偿 ; Select a reference time point t 参考 , and then calculate the time delay Δt of each sensor. The calculation formula of time delay Δt is: Δt = t 参考 -t 补偿 ;

[0021] According to the time delay Δt of each sensor, the timestamp of the corresponding sensor is adjusted to make it consistent with the reference time point t参考 Alignment.

[0022] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for generating the dynamic contact force of the pantograph is:

[0023] The pre-processed vibration data, contact force data and contact point temperature data are input into the data coupling model for comprehensive calculation, and the dynamic contact force of the pantograph is output. The calculation formula of the data coupling model is:

[0024]

[0025] Among them, F d is the dynamic contact force of the pantograph; F s is the static contact force of the pantograph; m i is the vibration mass of the pantograph in different directions, i represents the sequence number of different directions, and the value is 1, 2 or 3, m 1 is the component representing the vibrating mass of the pantograph in the X-axis direction; m 2 is the component representing the vibrating mass of the pantograph in the Y-axis direction; m 3 is the component representing the vibrating mass of the pantograph in the Z-axis direction; a 1 is the vibration acceleration component of the pantograph in the X-axis direction; a 2 is the vibration acceleration component of the pantograph in the Y-axis direction; a 3 is the vibration acceleration component of the pantograph in the Z-axis direction; β is the thermal expansion coefficient; ρ is the air density; C d is the aerodynamic drag coefficient; A is the frontal area of ​​the vehicle body; T 0 is the reference temperature; T c is the contact point temperature of the pantograph; v is the vehicle speed.

[0026] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for determining whether to trigger the pantograph pressure control instruction is:

[0027] Set the contact force interval threshold;

[0028] The dynamic contact force within a unit time period is collected to form a dynamic contact force sequence, the contact force mean of multiple dynamic contact forces in the dynamic contact force sequence is calculated, and the contact force mean is compared with the contact force interval threshold. If the value of the contact force mean does not meet the contact force interval threshold, the pressure control instruction needs to be triggered.

[0029] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for calculating the hard point impact probability by analyzing the vibration data and the dynamic contact force is:

[0030] The vibration data and dynamic contact force are input into the hard point impact probability model. The hard point impact probability model can extract the kurtosis coefficient through the vibration data. The formula is: Where K is the kurtosis coefficient, N is the number of input samples; σ is the standard deviation of the pantograph acceleration, and μ is the mean of the pantograph vibration acceleration component;

[0031] The hard point impact probability model can extract the contact force fluctuation rate through the dynamic contact force sequence, and the formula is: Among them, δ F is the contact force fluctuation rate, F d,max and F d,min are the maximum and minimum values ​​of the dynamic contact force in the dynamic contact force sequence, F d,avg is the mean value of the dynamic contact force in the dynamic contact force sequence; then the hard point impact probability is calculated and output through the judgment formula, the formula is:

[0032]

[0033] The method for judging whether to trigger the train speed limit instruction according to the hard point impact probability is:

[0034] Set hard point impact assessment threshold;

[0035] When the hard point impact probability is greater than or equal to the hard point impact assessment threshold, the train speed limit instruction is triggered.

[0036] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for obtaining the predicted damage depth is:

[0037] The network structure of the wear depth prediction model includes input layer, LSTM layer and output layer, input layer;

[0038] Input the wear data, contact point temperature data and contact force data in a past unit time period into the input layer of the wear depth prediction model;

[0039] The analysis and calculation are performed through the LSTM layer, and the formula is:

[0040] Among them, d s is the predicted wear depth of the pantograph, d w is the wear depth of the pantograph, t 当前 is the current time point, t 单 is a unit time period;

[0041] The output layer outputs the predicted wear depth in future unit time periods.

[0042] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for calculating the dynamic weight through the adaptive model is:

[0043] The dynamic contact force, hard point impact probability, predicted wear depth and wear area values ​​in the historical time are integrated into a multimodal data set. Then the multimodal data set and the vehicle speed, wind speed and contact point temperature at the corresponding time point are input into the adaptive model to calculate the dynamic weight of each parameter in the multimodal data set. The formula is as follows:

[0044]

[0045] Among them, q z is the dynamic weight of different data in the multimodal data set, z represents the sequence number of different data in the multimodal data set, and the sequence number is 1, 2, 3 or 4, which corresponds to the sequence number of dynamic contact force, hard point impact probability, predicted wear depth and wear area value respectively; S z It represents the parameter sensitivity of different data in the multimodal data set; Ez represents the information entropy of different data in the multimodal data set.

[0046] In the preferred scheme of the pantograph operation status monitoring and analysis method, the extracted different features and the corresponding dynamic weights are input into the multimodal data fusion neural network, and the multimodal fusion feature vector Rr=[q 1 ×F d ,q 2 ×P h ,q 3 ×d s ,q 4 ×d c ].

