A real-time online method for detecting polygonal faults in high-speed train wheels

Abstract

The present invention belongs to the technical field of high-speed train fault detection and discloses a method for real-time online detection of polygonal faults in high-speed train wheels. The method comprises the following steps: reading the vehicle speed and diameter, calculating the wheel rotation frequency harmonics, then inputting a vibration acceleration signal, performing a low-pass filter, intercepting the middle section, and simultaneously performing a FFTW frequency domain transform; calculating the time domain effective value (RMS) and then the effective value (Rmsd) of the signal obtained after intercepting the middle section; selecting the polygonal feature with the maximum amplitude to obtain the polygonal feature value; extracting the peak amplitude-frequency feature of the signal obtained by the FFTW frequency domain transform, then performing polygon feature coupling extraction to obtain the calculated polygonal index (Ad); and then calculating the polygonal evaluation index (Kd) to determine whether the threshold is exceeded and whether to trigger an alarm. The detection method of the present invention uses the weighted total of the effective value of the vibration signal and the polygonal feature index as the evaluation criterion, and has stable performance and accurate monitoring, and can monitor medium- and high-speed trains.

Description

Technical Field

[0001] The invention belongs to the technical field of high-speed train fault detection and relates to a method for real-time online detection of high-speed train wheel polygon faults. Background Art

[0002] After decades of development, by the end of 2020, China's operating railway mileage reached 146,000 kilometers, with 37,900 kilometers of high-speed railways in operation, firmly ranking first in the world. As of December 30, 2021, China's high-speed railway mileage exceeded 40,000 kilometers. As of June 20, 2022, nearly 3,200 kilometers of high-speed railways in China were operating at a regular speed of 350 kilometers per hour.

[0003] High-speed trains operate at high speeds, requiring numerous technical indicators to be monitored online. Wheel polygonal failure is one such indicator. Wheel polygonal failure, also known as wheel corrugation or periodic wheel non-circularity, can cause severe vibration and noise on the train, damaging both the track and vehicle components. Wheel polygonalization not only results in greater wheel-rail impact forces and rolling noise, but also generates high-frequency, high-amplitude vibrations that reduce ride comfort and, in severe cases, can pose safety risks and even lead to derailment.

[0004] Train wheel condition monitoring falls into two main categories: manual inspection and real-time online monitoring. Currently, most train maintenance depots in my country rely on manual inspections for wheel condition monitoring. Manual inspections require the vehicle to be disassembled for routine inspections and maintenance, and workers then use visual inspection, keystrokes, ear sounds, or various manual caliper tools to perform checks. Real-time monitoring methods primarily include eddy current, ultrasonic telemetry, laser sensors, vibration acceleration, and imaging.

[0005] Existing manual inspection methods result in excessively long train turnover times, and manual inspections are affected by factors such as human error and operating conditions, resulting in slow, inaccurate, and labor-intensive inspections. Furthermore, this method is only effective for wheel flat faults, and is powerless against wheel polygonal faults. Currently, all real-time monitoring methods for equipment are relatively good at identifying whether a wheel has polygonal faults, but have certain limitations in identifying the order of polygonal wheels. For example, some monitoring methods cannot quantitatively reflect the severity of the fault, and some monitoring methods have poor stability, slow and time-consuming calculations, or inaccurate monitoring results. Furthermore, when the sensor signal experiences a large impact, misjudgment is likely to occur. Summary of the Invention

[0006] The present invention provides a specific solution to this difficult problem. The method of the present invention uses an acceleration sensor for real-time monitoring, which can not only reflect defects but also accurately reflect the degree of damage caused by the defects, and the cost of monitoring installation is low.

