A method for online monitoring of train wheel polygonal faults

Abstract

The present invention belongs to the technical field of rail train fault detection, and discloses a method for online monitoring of train wheel polygon faults. The wheel polygon fault monitoring method of the present invention introduces a method of using the average value of polygon indicators within a certain period of time as a polygon evaluation indicator and a method of selecting the polygon order indicator that appears the most times within a certain period of time as a polygon order evaluation indicator, thereby ensuring that the evaluation indicator is stable and reliable, and effectively avoiding misdiagnosis and missed diagnosis; the vehicle speed is introduced as a condition for fault judgment and alarm status output, and a method of time dimension judgment and fault status value counting judgment is introduced, thereby effectively avoiding the influence of the track state on the wheel polygon fault status judgment, and reducing the interference of a large number of fault alarm information prompts on drivers and passengers. The wheel polygon fault detection method of the present invention has accurate and stable monitoring results, has low requirements on system performance, and meets the requirements of a low-cost embedded device real-time online monitoring system.

Description

Technical Field

[0001] The invention belongs to the technical field of rail train fault detection and relates to a method for online monitoring of train wheel polygonal faults. Background Art

[0002] During train operation, there are many technical indicators related to economy, comfort and safety that need to be focused on, and wheel polygon fault indicators are one of them. Wheel polygon, also known as wheel corrugation or wheel periodic non-circularity, is common in subway vehicles, ordinary railways and high-speed train operations. Wheel polygon will aggravate vehicle vibration and generate noise, affecting ride comfort; and vibration will also affect the service life of the wheel-rail system, affecting the stability, economy and safety of train operation.

[0003] The wheel polygon state detection methods currently in use are mainly divided into three categories: ground detection, trackside detection and real-time online monitoring. At present, most of the wheel polygon state detection in my country's train maintenance sections is mainly based on ground detection. Ground detection uses specific equipment for inspection. The advantage is high detection accuracy, but the wheelset needs to be disassembled for inspection. Generally, it is only carried out during level 4 or level 5 repairs, and the detection frequency is too low. Trackside detection generally uses image detection technology, using a high-definition camera array installed on the trackside to fully collect the tread image of the vehicle wheelset, and automatically analyze and identify fault information. The advantage is high detection accuracy, but the disadvantage is that the trackside detection equipment is fixed in position, and the train needs to pass by this equipment before detection can be carried out, and the detection result is affected by the train passing speed. Real-time online monitoring generally refers to the use of vibration acceleration sensors installed on the axle boxes of operating vehicles for detection, and the polygon evaluation index is obtained through the detection algorithm for real-time detection.

[0004] The existing real-time online monitoring method uses short-time indicators for alarm judgment, which can better identify the degree of wheel polygonal faults. However, during the actual operation of the train, due to differences in the running track conditions, when the train runs on a track section with poor conditions, the polygonal evaluation index value will be affected by the impact signal and will be high, which can easily lead to misjudgment. In addition, when there is a polygonal fault in the wheel, the use of short-time indicators for alarm judgment will continue to report polygonal faults, resulting in a large number of alarm information prompts. Summary of the invention

[0005] The present invention provides a specific solution to this difficult problem. The present invention adopts a method of analyzing and processing the vibration acceleration signal collected by the sensor, which can not only monitor the wheel polygon defect in real time online, but also effectively avoid the false alarm and other situations that are easy to be generated by using short-term indicator alarm judgment.

[0006] To achieve the above purpose, the technical solution adopted by the present invention to solve the problem is:

[0007] Obtain the real-time speed and wheel diameter information of the train; calculate the real-time rotation frequency of the wheel according to the real-time speed and wheel diameter, and then calculate the wheel polygon order frequency value sequence;

[0008] Acquire the collected vertical vibration acceleration signal of the train wheel; perform low-pass filtering on the vibration acceleration signal, intercept the middle section, and perform frequency domain transformation at the same time;

[0009] According to the vibration acceleration signal obtained after intercepting the middle section, the effective value in the time domain is calculated first, and then the vibration effective value index is calculated;

[0010] According to the frequency domain data obtained by frequency domain transformation, a certain number of peaks and corresponding frequencies in the frequency domain data are first extracted to form a peak sequence, and the frequency domain amplitude corresponding to the wheel polygon order frequency value sequence is extracted from the peak sequence using a difference threshold algorithm to form a wheel polygon amplitude-frequency feature sequence;

[0011] According to the wheel polygon amplitude-frequency characteristic sequence, a group of data with the largest amplitude is taken as the characterization quantity, and the polygon amplitude index and polygon order index are calculated;

[0012] According to the vibration effective value index and the polygon amplitude index, the polygon index is calculated;

[0013] According to the polygon index and the polygon order index, a polygon evaluation index and a polygon order evaluation index are calculated;

[0014] When the vehicle speed is greater than the preset value, the polygon evaluation index is compared with the alarm threshold, and combined with the polygon order index to determine whether the fault state counting conditions are met and count; when the vehicle speed is less than the preset value, it is determined whether the total number of fault states exceeds the alarm judgment number, the alarm state is output, and the relevant alarm parameters are set to 0.