[0047] In the preferred scheme of the pantograph operation status monitoring and analysis method, the method for predicting whether the pantograph has an operation failure risk in the future unit operation time is:

[0048] Several multimodal fusion feature vectors Rr in the historical records and the corresponding fault category labels Gl are used to construct several training samples, forming a training sample set [(Rr 1 , Gl 1 ), (Rr 2 , Gl 2 ), ..., (Rr n , Gl n )], where Rr n Represents the multimodal fusion feature vector of the nth training sample; Gl n Indicates the fault category label of the nth training sample; n is the sequence number of the training sample in the training sample set, and its value is a positive integer;

[0049] The support vector machine model is trained using the training sample set. After the model training is completed, the classification function f(x) of the new sample is calculated by inputting the multimodal fusion feature vector Rr of the new sample into the support vector machine model. The fault type matching of the new sample is achieved through the numerical value of the classification function f(x); the formula of the classification function f(x) is:

[0050]

[0051] Among them, ax j is the Lagrange multiplier; Rr j is the multimodal fusion feature vector of the jth sample, Gl j is the fault category label of the jth sample, K(Rr j ,Rr) is the kernel function; b is the bias term;

[0052] K(Rr j ,Rr) is calculated as:

[0053] K(Rr j ,Rr)=exp(-γ×||Rr j -Rr|| 2 );

[0054] Among them: γ is the kernel function parameter, which controls the width of the kernel function.

[0055] (III) Beneficial effects

[0056] The present invention provides a pantograph operating status monitoring and analysis method, which has the following beneficial effects:

[0057] (1) By synchronously collecting the pantograph's vibration data, contact force data, contact point temperature data, and wear-burn data, the pantograph's operating status can be fully reflected, improving the accuracy and reliability of monitoring. This multimodal sensor fusion technology can intuitively utilize the relationship between nodes in graph-structured data and can be extended to utilize intra-modal and inter-modal relationships in multimodal problems.

[0058] (2) Vibration data, contact force data and contact point temperature data are collected synchronously through a multimodal sensor system, and dynamic contact force is generated by dynamic calculation. The change of contact force can be monitored in real time, thereby improving the response speed and accuracy to abnormal contact force. According to the detection results, the passive warning strategy is changed to active regulation. By analyzing the vibration data and dynamic contact force, the probability of hard point impact is calculated, and the train speed limit instruction is triggered according to the probability value. Through the two intervention control instructions, the impact of hard point impact on train operation can be effectively reduced, and the problem of continuous deterioration of pantograph damage due to untimely adjustment can be avoided. It is more conducive to improving the safety of the train, and can reduce the train downtime caused by faults, thereby improving the overall efficiency of railway transportation.

[0059] (3) Through the multimodal data fusion neural network, the vibration data, contact force data, contact point temperature data and wear data are comprehensively analyzed, and the multimodal fusion feature vector is output. It is then input into the operation status assessment model for fault diagnosis, which can more comprehensively assess the operation status of the pantograph. Through intelligent monitoring and prediction technology, the frequency and cost of manual inspection are reduced, the accuracy of assessment is improved, and the maintenance costs and safety risks caused by the expansion of faults are avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of the steps of a pantograph operating status monitoring and analysis method of the present invention;

[0061] Figure 2 A schematic diagram of determining whether to trigger a pressure control instruction in a pantograph operation status monitoring and analysis method of the present invention;

[0062] Figure 3 A schematic diagram of whether a pantograph operation status monitoring and analysis method interruption triggers a train speed limit instruction. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] Example 1

[0065] See also Figure 1-3 The present invention provides a pantograph operation status monitoring and analysis method, comprising:

[0066] Step 1: Synchronously collect the pantograph's vibration data, contact force data, contact point temperature data, and wear data through a multimodal sensor system.

[0067] It should be noted that the multimodal sensor system includes at least a fiber grating sensor, an airbag pressure sensor, an infrared thermal imager and a laser profiler, which are used to collect vibration data, contact force data, contact point temperature data and wear data respectively.

[0068] Specifically, fiber Bragg grating sensors are embedded in the pantograph slide base, pull rod hinge points, frame support beams, etc., and the sampling rate can be selected to be 1kHz and the wavelength resolution can be 1pm. The wavelength offset is converted into vibration data through an optical signal demodulation module (such as SM130 demodulator), and the high-frequency noise is eliminated using a wavelet noise reduction algorithm (Daubechies 4 wavelet basis). The airbag pressure sensor is installed on the carbon slide support component of the pantograph to collect contact force data. The infrared thermal imager should be installed in a position where the contact area between the pantograph and the contact network can be clearly observed in order to accurately measure the contact point temperature. The best installation position needs to be determined based on the specific design and installation space of the pantograph, and then the contact point temperature data is collected. The laser profiler can use a line laser with a wavelength of 650nm to scan the slide surface at an incident angle of 45° through the scanning mode, and calculate the wear depth through the triangulation principle, such as Wherein, Δx is the displacement of the laser point; θ=45°, φ is the inclination angle of the slide, which can be provided by the inertial measurement unit.