[0007] To achieve the above purpose, the technical solution adopted by the present invention to solve the problem is:

[0008] A method for real-time online detection of high-speed train wheel polygon faults comprises the following steps:

[0009] Detect the real-time running speed and current wheel diameter of the high-speed train; calculate the current wheel rotation frequency and the corresponding wheel polygon order frequency value sequence based on the real-time speed and wheel diameter;

[0010] Detect the vertical acceleration signal of the high-speed train axle box vibration; after low-pass filtering the obtained data signal, intercept the middle section and perform frequency domain transformation;

[0011] After intercepting the middle segment of the data signal, first calculate the effective value in the time domain, and then calculate the effective value part of the evaluation index;

[0012] According to the frequency domain data obtained by frequency domain transformation, the maximum peak value and the corresponding frequency sequence in the frequency domain data are first extracted to form a peak sequence. The difference threshold algorithm is used to extract the polygon order feature sequence from the peak sequence, including the polygon order and the frequency domain amplitude corresponding to the order, to complete the extraction of peak amplitude-frequency features.

[0013] According to the extracted polygon order feature sequence, the largest set of polygon features is taken as the representation quantity, and the polygon index and polygon order are calculated;

[0014] The polygon evaluation index is calculated by using the obtained effective value partial evaluation index, polygon index value and weighting coefficient;

[0015] When the vehicle speed TSpd≥100km / h, the polygon evaluation index is compared with the alarm threshold to determine whether it exceeds the set threshold and whether to alarm; based on the comparison result, the polygon index and warning status are output.

[0016] A method for real-time online detection of high-speed train wheel polygon faults, specifically implemented as follows:

[0017] Step 1: Detect the real-time running speed TSpd and current wheel diameter TDia of the high-speed train;

[0018] Step 2: Based on the real-time vehicle speed TSpd and wheel diameter TDia, calculate the current wheel rotational frequency St = TSpd*1000 / (3.6*TDia*π), and calculate the corresponding wheel polygon order frequency value sequence On(St);

[0019] Step 3: Detect the vertical acceleration signal of the high-speed train axle box vibration;

[0020] Step 4: After the data signal obtained in step 3 is low-pass filtered, the middle segment is intercepted and FFTW frequency domain transformation is performed at the same time;

[0021] Step 5: Based on the data signal obtained after intercepting the middle segment in step 4, first calculate the time domain RMS value, and then calculate the RMS partial evaluation index Rmsd = 20log(RMS);

[0022] Step 6: Based on the frequency domain data obtained by FFTW frequency domain transformation in step 4, first extract the maximum peak value and the corresponding frequency P(x,y) sequence in the frequency domain data to form a peak sequence Px(x,y). Then, use the difference threshold algorithm to extract the polygon order feature sequence OPn(x,y) from the peak sequence, including the polygon order frequency and the frequency domain amplitude corresponding to the order frequency, to complete the extraction of peak amplitude-frequency features;

[0023] Step 7: Based on the multi-order feature sequence OPn(x,y) extracted in step 6, take the largest set of polygon features OPmax(x,y) as the characterization quantity, calculate the polygon index Ad=20log(OPmax(y)) and the polygon order Poly=OPmax(x) / St,

[0024] Step 8: The effective value partial evaluation index Rmsd is obtained through step 5, and the polygon index Ad value and weighting coefficient a are obtained in step 7 to calculate the polygon evaluation index Kd = a*Ad+(1-a)*Rmsd;

[0025] Step 9: When the vehicle speed TSpd ≥ 100 km / h, the polygon evaluation index Kd is compared with the alarm threshold to determine whether it exceeds the set threshold and whether to issue an alarm.

[0026] Step 10: Based on the comparison results of step 9, output polygon indicators and warning status.

[0027] In step 2, the wheel polygon order frequency value sequence On(St) is a sequence formed by the product of the wheel rotation frequency St and the polygon fault order On. The order On of the wheel polygon fault is preferably between 15-60.

[0028] Preferably, in step 4, when calculating the effective value part evaluation index, intercepting the middle section is intercepting the data signal between 10% and 90% of the data segment.