[0015] A method for online monitoring of train wheel polygonal faults, the specific implementation steps are as follows:

[0016] Step 1: Obtain the real-time speed TSpd and wheel diameter TDia of the train;

[0017] Step 2: Calculate the real-time wheel rotation frequency St = TSpd*1000 / (3.6*TDia*π) based on the real-time vehicle speed TSpd and the wheel diameter TDia, and then calculate the wheel polygon order frequency value sequence On(St);

[0018] Step 3: Obtain the collected vertical vibration acceleration signal of the train wheel;

[0019] Step 4: After low-pass filtering the vibration acceleration signal obtained in step 3, the middle section is intercepted and FFTW frequency domain transformation is performed at the same time;

[0020] Step 5: According to the vibration acceleration signal obtained after intercepting the middle section in step 4, first calculate the time domain effective value RMS, and then calculate the vibration effective value index RmsdB=20log(RMS);

[0021] Step 6: Based on the frequency domain data obtained by the FFTW frequency domain transformation in step 4, first extract a certain number of peaks and corresponding frequencies in the frequency domain data to form a P(x, y) sequence, then sort the P(x, y) sequence from large to small according to the frequency domain amplitude to form a peak sequence Py(x, y), and use the difference threshold algorithm to extract the frequency domain amplitude corresponding to the wheel polygon order frequency value sequence from the peak sequence to form the wheel polygon amplitude-frequency feature sequence OPn(x, y);

[0022] Step 7: Based on the wheel polygon amplitude-frequency characteristic sequence OPn(x,y) obtained in step 6, take the set of data OPmax(x,y) with the largest frequency domain amplitude as the characterization quantity, and calculate the polygon amplitude index AdB=20log(OPmax(y)) and the polygon order index Polyn=OPmax(x) / St;

[0023] Step 8: According to step 5, the vibration effective value index RmsdB is obtained, and the polygon amplitude index AdB value and the weighting coefficient a are obtained in step 7, and the polygon index KndB=a*AdB+(1-a)*RmsdB is calculated;

[0024] Step 9: Average the polygon index KndB from 1 second to N seconds as the polygon evaluation index at the Nth second The KdB value in the initial N-1 seconds is 0;

[0025] Step 10: Select the polygon order index Polyn that appears most times between 1 second and N seconds as the final polygon order evaluation index Poly;

[0026] Step 11: Output the polygon evaluation index KdB and the polygon order evaluation index Poly;

[0027] Step 12: When the vehicle speed is greater than the preset value, the polygon evaluation index KdB is compared with the alarm threshold, and combined with the polygon order evaluation index Poly, it is determined whether the fault state counting condition is met, and the fault state times of the prediction state times ALTn1 or the warning state times ALTn2 or the alarm state times ALTn3 are counted; when the vehicle speed is less than the preset value, it is determined whether the sum of the fault state times exceeds the alarm judgment times, that is, whether ALTn1+ALTn2+ALTn3 is greater than or equal to ALTN, if it exceeds, the alarm state ALTtr is the state value ALTst corresponding to the maximum value among ALTn1, ALTn2, and ALTn3, otherwise the alarm state ALTtr=0;

[0028] Step 13: Output the alarm status ALTtr; and set the values ​​of ALTn1, ALTn2, ALTn3, and ALTtr to 0.

[0029] As a preferred method, in step 2, the wheel polygon order frequency value sequence On(St) is the product of the wheel rotation frequency St and the polygon order On sequence, and the order On of the wheel polygon fault is generally between 15-40;

[0030] As a preferred manner, in step 4, intercepting the middle section is intercepting the vibration acceleration signal between 10% and 90% of the vibration acceleration signal.

[0031] As a preferred method, in step 6, the method for extracting the peak sequence Py(x,y) includes: using the least squares fitting algorithm to extract a certain number of peaks and corresponding frequencies in the frequency domain data to form a P(x,y) sequence, and then sorting the P(x,y) sequence from large to small according to the frequency domain amplitude to form a peak sequence Py(x,y), which is a set of coordinate values, the horizontal axis x represents the frequency, and the vertical axis y represents the corresponding frequency domain amplitude. Further, in the P(x,y) sequence, 40-60 peaks and corresponding frequencies are extracted.

[0032] As a preferred method, in step 6, the frequency domain amplitude corresponding to the wheel polygon order frequency value sequence is extracted from the peak sequence using a difference threshold algorithm to form a wheel polygon amplitude-frequency characteristic sequence OPn(x,y), including: setting a coupling coefficient, coupling the wheel polygon order frequency value sequence On(St) with the frequency values ​​in the peak sequence Py(x,y) one by one, when |On(St)-Py(x)|≤coupling coefficient, it indicates that the frequency value is a fault characteristic frequency, extracting the frequency value and the corresponding frequency domain amplitude output to form a wheel polygon amplitude-frequency characteristic sequence OPn(x,y), the sequence OPn(x,y) is a set of coordinate values, the horizontal axis x represents the wheel polygon order frequency value, and the vertical axis y represents the corresponding frequency domain amplitude.