[0069] Step 2: Time align and preprocess the collected data.

[0070] Specifically, the method for time alignment of the collected data is:

[0071] A unified time scale is added to each sensor of the multimodal sensor system, and then the data transmission time difference of each sensor is calculated by the optical fiber transmission delay compensation algorithm. The compensation formula is as follows:

[0072]

[0073] Among them, t 补偿 The transmission time difference that needs to be compensated for the sensor's data transmission process; L is the length of the optical fiber; n 纤 is the core refractive index; c is the speed of light; t 处理 Processing delay time of the demodulator during the data transmission process of the sensor;

[0074] By calculating the t of each sensor 补偿 ; Select a reference time point t 参考 , and then calculate the time delay Δt of each sensor. The calculation formula of time delay Δt is: Δt = t 参考-t 补偿 ; According to the time delay Δt of each sensor, adjust the timestamp of the corresponding sensor so that it is consistent with the reference time point t 参考 Alignment: if Δt is a positive value, the timestamp is adjusted forward by Δt; if Δt is a negative value, the timestamp is adjusted backward by Δt. By time aligning the collected data, the time uniformity of different types of data can be ensured, and the calculation results at different time points can be more accurate and the accuracy is guaranteed.

[0075] Specifically, preprocessing includes at least normalization preprocessing of data to eliminate the dimensions of different parameters to facilitate subsequent calculation and analysis, and may also include data preprocessing forms such as noise reduction.

[0076] Step 3: Generate the dynamic contact force of the pantograph through comprehensive calculation of the processed vibration data, contact force data and contact point temperature data, and determine whether to trigger the pantograph pressure control instruction based on the dynamic contact force; the pantograph pressure control instruction is to control and adjust the outlet pressure of the pneumatic system, and then adjust the pantograph-net contact force. When the dynamic contact force is too large, reduce the pressure, and when the dynamic contact force is too small, increase the pressure.

[0077] Step 4: Analyze the vibration data and dynamic contact force, and calculate the hard point impact probability; determine whether to trigger the train speed limit instruction according to the hard point impact probability;

[0078] Step 5: The wear data, contact point temperature data and contact force data are comprehensively analyzed to obtain the predicted damage depth; the wear area value of the pantograph friction area is collected through the quantum dot fluorescence monitoring system; the dynamic contact force, hard point impact probability, predicted wear depth and dynamic weights of the wear area value are calculated respectively through the adaptive model;

[0079] It should be noted that the method of collecting the wear area value of the pantograph friction area through the quantum dot fluorescence monitoring system is that in the pantograph friction area, quantum dots can be uniformly coated on the surface of the carbon slide in advance. These quantum dots will fall off due to wear during the friction process, thus forming a fluorescence signal distribution map. The fluorescence detector collects the fluorescence signal by irradiating the part coated with quantum dots with an excitation light source, and converts it into an electrical signal and transmits it to the signal processor; a three-dimensional measurement system (such as laser scanning or a digital matrix camera) is used to obtain high-precision three-dimensional point cloud data on the pantograph surface. These data can be combined with the distribution of quantum dot fluorescence signals, and the area of ​​the wear area can be calculated through digital modeling and image processing technology.

[0080] Step 6: Extract features of the preprocessed dynamic contact force, hard point impact probability, predicted wear depth and wear area values ​​respectively, and then input the extracted different features and corresponding dynamic weights into the multimodal data fusion neural network to output the multimodal fusion feature vector;

[0081] Step 7: By inputting the multimodal fusion feature vector into the operation status assessment model, the operation status assessment model is used to assess whether the pantograph has an operation fault.

[0082] The present invention synchronously collects vibration data, contact force data and contact point temperature data through a multimodal sensor system, generates dynamic contact force by combining dynamic calculation, can monitor the change of contact force in real time, thereby improving the response speed and accuracy to contact force anomalies, and according to the detection results, changes the passive warning strategy to active regulation; by analyzing the vibration data and dynamic contact force, the probability of hard point impact is calculated, and the train speed limit instruction is triggered according to the probability value. Through the two intervention control instructions, the influence of hard point impact on train operation can be effectively reduced, and the problem of continuous deterioration of pantograph damage due to untimely adjustment can be avoided, which is more conducive to improving the safety of the train, and can reduce the train downtime caused by faults, and improve the overall efficiency of railway transportation; through the multimodal data fusion neural network, the vibration data, contact force data, contact point temperature data and wear data are comprehensively analyzed, and the multimodal fusion feature vector is output, which is then input into the operation status evaluation model for fault diagnosis, which can more comprehensively evaluate the operation status of the pantograph, and reduce the frequency and cost of manual detection through intelligent monitoring and prediction technology, improve the evaluation accuracy, and avoid the maintenance costs and safety risks caused by the expansion of faults.