[0029] Preferably, in step 6, the method for extracting the peak sequence Px(x,y) includes: using a least squares fitting algorithm to extract the maximum peak and corresponding frequency P(x,y) sequence from the frequency domain data of multiple points; after extraction, sorting the P(x,y) sequence by peak amplitude to obtain the peak sequence Px(x,y), which is a set of coordinate values, where the horizontal axis x represents the frequency and the vertical axis y represents the corresponding peak. Furthermore, 40-60 maximum peaks and corresponding frequencies are extracted from the P(x,y) sequence.

[0030] As a preferred solution, in step 6, the method of extracting the polygonal order characteristic sequence OPn(x,y) from the peak sequence using the difference threshold algorithm includes: setting a coupling coefficient, coupling the wheel polygonal order frequency value sequence On(St) with the frequency values ​​in the peak sequence Px(x,y) one by one, when |On(St)-Px(x)|≤coupling coefficient, it indicates that the frequency value is a fault feature, extracting the frequency value and the corresponding peak output to form a polygonal order characteristic sequence OPn(x,y), the characteristic sequence OPn(x,y) is a set of coordinate values, the horizontal coordinate x represents the extracted order frequency value, and the vertical coordinate y represents the corresponding peak value.

[0031] In step 8, the weighting coefficient a is preferably in the range of 0.2-0.8.

[0032] As a preferred method, in step 9, the method for determining whether to alarm is as follows: compare the polygon evaluation index Kd with the alarm threshold. When the abnormality is continuously displayed and the polygon order Poly is the same, the abnormality is taken as the alarm prompt, otherwise, no alarm prompt is issued. During the actual operation of the train, due to the speed fluctuation and the dynamic change characteristics of the vibration signal, the calculated polygon evaluation index will produce numerical fluctuations. In order to diagnose the wheel rim polygon fault more stably and reliably, the present invention introduces a time dimension parameter, that is, continuously judging whether the polygon evaluation index exceeds the alarm threshold (that is, whether it continuously displays abnormalities) within a certain time period. If the conditions are met, it is diagnosed as a wheel rim polygon fault; preferably, when the abnormality is displayed for more than 10 seconds and the polygon order Poly is the same, an alarm prompt is issued.

[0033] Furthermore, the judgment criteria for whether to display an abnormality are: the polygon evaluation index Kd is less than the lowest value of the alarm threshold, the operation is normal, and the alarm state does not display an abnormality; the situations in which an abnormality is displayed include: the lowest value of the alarm threshold ≤ Kd < the middle value of the alarm threshold, the alarm state is set to a prejudgment alarm; the middle value of the alarm threshold ≤ Kd < the highest value of the alarm threshold, the alarm state is set to a warning alarm; Kd ≥ the highest alarm threshold, the alarm state is set to an alarm.

[0034] The beneficial effects of the present invention are as follows: The polygon fault monitoring method of the present invention selects a weighted approach to evaluate polygon faults, using the weighted sum of the effective value of the vibration signal and the polygon characteristic index as the evaluation criterion. The polygon fault detection method of the present invention has a fast detection process, accurate and stable monitoring results, and low system performance requirements. It can monitor medium- and high-speed trains (greater than 100 km / h), meeting the requirements of a low-cost embedded device real-time online monitoring system. Furthermore, the present invention introduces a time dimension parameter, ensuring more stable and reliable alarm output through a continuous time dimension. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1It is a flow chart of an implementation example of the method of the present invention.

[0036] Figure 2 It is a flow chart of the detection method of the present invention.

[0037] Figure 3 It is a communication monitoring diagram.

[0038] Figure 4 It is the waveform diagram of the measuring point.

[0039] Figure 5 It is the frequency domain amplitude diagram. DETAILED DESCRIPTION

[0040] The specific implementation of the present invention is described in detail below in conjunction with the technical solutions and drawings.