[0033] As a preferred method, in step 8, the weighting coefficient a ranges from 0.2 to 0.8, and is generally taken as a=0.5.

[0034] As a preferred method, in step 9, due to the dynamic change characteristics of the short-term force fluctuations and vibration signals of the wheels during the actual operation of the train, the calculated polygon index will produce numerical fluctuations and cause misjudgment. In order to calculate the polygon evaluation index more stably and reliably, the present invention introduces the average value of the polygon index within a certain period of time as the polygon evaluation index. Preferably, the average polygon index within 60 seconds is calculated as the polygon evaluation index.

[0035] As a preferred method, in step 10, due to the dynamic change characteristics of the short-term force fluctuations and vibration signals of the wheels during the actual operation of the train, the calculated polygon order index will produce numerical fluctuations and cause missed judgments. In order to calculate the polygon order evaluation index more stably and reliably, the present invention introduces a method of selecting the polygon order index that appears the most times within a certain period of time as the polygon order evaluation index. Preferably, the polygon order index that appears the most times within 60 seconds is selected as the polygon order evaluation index.

[0036] As a preferred method, in step 12, the method for determining whether to alarm is: when the vehicle speed is greater than a preset value, the polygon evaluation index KdB is compared with the alarm threshold. When KdB < the lowest value of the alarm threshold, the operation is normal, and the state value ALTst = 0; whenever the lowest value of the alarm threshold ≤ KdB < the middle value of the alarm threshold for n consecutive seconds, and the order Poly is the same, the state value ALTst = 1, and the number of pre-judgment states ALTn1 is increased by 1; whenever the middle value of the alarm threshold ≤ KdB < the highest value of the alarm threshold for n consecutive seconds, and the order Poly is the same, the state value ALTst = 2, and the number of warning states ALTn2 is increased by 1; whenever KdB ≥ the highest alarm threshold for n consecutive seconds, and the order Poly is the same, the state value ALTst = 3, and the number of alarm states ALTn3 is increased by 1. When the vehicle speed is less than the preset value, determine whether the total number of fault states exceeds the alarm judgment number, that is, whether ALTn1+ALTn2+ALTn3 is greater than or equal to ALTN. If it exceeds, the alarm state ALTtr is the state value ALTst corresponding to the maximum value of ALTn1, ALTn2, and ALTn3, and an alarm prompt is given; otherwise, the alarm state ALTtr=0, and no alarm prompt is given. If short-term indicators are used for alarm judgment, during the actual operation of the train, due to differences in the running track state, when the train runs on a track section with poor state, the evaluation index will be high, resulting in false alarms; and when there is a polygonal fault on the wheel, a polygonal fault will continue to be reported, resulting in a large number of alarm information prompts. Therefore, the present invention introduces vehicle speed as a condition for fault judgment and alarm state output, and introduces a method for time dimension judgment and fault state value counting judgment. That is, when the vehicle speed is greater than the preset value, it is determined whether the polygon evaluation index exceeds the alarm threshold continuously within a certain period of time, and whether the order evaluation index is the same, as a condition for counting the fault state; when the vehicle speed is less than the preset value, if the total number of fault states exceeds the threshold, it is diagnosed as a wheel polygon fault, and an alarm is given; further, the preset values ​​of ALTn1, ALTn2, and ALTn3 are 0; ALTst=1 is prediction, ALTst=2 is warning, and ALTst=3 is alarm; further, the value range of n is 1-60 seconds; further, the value range of the alarm judgment times ALTN is 50-300.

[0037] Beneficial effects of the present invention: The wheel polygon fault monitoring method of the present invention introduces the average value of polygon indicators within a certain period of time as the polygon evaluation indicator and the method of selecting the polygon order indicator that appears the most times within a certain period of time as the polygon order evaluation indicator, which ensures that the evaluation indicator is stable and reliable, and effectively avoids misdiagnosis and missed diagnosis; introduces vehicle speed as a condition for fault judgment and alarm status output, introduces time dimension judgment and fault status value counting judgment methods, effectively avoids the influence of track status on wheel polygon fault status judgment, and reduces the interference of a large number of fault alarm information prompts on drivers and passengers. The monitoring results of the wheel polygon fault detection method of the present invention are accurate and stable, with low requirements on system performance, meeting the requirements of low-cost embedded device real-time online monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the hardware structure of a system for implementing the method of the present invention.

[0039] Figure 2 It is a flow chart of the monitoring method of the present invention.

[0040] Figure 3 It is the time domain waveform of the measuring point.