[0083] Example 2

[0084] In the preferred embodiment of the above pantograph operation status monitoring and analysis method, the method for generating the dynamic contact force of the pantograph is:

[0085] The pre-processed vibration data, contact force data and contact point temperature data are input into the data coupling model for comprehensive calculation, and the dynamic contact force of the pantograph is output. The calculation formula of the data coupling model is:

[0086]

[0087] Among them, F d is the dynamic contact force of the pantograph; F s is the static contact force of the pantograph; m i is the vibration mass of the pantograph in different directions, which can be measured and obtained through vibration modal analysis system or finite element simulation system. i represents the serial number of different directions, which can be 1, 2 or 3. m 1 is the component representing the vibrating mass of the pantograph in the X-axis direction; m 2 is the component representing the vibrating mass of the pantograph in the Y-axis direction; m 3 is the component representing the vibrating mass of the pantograph in the Z-axis direction; a 1is the vibration acceleration component of the pantograph in the X-axis direction; a 2 is the vibration acceleration component of the pantograph in the Y-axis direction; a 3 is the vibration acceleration component of the pantograph in the Z-axis direction; β is the thermal expansion coefficient, which can be obtained by checking industry data or calibrated according to experiments. Here, the value is β = 4.5 × 10 -6 / ℃; ρ is the air density, which can be obtained through the air density measurement system. The air density measurement system generally includes temperature sensors, humidity sensors, pressure sensors, CO2 content sensors, data acquisition systems, display instruments and other components. By accurately measuring the temperature, humidity, pressure, CO2 content and other parameters in the quality environment system, the air density value in the quality measurement environment is determined by the CIPM formula. Here, the value is ρ = 1.225kg / m3; C d is the aerodynamic drag coefficient, which can be determined by wind tunnel testing and is taken as C d =0.85; A is the windward area of ​​the vehicle body, which can be modeled based on train data and simulated in the train running environment to obtain the windward area of ​​the vehicle body through modeling software; T 0 is the reference temperature, which can be taken as normal temperature, such as 25°C; T c is the contact point temperature of the pantograph; v is the vehicle speed, which can be measured by a sensor.

[0088] In the formula, m i a i It is used to represent the vibration inertia force term, which can reflect the dynamic characteristics of the mechanical structure of the pantograph; β(-0) is used to represent the thermal expansion force term, which can quantify the influence of thermal expansion effect on contact force and characterize the temperature-mechanical coupling effect; It is used to represent aerodynamic terms and characterize speed-related air resistance corrections. For example, for physical field interaction examples: vehicle speed increases → aerodynamic resistance increases → contact force increases. At the same time, frictional heat causes the contact point temperature to rise → thermal expansion forces are further superimposed. The model characterizes the multi-field coupling of mechanical vibration, thermodynamics, and aerodynamic effects, providing core algorithm support for the dynamic contact force calculation of high-speed pantographs. Through the synergy of multiple data, it can improve the authenticity and accuracy of the dynamic contact force calculation.

[0089] Furthermore, the method for determining whether to trigger the pantograph pressure control instruction is as follows:

[0090] Set the contact force interval threshold;

[0091] It should be noted that the contact force interval threshold can use a machine learning model to train historical data, and dynamically adjust the contact force interval threshold according to changes in real-time data and environmental conditions, thereby improving the real-time performance of the solution and making the control more accurate. It can also be set according to industry standards and specifications.

[0092] The dynamic contact force in a unit time period is collected to form a dynamic contact force sequence {Fd 1 ,Fd 2 ,…,Fd q}, represents the dynamic contact force at different time points in a unit time period, calculates the contact force mean of multiple dynamic contact forces in the dynamic contact force sequence, compares the contact force mean with the contact force interval threshold, and if the contact force mean value does not meet the contact force interval threshold, it is necessary to trigger the pressure control instruction; by calculating the mean of the dynamic contact force sequence with a shorter unit time, the false triggering problem caused by occasional data anomalies can be avoided, thereby ensuring the rationality of monitoring.