[0041] Example 1

[0042] like Figure 1 As shown, vibration sensors are placed above the train wheel bearings. Each of the eight wheelsets in each carriage is equipped with a vibration acceleration sensor. The data acquisition module acquires the radial vibration signal of the wheel and sends it to the diagnostic analysis algorithm processing module. The data processing module obtains the train synchronous speed and current wheel diameter value from the onboard central control unit via Ethernet to calculate the real-time wheel rotation frequency. The calculation result and alarm information are then sent to the onboard central control unit in real time via the Ethernet port. This method is implemented in the data processing module. The present invention specifically includes the following steps:

[0043] Step 1: Detect the real-time running speed TSpd and current wheel diameter TDia of the high-speed train;

[0044] Step 2: Based on the real-time vehicle speed TSpd and wheel diameter TDia, calculate the current wheel rotational frequency St = TSpd*1000 / (3.6*TDia*π), and calculate the corresponding wheel polygon order frequency value sequence On(St);

[0045] Step 3: Detect the vertical acceleration signal of the high-speed train axle box vibration;

[0046] Step 4: After the data signal obtained in step 3 is low-pass filtered at 2500 Hz, the middle segment is intercepted and FFTW frequency domain transformation is performed at the same time;

[0047] Step 5: Based on the data signal obtained after intercepting the middle segment in step 4, first calculate the time domain RMS value, and then calculate the RMS partial evaluation index Rmsd = 20log(RMS);

[0048] Step 6: Based on the frequency domain data obtained by FFTW frequency domain transformation in step 4, first extract the maximum peak and the corresponding frequency P(x,y) sequence in the frequency domain data to form a peak sequence Px(x,y). Then use the interpolation threshold algorithm to extract the polygon order feature sequence OPn(x,y) from the peak sequence, including the polygon order and the frequency domain amplitude corresponding to the order, to complete the extraction of peak amplitude-frequency features.

[0049] Step 7: Based on the multi-order feature sequence OPn(x,y) extracted in step 6, take the largest set of polygonal features OPmax(x,y) as the representation, calculate the polygon index Ad=20log(OPmax(y)) and the polygon order Poly=OPmax(x) / St,

[0050] Step 8: The effective value partial evaluation index Rmsd is obtained through step 5, the polygon characteristic value and the polygon index Ad value are obtained in step 7, and the polygon evaluation index Kd = a*Ad+(1-a)*Rmsd is calculated with the weighting coefficient a;

[0051] Step 9: When the vehicle speed TSpd ≥ 100 km / h, the polygon evaluation index Kd is compared with the alarm threshold to determine whether it exceeds the set threshold and whether to issue an alarm.

[0052] Step 10: Based on the comparison results of step 9, output polygon indicators and warning status.

[0053] Example 2

[0054] like Figure 2 Shown is a flowchart of the present invention.

[0055] Step 1: Read the real-time train speed TSpd sent by the system and read the current wheel diameter TDia;

[0056] Step 2: Calculate the current wheel rotation frequency St = TSpd * 1000 / (3.6 * TDia * π) based on the real-time vehicle speed and wheel diameter. Calculate the corresponding wheel polygon order frequency value sequence On(St) as the polygon fault feature reference value. The wheel polygon order frequency value sequence On(St) is the product of the wheel rotation frequency St and the polygon fault order On. The polygon fault order On of a wheel is generally between 15 and 60.

[0057] Step 3: Read the vertical acceleration signal of the axle box vibration and perform a 2500Hz low-pass filter. Select a 6th-order Butterworth filter as the polygon evaluation data. The filter transfer function is as follows:

[0058]

[0059] Step 4: Calculate the RMS value of the filtered vibration data. The filtered data may have edge jitter and cause errors, so a portion of the signal is intercepted to calculate the RMS value of the vibration data. The intercepted signal is generally between 10% and 90% of the data segment. Then, the RMS evaluation index Rmsd = 20log(RMS) is calculated.

[0060] Step 5: Perform Fourier transform on the filtered vibration data, obtain the frequency domain data, and use the FFTW formula to calculate the Fourier transform. The calculation formula is as follows:

[0061] in,

[0062] 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;

[0063] Step 6: Extract the maximum peak value and the corresponding frequency P(x,y) sequence from the frequency domain data. The extraction algorithm is the least squares fitting algorithm. It is best to extract 40-60 maximum peak sequences. After extraction, sort them according to the peak amplitude to obtain the peak sequence Px(x,y). This sequence is a set of coordinate values, where the horizontal axis x represents the frequency and the vertical axis y represents the corresponding peak value. In this step, the maximum peak sequence of multiple points is first extracted to ensure that the extracted peak sequence is the possible fault location. Then, the least squares fitting algorithm is used to extract data. This is stable, reliable, time-saving, and fast.