[0041] Figure 4 It is the frequency domain waveform of the measurement point.

[0042] Figure 5 It is the trend chart of polygon evaluation index of the original method.

[0043] Figure 6 It is the trend chart of polygon evaluation index of this method.

[0044] Figure 7 It is the trend chart of polygonal order evaluation index of the original method.

[0045] Figure 8 It is the trend chart of polygonal order evaluation index of this method.

[0046] Fig. 9 It is the trend of polygon evaluation index and polygon order evaluation index of a certain wheel throughout the day.

[0047] Fig.10 It is the trend of polygonal fault status value of a certain wheel throughout the day.

[0048] Fig.11 It is the trend of polygonal fault alarm status value of a certain wheel throughout the day. DETAILED DESCRIPTION

[0049] The specific implementation of the present invention is described in detail below in conjunction with the technical scheme and the accompanying drawings.

[0050] Example 1

[0051] like Figure 1 As shown in the schematic diagram of the system hardware structure, the vibration sensor is arranged above the train axle box bearing, and a vibration acceleration sensor is arranged on each of the 8 wheels of each carriage. The data acquisition module collects the vertical vibration acceleration signal of the wheel obtained by the sensor and sends it to the diagnosis and analysis algorithm processing module; the communication module obtains the real-time speed and wheel diameter information of the train sent by the on-board central control unit through Ethernet, and sends it to the diagnosis and analysis algorithm processing module for calculating the real-time rotation frequency of the wheel; the wheel polygon evaluation index, wheel polygon order evaluation index and alarm information calculated by the diagnosis and analysis algorithm processing module are sent from the communication module to the on-board central control unit in real time through Ethernet. This method is implemented in the diagnosis and analysis algorithm processing module.

[0052] like Figure 2 The monitoring method flow chart shows that the specific steps are as follows:

[0053] Step 1: Obtain the real-time speed TSpd and wheel diameter TDia of the train;

[0054] Step 2: Calculate the real-time wheel rotation frequency St = TSpd*1000 / (3.6*TDia*π) based on the real-time vehicle speed TSpd and the wheel diameter TDia, and then calculate the wheel polygon order frequency value sequence On(St); the wheel polygon order frequency value sequence On(St) is the product of the wheel rotation frequency St and the polygon order On sequence. The order On of the wheel polygon fault is generally between 15-40;

[0055] Step 3: Obtain the collected vertical vibration acceleration signal of the train wheel;

[0056] Step 4: Perform 2500Hz low-pass filtering on the vibration acceleration signal. The filter selects the 6th-order Butterworth filter as the wheel polygon evaluation data. The filter transfer function is as follows:

[0057]

[0058] Perform Fourier transform on the filtered signal to obtain frequency domain data. The Fourier transform formula uses the FFTW formula for calculation. The calculation formula is as follows:

[0059] in,

[0060] According to the properties of the correlation function and the use of recursive divide and conquer, all coefficients, that is, the frequency domain amplitudes, can be quickly calculated.

[0061] Step 5: Calculate the effective value RMS of the vibration time domain of the filtered signal. Since the edge jitter of the filtered signal will cause calculation errors, part of the signal is intercepted to calculate the effective value in the time domain. When intercepting the signal, the value is generally between 10% and 90% of the data segment, and then calculate the effective value index of vibration RmsdB = 20log (RMS);

[0062] Step 6: Extract a certain number of peaks and corresponding frequencies from the frequency domain data to form a P(x,y) sequence. The extraction algorithm is the least squares fitting algorithm. Generally, 40-60 peaks and corresponding frequencies are extracted, and then the P(x,y) sequence is sorted from large to small according to the frequency domain amplitude to form a peak sequence Py(x,y). This sequence is a set of coordinate values, with the horizontal axis x representing the frequency and the vertical axis y representing the corresponding frequency domain amplitude. In this step, the peak sequence of multiple points is extracted first to ensure that the extracted peak sequence is a possible fault location, and then the least squares fitting algorithm is used to extract data, which is stable and reliable, time-saving and fast;

[0063] The difference threshold algorithm is used for fault feature coupling, and the wheel polygon order frequency value sequence On(St) is coupled with the frequency value in the peak sequence Px(x,y), with a coupling coefficient of 2.1. The specific calculation method is to couple the wheel polygon order frequency value sequence On(St) with the frequency value in the peak sequence Py(x,y) one by one. When |On(St)-Py(x)|≤2.1, it means that the frequency value is the fault characteristic frequency. The frequency value and the corresponding frequency domain amplitude output are extracted to form the wheel polygon amplitude-frequency feature sequence OPn(x,y). The sequence OPn(x,y) is a set of coordinate values, the horizontal axis x represents the wheel polygon order frequency value, and the vertical axis y represents the corresponding frequency domain amplitude. The fault feature coupling in this step adopts the difference threshold algorithm, which can eliminate the frequency error caused by floating-point calculation or speed fluctuation, and avoid feature extraction failure;

[0064] Step 7: Take the set of data OPmax(x,y) with the largest frequency domain amplitude in the wheel polygon amplitude-frequency characteristic sequence OPn(x,y) as the characterization quantity. The characterization quantity is a set of coordinates. OPmax(x) represents the wheel polygon order frequency value, and OPmax(y) represents the corresponding frequency domain amplitude. Calculate the polygon amplitude index AdB=20log(OPmax(y)) and the polygon order index Polyn=OPmax(x) / St.