[0093] In the above embodiment, by comprehensively considering multiple influencing factors such as static contact force, vibration, thermal expansion and aerodynamics, the calculation results are made more comprehensive and accurate, and the dynamic contact force is calculated to provide a basis for real-time monitoring and adjustment of the pantograph; the accuracy of the calculation can be further improved to avoid the comprehensive influence of multiple factors that may be ignored by the single factor calculation method, reduce calculation errors, and improve the reliability of pantograph operation status monitoring; by setting the contact force range threshold and comparing it with the dynamic contact force, it is possible to facilitate the regulation of the contact force, thereby effectively avoiding the continuous deterioration of the pantograph due to contact force problems, thereby improving the safety of train driving.

[0094] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for calculating the hard point impact probability by analyzing the vibration data and the dynamic contact force is:

[0095] The vibration data and dynamic contact force are input into the hard point impact probability model. The hard point impact probability model can extract the kurtosis coefficient through the vibration data. The formula is: Among them, K is the kurtosis coefficient, which is used to measure the sharpness of vibration data, N is the number of input samples; σ is the standard deviation of the pantograph acceleration, which reflects the discreteness of the vibration data, and μ is the mean value of the vibration acceleration component of the pantograph; the kurtosis coefficient is very sensitive to extreme values ​​in the data and can effectively capture the vibration anomalies caused by hard point impact. When the kurtosis coefficient is large, it means that the data distribution has more extreme values, that is, the probability of hard point impact is high. The traditional method may be affected by the installation position and accuracy of the contact force sensor and is only applicable to specific contact network structures. The kurtosis coefficient method is based on vibration data and has less dependence on the sensor installation position. It is applicable to various contact network structures and operating conditions and has higher adaptability.

[0096] The hard point impact probability model can extract the contact force fluctuation rate through the dynamic contact force sequence, and the formula is: Among them, δ F is the contact force fluctuation rate, F d,max and Fd,min are the maximum and minimum values ​​of the dynamic contact force in the dynamic contact force sequence, F d,avg is the mean value of the dynamic contact force in the dynamic contact force sequence; then the hard point impact probability is calculated and output through the judgment formula, the formula is:

[0097]

[0098] The method for judging whether to trigger the train speed limit instruction according to the hard point impact probability is:

[0099] Set hard point impact assessment threshold;

[0100] It should be noted that the hard point impact assessment threshold can use a machine learning model to train historical data, and dynamically adjust according to changes in real-time data and environmental conditions, thereby improving the real-time performance of the solution and making the control more accurate. It can also be set according to industry standards and specifications.

[0101] When the hard point impact probability is greater than or equal to the hard point impact assessment threshold, the train speed limit instruction is triggered to reduce the train speed and avoid excessive damage to the pantograph.

[0102] In the above embodiment, the hard point impact can be identified more accurately by combining the kurtosis coefficient and the dynamic contact force fluctuation rate. The kurtosis coefficient reflects the distribution pattern of the contact force sequence, while the dynamic contact force fluctuation rate reflects the degree of fluctuation of the contact force. The combination of the two can more comprehensively evaluate the possibility of hard point impact; according to the different ranges of the kurtosis coefficient and the dynamic contact force fluctuation rate, the calculation formula of the hard point impact probability is dynamically adjusted. It can effectively reduce false alarms and missed alarms. Using different calculation formulas under different conditions can more accurately judge the hard point impact and avoid misjudgment caused by a single indicator. This dynamic adjustment can adapt to different operating conditions and contact force changes, and improve the flexibility and accuracy of monitoring.

[0103] Example 3

[0104] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for obtaining the predicted damage depth is:

[0105] The network structure of the wear depth prediction model includes input layer, LSTM layer and output layer, input layer;

[0106] Input the wear data, contact point temperature data and contact force data in a past unit time period into the input layer of the wear depth prediction model;

[0107] The analysis and calculation are performed through the LSTM layer, and the formula is:

[0108] Among them, d sis the predicted wear depth of the pantograph, d w is the wear depth of the pantograph, t 当前 is the current time point, t 单 is a unit time period;

[0109] The output layer outputs the predicted wear depth in future unit time periods.

[0110] It should be noted that Indicates the wear depth value of the pantograph generated at different time points in the previous historical unit time period from the current time to the end; It represents the dynamic contact force generated at different time points in the previous historical unit time period from the current time to the end; It indicates the contact point temperature generated at different time points in the previous historical unit time period from the current time as the end; the unit time period can be selected as 60 seconds, and the output layer can output the predicted wear depth at multiple time points in the future unit time period.

[0111] In the above embodiment, the output layer can output the predicted wear depth at multiple time points in the future unit time period, which enables the model to provide more detailed prediction results, helps to more finely monitor and manage the operating status of the pantograph, and solves the problem that traditional wear depth prediction methods are difficult to capture long-term dependencies and complex nonlinear relationships, resulting in low prediction accuracy. The LSTM layer can effectively integrate multi-factor information such as wear depth, dynamic contact force, and contact point temperature to provide more comprehensive prediction results, effectively solve these problems, provide more accurate prediction results, help to achieve more refined real-time monitoring and management, and improve the reliability and safety of the system.