[0064] Step 7: Use the difference threshold algorithm to couple the fault features. The wheel polygon order frequency value sequence On(St) is coupled with the frequency value in the peak sequence Px(x,y). The coupling coefficient is 2.1. The specific calculation method is to compare the absolute value of the difference between the polygon order frequency value sequence and the frequency value in the peak sequence. When |On(St)-Px(x)|≤2.1, it means that the frequency value is a fault feature. The frequency value and the corresponding peak output are recorded. All polygon order frequency values ​​are compared with the peak sequence frequency values ​​in turn to extract the polygon order feature sequence OPn(x,y). The feature sequence OPn(x,y) is a set of coordinate values. The horizontal coordinate x represents the extracted order frequency, and the vertical coordinate y represents the corresponding peak value. The fault feature coupling in this step uses the difference threshold algorithm to eliminate the frequency error caused by floating-point calculations or speed fluctuations, thereby avoiding feature extraction failure.

[0065] Step 8: Take the largest set of polygon features OPmax(x,y) from OPn(x,y) as the characterization quantity. The characterization quantity is a set of coordinates, OPmax(x) represents the order frequency, and OPmax(y) represents the corresponding amplitude. Calculate the polygon index Ad = 20log(OPmax(y)) and the polygon order Poly = OPmax(x) / St.

[0066] Step 9: Calculate the polygonal evaluation index Kd = a*Ad+(1-a)*Rmsd+Kfac. The weighted coefficient a ranges from 0.2 to 0.8, with the optimal value being a=0.5. Kfac is the correction coefficient, ranging from 0 to 50, which is used to adapt to the alarm value standardization of various vehicle models. Introducing a weighting algorithm into the polygonal evaluation index and using the effective value of the vibration data as part of the judgment parameter can ensure more accurate monitoring results.

[0067] Step 10: Vehicle speed TSpd≥100km / h, polygon evaluation index Kd is compared with the alarm threshold: when Kd<44(dB), the operation is normal, and the alarm state ALTst=0; when 44(dB)≤Kd<49(dB), the alarm state is set to prediction ALTst=1; when 49(dB)≤Kd<54(dB), the alarm state is set to warning ALTst=2; when Kd≥54(dB), the alarm state is set to alarm ALTst=3. When the alarm is continuous and the order Poly is the same, the alarm state ALTtr=ALTst is taken; otherwise, the alarm state ALTtr=0 is taken; due to speed fluctuations and the dynamic change characteristics of the vibration signal, the calculated polygon evaluation index will produce numerical fluctuations. In order to stably and reliably diagnose the wheel rim polygon fault, the time dimension parameter is introduced, that is, continuously judging whether the polygon evaluation index exceeds the alarm threshold within a certain time period (more than 10 seconds). If the conditions are met, it is diagnosed as a wheel rim polygon fault;

[0068] Step 11: Output polygon evaluation index Kd, polygon order Poly, and alarm status ALTtr.

[0069] Example 3

[0070] The method of the present invention is applied to a domestic intelligent high-speed railway dynamic safety monitoring system to perform polygonal fault diagnosis, as follows:

[0071] Step 1: Figure 3 shows some public data information sent from the high-speed train central control unit CCU to the vibration data processing module. The information includes train composition, car number, ambient temperature, speed, wheel diameter, etc. The real-time train speed TSpd = 344.34 km / h sent by the reading system and the current wheel diameter TDia = 920 mm are read;

[0072] Step 2: Calculate the current wheel rotation frequency St = TSpd * 1000 / (3.6 * TDia * π) = 33.11 Hz based on the real-time vehicle speed and wheel diameter. Calculate the corresponding wheel polygon order frequency values ​​On(St) = 496.65 Hz, 529.76 Hz ... 662.2 Hz ... 1986.6 Hz sequence.