[0065] Step 8: Calculate the polygon index KndB = a*AdB+(1-a)*RmsdB+Kfac, the weighted coefficient a ranges from 0.2 to 0.8, the preferred value is a=0.5, Kfac is the correction coefficient, the range is 0-50, and it is used to adapt to the alarm value of various types of vehicles; introduce the weight algorithm into the polygon index, and take the effective value of vibration as a component of the index parameter to ensure more accurate monitoring results;

[0066] Step 9: Average the polygon index KndB from 1 second to N seconds as the polygon evaluation index at the Nth second The KdB value within the initial N-1 seconds is 0; the value range of N is 20-100 seconds, and the preferred value is N=60; the average value of the polygon index within a certain period of time is introduced as the polygon evaluation index, which can effectively avoid the misjudgment caused by the fluctuation of the polygon index value due to the short-term force fluctuation of the wheel and the dynamic change characteristics of the vibration signal during the actual operation of the train, and ensure that the polygon evaluation index is more stable and reliable;

[0067] Step 10: Select the order of the polygon order Polyn that appears most often within a time period of 1 second to N seconds as the final polygon order evaluation index Poly; the value range of N is 20-100 seconds, and the optimal value is N=60; the method of selecting the polygon order index that appears most often within a certain period of time as the polygon order evaluation index is introduced, which can effectively avoid missed judgment caused by polygon order index fluctuations caused by short-term wheel force fluctuations and dynamic change characteristics of vibration signals during actual train operation, and ensure that the polygon order evaluation index is more stable and reliable;

[0068] Step 11: Output the polygon evaluation index KdB and the polygon order evaluation index Poly;

[0069] Step 12: The preset values ​​of ALTn1, ALTn2, and ALTn3 are 0; when the vehicle speed TSpd≥100km / h, the polygon evaluation index KdB is compared with the alarm threshold. When KdB<10(dB), the operation is normal, and the state value ALTst=0; whenever the polygon evaluation index KdB is 10(dB)≤KdB<18(dB) for n consecutive seconds, and the order Poly is the same, the state value ALTst=1, and the number of predicted states ALTn1 is increased by 1; whenever the polygon evaluation index KdB is 18(dB)≤KdB<23(dB) for n consecutive seconds, and the order Poly is the same, the state value ALTst=2, and the number of warning states ALTn 2 plus 1; whenever the polygon evaluation index KdB is KdB ≥ 23 (dB) for n consecutive seconds and the order Poly is the same, the state value ALTst = 3, and the alarm state number ALTn3 is increased by 1; when the vehicle speed TSpd ≤ 100km / h, the counting of ALTn1, ALTn2, and ALTn3 is ended; it is determined whether the total number of fault states exceeds the alarm judgment number, that is, whether ALTn1+ALTn2+ALTn3 is greater than or equal to ALTN, and the alarm state ALT tr is the state value ALTst corresponding to the maximum value among ALTn1, ALTn2, and ALTn3; otherwise, the alarm state ALTtr=0; the vehicle speed is introduced as a condition for fault judgment and alarm state output, and the method of time dimension judgment and fault state value counting judgment is introduced, that is, when the vehicle speed is greater than the preset value, it is judged whether the polygon evaluation index exceeds the threshold continuously within a certain period of time, and whether the order evaluation index is the same as the fault state counting condition; when the vehicle speed is less than the preset value, if the total number of fault states exceeds the threshold, it is diagnosed as a wheel polygon fault, and an alarm prompt is issued; it avoids the use of short-term indicators for alarm judgment, and when the train runs on a track section with poor condition, the evaluation index is too high to cause a false alarm, and avoids the situation where when there is a polygon fault in the wheel, the polygon fault is continuously reported, resulting in a large number of alarm information prompts; the value range of n is 1-60, and the preferred value is n=20; the value range of ALTN is 50-300, and the preferred value is ALTN=100.

[0070] Step 13: Output the alarm status ALTtr; and set the values ​​of ALTn1, ALTn2, ALTn3, and ALTtr to 0.