[0112] Example 4

[0113] In the preferred embodiment of the pantograph operation status monitoring and analysis method, the method for calculating the dynamic weight through the adaptive model is:

[0114] The dynamic contact force, hard point impact probability, predicted wear depth and wear area values ​​of continuous time periods in the historical time are integrated into a multimodal data set. Then the multimodal data set and the vehicle speed, wind speed and contact point temperature at the corresponding time point are input into the adaptive model to calculate the dynamic weight of each parameter in the multimodal data set. The formula is as follows:

[0115]

[0116] Among them, q z is the dynamic weight of different data in the multimodal data set, z represents the sequence number of different data in the multimodal data set, and the sequence number is 1, 2, 3 or 4, which corresponds to the sequence number of dynamic contact force, hard point impact probability, predicted wear depth and wear area value, respectively. For example, q1 The dynamic weight corresponding to the dynamic contact force; q 2 Dynamic weight corresponding to the probability of hard point impact; q 3 The dynamic weight corresponding to the predicted wear depth; q 4 Dynamic weight corresponding to the wear area; S z Indicates the parameter sensitivity of different data in a multimodal data set; for example, S 1 Parameter sensitivity corresponding to dynamic contact force; S 2 Parameter sensitivity corresponding to hard point impact probability; S 3 Corresponding to the parameter sensitivity of predicting wear depth; S 4 The parameter sensitivity corresponding to the wear area; Ez represents the information entropy of different data in the multimodal data set; 1 The information entropy corresponding to the dynamic contact force; E 2 The information entropy corresponding to the hard point impact probability; E 3 The information entropy corresponding to the predicted wear depth; E 4 Information entropy corresponding to the wear area.

[0117] It should be noted that parameter sensitivity indicates the degree of influence of different data in a multimodal data set on the operating state of the pantograph. It can be obtained through industry standards and can also be calculated through experimental data, such as designing a series of experiments to change the values ​​of various parameters (such as dynamic contact force, hard point impact probability, predicted wear depth and wear area value) while keeping other parameters unchanged. Under different parameter settings, data on the operating state of the pantograph is collected, and the influence of different parameter changes on the operating state of the pantograph is analyzed. The sensitivity of each parameter is calculated. The sensitivity can be calculated through regression analysis, sensitivity analysis and other methods. Information entropy represents the information uncertainty of different data in a multimodal data set. By statistically analyzing the distribution of each parameter in historical data, the information entropy of each parameter is calculated according to the entropy formula in information theory. The calculation formula is:

[0118]

[0119] Among them, P z It represents the probability of different parameters in the multimodal data set under different value intervals. u is the serial number of the value interval, which is a positive integer. It can be achieved by dividing some data in the historical data into several value intervals according to the needs, and numbering each interval, which is replaced by the subscript u. For example, P 1，u The probability corresponding to the dynamic contact force in the u-th value interval.

[0120] Furthermore, the extracted different features and the corresponding dynamic weights are input into the multimodal data fusion neural network, and the multimodal fusion feature vector Rr = [q 1 ×F d ,q2 ×P h ,q 3 ×d s ,q 4 ×d c ].

[0121] It should be noted that the features of different parameters can be extracted by extraction methods known in the art, and feature extraction can be performed by mathematical models such as Fourier transform.

[0122] Furthermore, the method for predicting whether there is a risk of operating failure of the pantograph in the future unit operating time is:

[0123] Several multimodal fusion feature vectors Rr in the historical records and the corresponding fault category labels Gl are used to construct several training samples, forming a training sample set [(Rr 1 , Gl 1 ), (Rr 2 , Gl 2 ), ..., (Rr n , Gl n )], where Rr n Represents the multimodal fusion feature vector of the nth training sample; Gl n Indicates the fault category label of the nth training sample; n is the sequence number of the training sample in the training sample set, and its value is a positive integer;

[0124] It should be noted that the fault category label Gl can be obtained by first determining the fault type and then calculating the multimodal fusion feature vector Rr corresponding to the fault type so that the two have a clear correspondence and become a set of training samples.

[0125] The support vector machine model is trained using the training sample set. After the model training is completed, the classification function f(x) of the new sample is calculated by inputting the multimodal fusion feature vector Rr of the new sample into the support vector machine model. The fault type matching of the new sample is achieved through the numerical value of the classification function f(x); the formula of the classification function f(x) is:

[0126]

[0127] Among them, ax j The Lagrange multiplier is a parameter obtained during the SVM training process, which indicates the influence of each training sample on the decision boundary; Rr j is the multimodal fusion feature vector of the jth sample, Gl j is the fault category label of the jth sample, K(Rr j ,Rr) is the kernel function; b is the bias term, which is a parameter obtained during the SVM training process and is used to adjust the position of the decision boundary.