[0073] Step 3: Read the vertical acceleration signal of the axle box vibration, such as Figure 4 , do 2500Hz low-pass filtering, and choose 6th-order Butterworth filter.

[0074] Step 4: Calculate the RMS value of the filtered data. The filtered data will have edge jitter, so the calculation of the RMS intercepts part of the signal, generally between 10% and 90% of the data segment, and then calculate the RMS partial evaluation index Rmsd = 20log(RMS) = 20log(5.6597g) = 15.055dB;

[0075] Step 5: Calculate the filtered data and perform Fourier transform to obtain frequency domain data, such as Figure 5 , the Fourier transform formula selects FFTW formula to calculate the frequency domain amplitude;

[0076] Step 6: Extract the maximum peak and the corresponding frequency P(x,y) sequence from the frequency domain data. The extraction algorithm is the least squares fitting algorithm. Generally, 40-60 maximum peak sequences are extracted. After extraction, they are sorted according to the peak amplitude Px(x,y). Table 1 shows the top 10 peak sequences.

[0077] Table 1 Peak sequence ranking

[0078] Serial number Frequency / Hz Peak value / g 1 662.00 7.2542 2 628.97 1.2796 3 595.98 0.8245 4 529.92 0.7041 5 1324.25 0.7018 6 666.62 0.5412 7 658.34 0.4571 8 430.26 0.4101 9 648.66 0.4033 10 695.15 0.3688

[0079] Step 7: According to the frequency value of the wheel polygon order On(St) sequence and the frequency value in the peak sequence Px(x, y), the coupling coefficient is 2.1. The specific calculation method is to compare the absolute value of the difference between the polygon order and the frequency value in the peak sequence. When On(St)-Px(x)≤2.1, record the frequency value and the corresponding peak output, and compare all polygon orders with the peak sequence frequency values ​​in turn to extract the polygon order feature sequence OPn(x, y). Table 2 shows some of the extraction results.

[0080] Table 2 Order index extraction results

[0081] Serial number Frequency / Hz Peak value / g Polygonal order sequence frequency 1 662.00 7.2542 662.2 2 628.97 1.2796 629.09 3 595.98 0.8245 595.98 4 529.92 0.7041 529.76 5 1324.25 0.7018 1324.4 6 430.26 0.4101 430.43 7 695.15 0.3688 695.31 8 1125.68 0.2466 1125.74 9 728.98 0.2336 728.42 10 828.66 0.1923 827.75

[0082] Step 8: Take the largest set of polygon features OPmax(x, y) from OPn(x, y) as the characterization quantity, calculate the polygon index Ad = 20log(OPmax(y)) = 20log(7.2542g) = 17.212dB, and calculate the polygon order Poly = OPmax(x) / St = 662.00 / 33.11 = 19.99, rounded to 20th order;

[0083] Step 9: Calculate polygon evaluation index

[0084] Kd=0.5*17.212dB+0.5*15.055dB=16.134dB.

[0085] High-speed rail dynamic safety monitoring systems typically set a unified alarm threshold for standardized management. The evaluation values ​​obtained based on different algorithms are different, but the preset alarm thresholds are not adjusted accordingly. Therefore, the algorithm provider can set a corresponding correction factor based on the system's preset alarm thresholds to match the system's usage requirements. In this embodiment, the correction factor is 30dB, and the corrected polygon evaluation index value is:

[0086] Kd=0.5*17.212dB+0.5*15.055dB+30dB=46.134dB

[0087] Step 10: When vehicle speed TSpd ≥ 100 km / h, compare the polygon evaluation index Kd with the alarm threshold: When Kd < 44 (dB), normal operation is achieved, and the alarm state ALTst = 0; when 44 (dB) ≤ Kd < 49 (dB), the alarm state is set to prejudgment ALTst = 1; when 49 (dB) ≤ Kd < 54 (dB), the alarm state is set to warning ALTst = 2; when Kd ≥ 54 (dB), the alarm state is set to alarm ALTst = 3. When the alarm lasts for more than 10 seconds and the order Poly is the same, the alarm state ALTtr = ALTst is assumed; otherwise, the alarm state ALTtr = 0 is assumed.