[0071] Example 2

[0072] The method of the present invention is applied to a domestic intelligent high-speed rail dynamic safety monitoring system to perform polygonal fault diagnosis, as follows:

[0073] Step 1: Get the real-time speed of the train TSpd = 344.5Km / h and the wheel diameter TDia = 920mm;

[0074] Step 2: Calculate the current wheel rotation frequency St = TSpd*1000 / (3.6*TDia*π) = 33.12 Hz according to the real-time vehicle speed and wheel diameter, and calculate the corresponding wheel polygon order frequency value On(St) = 496.8 Hz, 529.92 Hz...662.4 Hz...1324.8 Hz sequence;

[0075] Step 3: Obtain the collected vertical vibration acceleration signal of the train wheel;

[0076] Step 4: Figure 3 to Figure 4 , perform 2500Hz low-pass filtering on the vibration acceleration signal, and select the 6th-order Butterworth filter;

[0077] Perform Fourier transform on the filtered signal to obtain frequency domain data, and use FFTW formula to calculate the Fourier transform formula;

[0078] Step 5: Calculate the effective value RMS of the filtered signal in the time domain. Since the edge jitter of the filtered signal will cause calculation errors, part of the signal is intercepted to calculate the effective value of the vibration in the time domain. When intercepting the signal, the value is generally between 10% and 90% of the data segment. Then calculate the effective value index of vibration RmsdB = 20log (RMS)

[0079] =20log(8.9805g)=19.066dB;

[0080] Step 6: Extract a certain number of peaks and corresponding frequencies from the frequency domain data to form a P(x,y) sequence. The extraction algorithm is the least squares fitting algorithm. Generally, 40-60 peaks and corresponding frequencies are extracted, and then the P(x,y) sequence is sorted from large to small according to the frequency domain amplitude to form a peak sequence Py(x,y). Table 1 shows the peak sequences of the first 15 sorted peaks;

[0081] Table 1 Peak sequence ranking

[0082] Serial number Frequency (Hz) Amplitude(g) 1 661.932 10.3148 2 694.946 1.8826 3 1323.855 0.9097 4 628.838 0.6765 5 52.013 0.5837 6 641.694 0.5707 7 658.646 0.5704 8 645.331 0.5659 9 646.87 0.5536 10 665.046 0.4974 11 640.042 0.4966 12 669.579 0.4741 13 992.817 0.4549 14 675.111 0.4404 15 618.054 0.4013

[0083] The difference threshold algorithm is used for fault feature coupling, and the wheel polygon order frequency value sequence On(St) is coupled with the frequency value in the peak sequence Px(x,y), with a coupling coefficient of 2.1. The specific calculation method is to couple the wheel polygon order frequency value sequence On(St) with the frequency value in the peak sequence Py(x,y) one by one. When |On(St)-Py(x)|≤2.1, it means that the frequency value is the fault characteristic frequency. The frequency value and the corresponding frequency domain amplitude output are extracted to form the wheel polygon amplitude-frequency feature sequence OPn(x,y). Table 2 shows some of the extraction results;

[0084] Table 2 Order index extraction results

[0085]

[0086]

[0087] Step 7: Take the data set OPmax(x,y) with the largest frequency domain amplitude in the wheel polygon amplitude-frequency characteristic sequence OPn(x,y) as the characterization quantity, calculate the polygon amplitude index AdB=20log(OPmax(y))=20log(10.3148g)=20.269dB and the polygon order index Polyn=OPmax(x) / St=661.932 / 33.12=19.986, rounded to the 20th order;

[0088] Step 8: Calculate the polygon index KndB = a*AdB+(1-a)*RmsdB=0.5*20.269dB+0.5

[0089] *19.066dB=19.668dB;

[0090] The high-speed rail dynamic safety monitoring system usually sets a unified alarm threshold for standardized management. The evaluation values ​​obtained based on different algorithms are different, but the preset alarm threshold will not be adjusted accordingly according to different algorithms. In this regard, the algorithm provider can set the corresponding correction coefficient according to the system preset alarm threshold to match the system usage requirements. In this embodiment, the correction coefficient is -4dB, and the corrected polygon index is:

[0091] KndB=19.668dB-4dB=15.668dB;

[0092] Step 9: Figure 5 Without averaging, the polygon index value fluctuates greatly due to the short-term force fluctuations of the wheel and the dynamic change characteristics of the vibration signal. In some periods, it has exceeded the warning threshold of 18dB, resulting in misjudgment. Figure 6 ,After 60 seconds of averaging, the polygon evaluation index is more stable and reliable;

[0093] Step 10: Figure 7 Before processing, the polygon order index data fluctuates due to the short-term force fluctuations of the wheel and the dynamic change characteristics of the vibration signal, which interferes with the judgment condition of "the polygon evaluation index KdB exceeds the threshold for n seconds continuously and the order Poly is the same", which will cause the fault to be missed. Figure 8 ,After adopting the method of selecting the polygon order index that appears most times within 60 seconds as the polygon order evaluation index, the polygon order evaluation index is more stable and reliable;

[0094] Step 11: Output polygon evaluation index KdB=14.233dB, polygon order evaluation index Poly=20;