[0128] K(Rr j ,Rr) is used to calculate the similarity of two samples in the feature space, and the calculation formula is:

[0129] K(Rr j ,Rr)=exp(-γ×||Rr j -Rr|| 2 );

[0130] Among them: γ is the kernel function parameter, which controls the width of the kernel function.

[0131] It should be noted that the support vector machine model distinguishes the samples and inputs the multimodal fusion feature vector of the new sample. The support vector machine model can compare it with the multimodal fusion feature vector of the sample during the training process, match the closest sample, find the corresponding fault category label, and then determine the operation fault.

[0132] In the above embodiment, the multimodal fusion feature vector is used to integrate multiple sensor data for fault classification, which improves the accuracy and robustness of classification, and can map nonlinearly separable data into a high-dimensional feature space for classification, thereby improving the classification ability of the model.

[0133] Example 5

[0134] The present invention also discloses a pantograph operation status monitoring and analysis system, which is used to implement the above-mentioned pantograph operation status monitoring and analysis method.

[0135] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware or in combination with computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0136] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0137] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A pantograph operating status monitoring and analysis method, characterized in that: include: Step 1: synchronously collect the pantograph vibration data, contact force data, contact point temperature data and wear data through a multi-modal sensor system; Step 2: Time alignment and data preprocessing of the collected data; Step 3: Perform comprehensive calculations on the processed vibration data, contact force data, and contact point temperature data to generate the dynamic contact force of the pantograph, and determine whether to trigger the pantograph pressure control instruction based on the dynamic contact force; Step 4: Analyze the vibration data and dynamic contact force, and calculate the hard point impact probability; determine whether to trigger the train speed limit instruction according to the hard point impact probability; Step 5: The wear data, contact point temperature data and contact force data are comprehensively analyzed to obtain the predicted damage depth; the wear area value of the pantograph friction area is collected through the quantum dot fluorescence monitoring system; the dynamic contact force, hard point impact probability, predicted wear depth and dynamic weights of the wear area value are calculated respectively through the adaptive model; Step 6: Extract features of the preprocessed dynamic contact force, hard point impact probability, predicted wear depth and wear area values ​​respectively, and then input the extracted different features and corresponding dynamic weights into the multimodal data fusion neural network to output the multimodal fusion feature vector; Step 7: By inputting the multimodal fusion feature vector into the operation status assessment model, the operation status assessment model is used to assess whether the pantograph has an operation fault.

2. A pantograph operating status monitoring and analysis method according to claim 1, characterized in that: The method for time alignment of the acquired data is: A unified time scale is added to each sensor of the multimodal sensor system, and then the data transmission time difference of each sensor is calculated by the optical fiber transmission delay compensation algorithm. The compensation formula is as follows: Among them, t 补偿 The transmission time difference that needs to be compensated for the sensor's data transmission process; L is the length of the optical fiber; n 纤 is the core refractive index; c is the speed of light; t 处理 Processing delay time of the demodulator during the data transmission process of the sensor; Calculate t for each sensor 补偿 ; Select a reference time point t 参考 , and then calculate the time delay Δt of each sensor. The calculation formula of time delay Δt is: Δt = t 参考 -t 补偿 ; According to the time delay Δt of each sensor, the timestamp of the corresponding sensor is adjusted to make it consistent with the reference time point t 参考 Alignment.

3. A pantograph operating status monitoring and analysis method according to claim 1, characterized in that: The method to generate the dynamic contact force of the pantograph is: The pre-processed vibration data, contact force data and contact point temperature data are input into the data coupling model for comprehensive calculation, and the dynamic contact force of the pantograph is output. The calculation formula of the data coupling model is: Among them, F d is the dynamic contact force of the pantograph; F s is the static contact force of the pantograph; m i is the vibration mass of the pantograph in different directions, i represents the serial number of different directions, and its value is 1, 2 or 3, m1 represents the component of the vibration mass of the pantograph in the X-axis direction; m2 represents the component of the vibration mass of the pantograph in the Y-axis direction; m3 represents the component of the vibration mass of the pantograph in the Z-axis direction; a1 represents the vibration acceleration component of the pantograph in the X-axis direction; a2 represents the vibration acceleration component of the pantograph in the Y-axis direction; a3 represents the vibration acceleration component of the pantograph in the Z-axis direction; β is the thermal expansion coefficient; ρ is the air density; C d is the aerodynamic drag coefficient; A is the frontal area of ​​the vehicle body; T0 is the reference temperature; T c is the contact point temperature of the pantograph; v is the vehicle speed.