[0088] Since the polygon evaluation index is 46.134 dB, reaching the predicted alarm level and the alarm continues for 10 seconds, ALTtr=1.

[0089] Step 11: Output polygon evaluation index Kd=46.134, polygon order Poly=20, and alarm status ALTtr=1.

[0090] The intelligent high-speed rail dynamic safety monitoring system outputs a test result every 0.5 seconds. This algorithm takes 50 milliseconds to calculate, which is superior to the 0.1-second operation speed of other algorithms. Verification by wheel polygon bench testing conducted at the client's National Engineering Laboratory shows that the test results are consistent with the actual state of the wheel. Furthermore, the intelligent high-speed rail dynamic safety monitoring system using this algorithm has been in continuous operation for 15 months, with two alarms. Manual wheel roundness measurements of the wheels at the alarm location confirmed the presence of polygonal defects identical to those in the monitoring results, with 100% accuracy. The monitoring results using the method of the present invention are accurate, and the monitoring method has extremely high monitoring stability.

Claims

1. A method for real-time online detection of high-speed train wheel polygonal faults, characterized in that: include: Detect the real-time running speed and current wheel diameter of high-speed trains; According to the real-time vehicle speed and wheel diameter, the current wheel rotation frequency is calculated, and the corresponding wheel polygon order frequency value sequence is calculated; Detect the vertical acceleration signal of the high-speed train axle box vibration; after low-pass filtering the obtained data signal, intercept the middle section and perform frequency domain transformation; After intercepting the middle segment of the data signal, first calculate the effective value in the time domain, and then calculate the effective value part of the evaluation index; According to the frequency domain data obtained by frequency domain transformation, the maximum peak value and the corresponding frequency sequence in the frequency domain data are first extracted to form a peak sequence. The difference threshold algorithm is used to extract the polygon order feature sequence from the peak sequence, including the polygon order and the frequency domain amplitude corresponding to the order, to complete the extraction of peak amplitude-frequency features. According to the extracted polygon order feature sequence, the largest set of polygon features is taken as the representation quantity, and the polygon index and polygon order are calculated; The polygon evaluation index is calculated by using the obtained effective value partial evaluation index, polygon index value and weighting coefficient; When the vehicle speed TSpd≥100km / h, the polygon evaluation index is compared with the alarm threshold to determine whether it exceeds the set threshold and whether to alarm; based on the comparison result, the polygon index and warning status are output.

2. The method for real-time online detection of high-speed train wheel polygon faults according to claim 1, characterized in that: The steps include: Step 1: Detect the real-time running speed TSpd and current wheel diameter TDia of the high-speed train; Step 2: Based on the real-time vehicle speed TSpd and wheel diameter TDia, calculate the current wheel rotational frequency St = TSpd*1000 / (3.6*TDia*π), and calculate the corresponding wheel polygon order frequency value sequence On(St); Step 3: Detect the vertical acceleration signal of the high-speed train axle box vibration; Step 4: After the data signal obtained in step 3 is low-pass filtered, the middle segment is intercepted and FFTW frequency domain transformation is performed at the same time; Step 5: Based on the data signal obtained after intercepting the middle segment in step 4, first calculate the time domain RMS value, and then calculate the RMS partial evaluation index Rmsd = 20log(RMS); Step 6: Based on the frequency domain data obtained by FFTW frequency domain transformation in step 4, first extract the maximum peak value and the corresponding frequency P(x,y) sequence in the frequency domain data to form a peak sequence Px(x,y). Then, use the difference threshold algorithm to extract the polygon order feature sequence OPn(x,y) from the peak sequence, including the polygon order frequency and the frequency domain amplitude corresponding to the order frequency, to complete the extraction of peak amplitude-frequency features; Step 7: Based on the multi-order feature sequence OPn(x,y) extracted in step 6, take the largest set of polygon features OPmax(x,y) as the characterization quantity, calculate the polygon index Ad=20log(OPmax(y)) and the polygon order Poly=OPmax(x) / St, Step 8: The effective value partial evaluation index Rmsd is obtained through step 5, and the polygon index Ad value and weighting coefficient a are obtained in step 7 to calculate the polygon evaluation index Kd = a*Ad+(1-a)*Rmsd; Step 9: When the vehicle speed TSpd ≥ 100 km / h, the polygon evaluation index Kd is compared with the alarm threshold to determine whether it exceeds the set threshold and whether to issue an alarm; Step 10: Based on the comparison results of step 9, output polygon indicators and warning status.