[0095] Step 12: Fig. 9 and Fig.10 , when the vehicle speed TSpd≥100km / h, the polygon evaluation index KdB is compared with the alarm threshold: when KdB<10(dB), the operation is normal, and the state value ALTst=0; whenever the polygon evaluation index KdB is 10(dB)≤KdB<18(dB) for 20 consecutive seconds, and the order Poly is the same, the state value ALTst=1, and the predicted state times ALTn1 are increased by 1;

[0096] like Fig.11 , when the vehicle speed TSpd≤100km / h, the counting of ALTn1, ALTn2, and ALTn3 is terminated; when the total number of fault states ALTn1+ALTn2+ALTn3≥alarm judgment times ALTN=100, the alarm state ALTtr is the state value ALTst=1 corresponding to the maximum value ALTn1 among ALTn1, ALTn2, and ALTn3;

[0097] Step 13: Output alarm status ALTtr=1; and set the values ​​of ALTn1, ALTn2, ALTn3, and ALTtr to 0.

[0098] A total of 6 prediction alarms were output throughout the day;

[0099] The intelligent high-speed rail dynamic safety monitoring system has been running for 2 months using this algorithm, and has prompted alarms for 136 times in 14 wheel positions. After manual roundness measurement of the wheels at the alarm positions, it is found that there are indeed polygonal defects that are the same as the monitoring results, with an accuracy rate of 100%. The monitoring results of the method of the present invention are accurate, and the monitoring method has extremely high monitoring stability.

Claims

1. A method for online monitoring of train wheel polygonal faults, characterized in that: Obtain the real-time speed and wheel diameter information of the train; calculate the real-time rotation frequency of the wheel according to the real-time speed and wheel diameter, and then calculate the wheel polygon order frequency value sequence; Acquire the collected vertical vibration acceleration signal of the train wheel; perform low-pass filtering on the vibration acceleration signal, intercept the middle section, and perform frequency domain transformation at the same time; According to the vibration acceleration signal obtained after intercepting the middle section, the effective value in the time domain is calculated first, and then the vibration effective value index is calculated; According to the frequency domain data obtained by frequency domain transformation, the peak value and the corresponding frequency in the frequency domain data are first extracted to form a peak sequence, and the frequency domain amplitude corresponding to the wheel polygon order frequency value sequence is extracted from the peak sequence using a difference threshold algorithm to form a wheel polygon amplitude-frequency feature sequence; According to the wheel polygon amplitude-frequency characteristic sequence, a group of data with the largest amplitude is taken as the characterization quantity, and the polygon amplitude index and polygon order index are calculated; According to the vibration effective value index and the polygon amplitude index, the polygon index is calculated; According to the polygon index and the polygon order index, a polygon evaluation index and a polygon order evaluation index are calculated; When the vehicle speed is greater than the preset value, the polygon evaluation index is compared with the alarm threshold, and combined with the polygon order evaluation index, it is judged whether the fault state counting condition is met and counted; when the vehicle speed is less than the preset value, it is judged whether the total number of fault states exceeds the alarm judgment number, the alarm state is output, and the relevant alarm parameters are set to 0; The steps include: Step 1: Obtain the real-time speed TSpd and wheel diameter TDia of the train; Step 2: Calculate the real-time wheel rotation frequency St = TSpd*1000 / (3.6*TDia*π) based on the real-time vehicle speed TSpd and the wheel diameter TDia, and then calculate the wheel polygon order frequency value sequence On(St); Step 3: Obtain the collected vertical vibration acceleration signal of the train wheel; Step 4: After low-pass filtering the vertical vibration acceleration signal of the train wheel obtained in step 3, the middle section is intercepted and FFTW frequency domain transformation is performed at the same time; Step 5: According to the vibration acceleration signal obtained after intercepting the middle section in step 4, first calculate the time domain effective value RMS, and then calculate the vibration effective value index RmsdB=20log(RMS); Step 6: Based on the frequency domain data obtained by the FFTW frequency domain transformation in step 4, first extract a certain number of peaks and corresponding frequencies in the frequency domain data to form a P(x, y) sequence, then sort the P(x, y) sequence from large to small according to the frequency domain amplitude to form a peak sequence Py(x, y), and use the difference threshold algorithm to extract the frequency domain amplitude corresponding to the wheel polygon order frequency value sequence from the peak sequence to form the wheel polygon amplitude-frequency feature sequence OPn(x, y); Step 7: Based on the wheel polygon amplitude-frequency characteristic sequence OPn(x,y) obtained in step 6, take the set of data OPmax(x,y) with the largest frequency domain amplitude as the characterization quantity, and calculate the polygon amplitude index AdB=20log(OPmax(y)) and the polygon order index Polyn=OPmax(x) / St; Step 8: According to the vibration effective value index RmsdB obtained in step 5, the polygon amplitude index AdB value and the weighting coefficient a obtained in step 7, the polygon index KndB=a*AdB+(1-a)*RmsdB is calculated; Step 9: Average the polygon index KndB from 1 second to N seconds, and use it as the polygon evaluation index KdB of the Nth second = Average(∑1 N KndB, N), the KdB value in the initial N-1 seconds is 0; step 10: Select the polygon order index Polyn that appears most times between 1 second and N seconds as the final polygon order evaluation index Poly; Step 11: Output the polygon evaluation index KdB and the polygon order evaluation index Poly; Step 12: When the vehicle speed is greater than the preset value, the polygon evaluation index KdB is compared with the alarm threshold, and combined with the polygon order evaluation index Poly, it is determined whether the fault state counting condition is met, and the fault state times of the prediction state times ALTn1 or the warning state times ALTn2 or the alarm state times ALTn3 are counted; when the vehicle speed is less than the preset value, it is determined whether the sum of the fault state times exceeds the alarm judgment times, that is, whether ALTn1+ALTn2+ALTn3 is greater than or equal to ALTN, if it exceeds, the alarm state ALTtr is the state value ALTst corresponding to the maximum value among ALTn1, ALTn2, and ALTn3, otherwise the alarm state ALTtr=0; Step 13: Output the alarm status ALTtr; and set the values ​​of ALTn1, ALTn2, ALTn3, and ALTtr to 0; the wheel polygon order frequency value sequence On(St) is the product of the wheel rotation frequency St and the polygon order On sequence, and the order On of the wheel polygon fault is between 15-40.