4. A pantograph operating status monitoring and analysis method according to claim 3, characterized in that: The method for determining whether to trigger the pantograph pressure control instruction is: Set the contact force interval threshold; The dynamic contact force within a unit time period is collected to form a dynamic contact force sequence, the contact force mean of multiple dynamic contact forces in the dynamic contact force sequence is calculated, and the contact force mean is compared with the contact force interval threshold. If the value of the contact force mean does not meet the contact force interval threshold, the pressure control instruction needs to be triggered.

5. The pantograph operation status monitoring and analysis method according to claim 1, characterized in that: By analyzing the vibration data and dynamic contact force, the method for calculating the probability of hard point impact is: The vibration data and dynamic contact force are input into the hard point impact probability model. The hard point impact probability model can extract the kurtosis coefficient through the vibration data. The formula is: Where K is the kurtosis coefficient, N is the number of input samples; σ is the standard deviation of the pantograph acceleration, and μ is the mean of the pantograph vibration acceleration component; The hard point impact probability model can extract the contact force fluctuation rate through the dynamic contact force sequence, and the formula is: Among them, δ F is the contact force fluctuation rate, F d,max and F d,min are the maximum and minimum values ​​of the dynamic contact force in the dynamic contact force sequence, F d,avg is the mean value of the dynamic contact force in the dynamic contact force sequence; then the hard point impact probability is calculated and output through the judgment formula, the formula is: The method for judging whether to trigger the train speed limit instruction according to the hard point impact probability is: Set hard point impact assessment threshold; When the hard point impact probability is greater than or equal to the hard point impact assessment threshold, the train speed limit instruction is triggered.

6. A pantograph operating status monitoring and analysis method according to claim 3, characterized in that: The method for obtaining the predicted damage depth is: The network structure of the wear depth prediction model includes input layer, LSTM layer and output layer, input layer; Input the wear data, contact point temperature data and contact force data in a past unit time period into the input layer of the wear depth prediction model; The analysis and calculation are performed through the LSTM layer, and the formula is: Among them, d s is the predicted wear depth of the pantograph, d w is the wear depth of the pantograph, t 当前 is the current time point, t 单 is a unit time period; The output layer outputs the predicted wear depth in future unit time periods.

7. A pantograph operating status monitoring and analysis method according to claim 6, characterized in that: The method for calculating dynamic weights through the adaptive model is: The dynamic contact force, hard point impact probability, predicted wear depth and wear area values ​​in the historical time are integrated into a multimodal data set. Then the multimodal data set and the vehicle speed, wind speed and contact point temperature at the corresponding time point are input into the adaptive model to calculate the dynamic weight of each parameter in the multimodal data set. The formula is as follows: Among them, q z is the dynamic weight of different data in the multimodal data set, z represents the sequence number of different data in the multimodal data set, and the sequence number is 1, 2, 3 or 4, which corresponds to the sequence number of dynamic contact force, hard point impact probability, predicted wear depth and wear area value respectively; S z It represents the parameter sensitivity of different data in the multimodal data set; Ez represents the information entropy of different data in the multimodal data set.

8. A pantograph operating status monitoring and analysis method according to claim 7, characterized in that: The extracted different features and the corresponding dynamic weights are input into the multimodal data fusion neural network, and the multimodal fusion feature vector Rr = [q1×F d ,q2×P h ,q3×d s ,q4×d c ].

9. A pantograph operating status monitoring and analysis method according to claim 8, characterized in that: The method for predicting whether there is a risk of operating failure of the pantograph in the future unit operating time is: Several multimodal fusion feature vectors Rr and corresponding fault category labels Gl in historical records are used to construct several training samples to form a training sample set [(Rr1, Gl1), (Rr2, Gl2),..., (Rr n , Gl n )], where Rr n Represents the multimodal fusion feature vector of the nth training sample; Gl n Indicates the fault category label of the nth training sample; n is the sequence number of the training sample in the training sample set, and its value is a positive integer; The support vector machine model is trained using the training sample set. After the model training is completed, the classification function f(x) of the new sample is calculated by inputting the multimodal fusion feature vector Rr of the new sample into the support vector machine model. The fault type matching of the new sample is achieved through the numerical value of the classification function f(x); the formula of the classification function f(x) is: Among them, ax j is the Lagrange multiplier; Rr j is the multimodal fusion feature vector of the jth sample, Gl j Rr j The corresponding fault category label, K(Rr j ,Rr) is the kernel function; b is the bias term; K(Rr j ,Rr) is calculated as: K(Rr j ,Rr)=exp(-γ×||Rr j -Rr|| 2 ); Among them: γ is the kernel function parameter, which controls the width of the kernel function.

10. A pantograph operation status monitoring and analysis system, characterized in that: A pantograph operating status monitoring and analysis method for implementing any one of claims 1-9 above.

Images