3. The method for real-time online detection of high-speed train wheel polygon faults according to claim 2, further characterized by: The wheel polygon order frequency value sequence On(St) is a sequence formed by the product of the wheel rotation frequency St and the polygon fault order On. The order On of the wheel polygon fault is between 15 and 60.

4. The method for real-time online detection of high-speed train wheel polygon faults according to claim 2 or 3, further characterized in that: When calculating the effective value part of the evaluation index, the middle segment is intercepted to intercept the data signal between 10% and 90% of the data segment.

5. The method for real-time online detection of high-speed train wheel polygon faults according to claim 2, further characterized by: The peak sequence Px(x,y) extraction method includes: using a least squares fitting algorithm to extract the maximum peak and the corresponding frequency P(x,y) sequence in the frequency domain data of multiple points, and sorting them according to the peak amplitude after extraction to obtain the peak sequence Px(x,y), which is a set of coordinate values, where the horizontal coordinate x represents the frequency and the vertical coordinate y represents the corresponding peak.

6. The method for real-time online detection of high-speed train wheel polygon faults according to claim 5, characterized in that: Extract 40-60 maximum peaks and corresponding frequencies to form the P(x,y) sequence.

7. A method for real-time online detection of high-speed train wheel polygon faults according to claim 2, 3 or 5, characterized in that: The method of extracting a polygonal order characteristic sequence OPn(x,y) from a peak sequence using a difference threshold algorithm includes: setting a coupling coefficient, coupling the wheel polygonal order frequency value sequence On(St) with the frequency values ​​in the peak sequence Px(x,y) one by one, when |On(St)-Px(x)|≤coupling coefficient, it indicates that the frequency value is a fault feature, extracting the frequency value and the corresponding peak output to form a polygonal order characteristic sequence OPn(x,y), the characteristic sequence OPn(x,y) is a set of coordinate values, the horizontal coordinate x represents the extracted order frequency value, and the vertical coordinate y represents the corresponding peak value.

8. A method for real-time online detection of high-speed train wheel polygonal faults according to claim 2, 3 or 5, characterized in that: The weighting coefficient a ranges from 0.2 to 0.

8.

9. A method for real-time online detection of high-speed train wheel polygonal faults according to claim 2, 3 or 5, characterized in that: The method for determining whether to alarm is: compare the polygon evaluation index Kd with the alarm threshold. When an abnormality is displayed for a continuous period of time and the polygon order Poly is the same, the abnormality is taken as the alarm prompt, otherwise, no alarm prompt is issued.

10. The method for real-time online detection of high-speed train wheel polygon faults according to claim 9, characterized in that: When the polygon evaluation index Kd is less than the lowest value of the alarm threshold, the operation is normal and the alarm status does not show abnormalities; the situations in which abnormalities are shown include: when the lowest value of the alarm threshold ≤ Kd < the middle value of the alarm threshold, the alarm status is set to a prejudgment alarm; when the middle value of the alarm threshold ≤ Kd < the highest value of the alarm threshold, the alarm status is set to a warning alarm; when Kd ≥ the highest alarm threshold, the alarm status is set to an alarm.

Images