2. The method for online monitoring of train wheel polygonal faults according to claim 1 is further characterized in that: The intercepted middle section intercepts the vibration acceleration signal between 10% and 90% of the vibration acceleration signal.

3. The method for online monitoring of train wheel polygonal faults according to claim 1 is further characterized in that: The method for extracting the peak sequence Py(x,y) includes: using a least squares fitting algorithm to extract 40-60 maximum peaks and corresponding frequencies in the frequency domain data to form a P(x,y) sequence, and then sorting the P(x,y) sequence from large to small according to the frequency domain amplitude to form a peak sequence Py(x,y), which is a set of coordinate values, the horizontal axis x represents the frequency, and the vertical axis y represents the corresponding frequency domain amplitude.

4. The method for online monitoring of train wheel polygonal faults according to claim 1 is further characterized in that: The method of using a difference threshold algorithm to extract the frequency domain amplitude corresponding to the wheel polygon order frequency value sequence from the peak sequence to form a wheel polygon amplitude-frequency characteristic sequence OPn(x,y) includes: setting a coupling coefficient, coupling the wheel polygon order frequency value sequence On(St) with the frequency values ​​in the peak sequence Py(x,y) one by one, when |On(St)-Py(x)|≤coupling coefficient, it means that the frequency value is a fault characteristic frequency, extracting the frequency value and the corresponding frequency domain amplitude output to form a wheel polygon amplitude-frequency characteristic sequence OPn(x,y), the sequence OPn(x,y) is a set of coordinate values, the horizontal axis x represents the wheel polygon order frequency value, and the vertical axis y represents the corresponding frequency domain amplitude.

5. The method for online monitoring of train wheel polygonal faults according to claim 1 is further characterized in that: The weighting coefficient a ranges from 0.2 to 0.

8.

6. The method for online monitoring of train wheel polygonal faults according to claim 1 is further characterized in that: The method for determining whether to alarm is as follows: when the vehicle speed is greater than the preset value, the polygon evaluation index KdB is compared with the alarm threshold; when KdB is less than the minimum value of the alarm threshold, the operation is normal, and the state value ALTst = 0; Whenever the lowest alarm threshold value is ≤KdB<the middle alarm threshold value for n consecutive seconds, and the order Poly is the same, the state value ALTst=1, and the number of predicted states ALTn1 increases by 1; whenever the middle alarm threshold value is ≤KdB<the highest alarm threshold value for n consecutive seconds, and the order Poly is the same, the state value ALTst=2, and the number of early warning states ALTn2 increases by 1; whenever KdB≥the highest alarm threshold value for n consecutive seconds, and the order Poly is the same, the state value ALTst=3, and the number of alarm states ALTn3 increases by 1 ; When the vehicle speed is less than the preset speed, determine the number of fault states and whether it exceeds the alarm judgment number, that is, whether ALTn1+ALTn2+ALTn3 is greater than or equal to ALTN. If it exceeds, the alarm state ALTtr is taken as the state value ALTst corresponding to the maximum value of ALTn1, ALTn2, and ALTn3, and an alarm prompt is given; otherwise, the alarm state ALTtr=0, and no alarm prompt is given; further, ALTst=1 is a pre-judgment, ALTst=2 is a warning, and ALTst=3 is an alarm.

7. The method for online monitoring of train wheel polygonal faults according to claim 6, characterized in that: In the continuous n seconds, the value range of n is 1-60.

8. The method for online monitoring of train wheel polygonal faults according to claim 6, characterized in that: The value range of the alarm determination times ALTN is 50-300.

Images