Railway wheel flat detection system, railway wheel flat detection method, and railway wheel flat detection program

Abstract

To provide a railroad wheel flat detection system that can detect a wheel flat at low cost even when detecting a wheel flat for a large number of wheels.SOLUTION: A railroad wheel flat detection system 1A includes: a frequency analysis unit 10A for converting a vibration sensor signal measured by a vibration sensor set in a rail to a first time signal showing a signal frequency of each frequency band; a joint vibration removing unit 40A for calculating a second time signal obtained by removing a signal of a vibration resulting from a rail joint from a correspondence signal corresponding to the first time signal; a flat detection unit 80A for determining whether there is a wheel flat from the second time signal; and an output unit 90 for outputting the result of determination of the flat detection unit 80A. The joint vibration removal unit 40A calculates a second time signal by removing a third time signal from the correspondence signal when the signal intensity of a third time signal including the frequency of a vibration resulting from the rail joint extracted from the correspondence signal is larger than a specific value.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a railway wheel flat detection system, a railway wheel flat detection method, and a railway wheel flat detection program for detecting a railway wheel flat. [Background technology]

[0002] Wheel flats are scratches that occur on the wheel tread due to friction between the rail and wheel when a moving train applies heavy braking, such as sudden braking. Wheel flats can cause abnormal noise or vibration, resulting in a worsening ride and damage to the rail, so early detection of wheel flats is essential. One technology for detecting wheel flats uses vibration data obtained from vibration sensors installed on the axles of the train.

[0003] For example, the abnormality diagnosis device in Patent Document 1 determines whether abnormal vibration is caused by a wheel flat or an axle bearing based on vibration data obtained from vibration sensors attached to the bearing housing of each wheel. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2007-170815 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the technology in Patent Document 1 uses a vibration sensor attached to the bearing housing of each wheel, so vibration sensors are required for each wheel for which wheel flats need to be detected. As a result, the technology in Patent Document 1 has the problem of high costs when detecting wheel flats on multiple wheels.

[0006] The present disclosure has been made in consideration of the above, and aims to provide a railway wheel flat detection system that can detect wheel flats at low cost even when detecting wheel flats on a large number of wheels. [Means for solving the problem]

[0007] To solve the above-mentioned problems and achieve the object, the railway wheel flat detection system disclosed herein includes a frequency analysis unit that converts a vibration sensor signal, which indicates vibration information measured by a vibration sensor installed on the rail, into a first time signal indicating signal strength for each frequency band, and a joint vibration elimination unit that calculates a second time signal by removing a signal of vibration caused by the rail joint from a corresponding signal corresponding to the first time signal.The railway wheel flat detection system also includes a flat detection unit that determines whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail based on the second time signal, and an output unit that outputs the determination result from the flat detection unit.The joint vibration elimination unit extracts a third time signal from the corresponding signal, which includes the frequency of vibration caused by the rail joint, and calculates the second time signal by removing the third time signal from the corresponding signal if the signal strength of the third time signal exceeds a specific value. [Effects of the Invention]

[0008] The railway wheel flat detection system according to the present disclosure has the advantage of being able to detect wheel flats at low cost even when detecting wheel flats on a large number of wheels. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram showing a configuration of a railway wheel flat detection system according to a first embodiment. [Figure 2] FIG. 1 is a diagram for explaining the arrangement positions of a vibration sensor and a wheel detection sensor connected to the railway wheel flat detection system according to the first embodiment. [Figure 3]FIG. 1 is a diagram for explaining an example of a vibration sensor signal input to the railway wheel flat detection system according to the first embodiment; [Figure 4] FIG. 1 is a diagram for explaining an example of a frequency spectrum signal calculated by a normalization unit of the railway wheel flat detection system according to the first embodiment; [Figure 5] FIG. 1 is a diagram showing an example of a first independent component signal calculated by an independent component decomposition unit of the railway wheel flat detection system according to the first embodiment; [Figure 6] FIG. 10 is a diagram showing an example of a second independent component signal calculated by the independent component decomposition unit of the railway wheel flat detection system according to the first embodiment. [Figure 7] FIG. 10 is a diagram showing an example of a third independent component signal calculated by the independent component decomposition unit of the railway wheel flat detection system according to the first embodiment. [Figure 8] FIG. 10 is a diagram showing an example of a fourth independent component signal calculated by the independent component decomposition unit of the railway wheel flat detection system according to the first embodiment. [Figure 9] FIG. 1 is a diagram showing an example of a frequency spectrum signal calculated by a removal width expansion unit of the railway wheel flat detection system according to the first embodiment; [Figure 10] FIG. 1 is a diagram showing an example of a frequency spectrum signal calculated by a similarity filter unit of the railway wheel flat detection system according to the first embodiment; [Figure 11] FIG. 10 is a diagram showing an example of a time signal calculated by a flat detection unit of the railway wheel flat detection system according to the first embodiment. [Figure 12] FIG. 10 is a diagram showing an example of a wheel detection output and a flat detection determination result displayed by a display unit of the railway wheel flat detection system according to the first embodiment. [Figure 13] 1 is a flowchart showing an operation processing procedure of the railway wheel flat detection system according to the first embodiment; [Figure 14] FIG. 10 is a diagram showing an example of a norm calculated by an independent component decomposition unit of the railway wheel flat detection system according to the second embodiment. [Figure 15] FIG. 10 is a diagram showing an example of a first frequency set by an independent component remover of the railway wheel flat detection system according to the third embodiment when calculating signal elements of a frequency spectrum signal group. [Figure 16] FIG. 10 is a diagram showing an example of the norm of a first peak calculated by the independent component removal unit of the railway wheel flat detection system according to the third embodiment; [Figure 17] FIG. 10 is a diagram showing an example of the norm of a second peak calculated by the independent component removal unit of the railway wheel flat detection system according to the third embodiment; [Figure 18] FIG. 10 is a diagram showing an example of the norm of a third peak calculated by the independent component removal unit of the railway wheel flat detection system according to the third embodiment. [Figure 19] FIG. 10 is a diagram showing an example of the norm of the reverberation noise of the third peak calculated by the independent component removal unit of the railway wheel flat detection system according to the third embodiment. [Figure 20] FIG. 10 is a diagram showing the configuration of a railway wheel flat detection system according to a fourth embodiment. [Figure 21] FIG. 10 is a diagram showing an example of a time signal limited to a first frequency band calculated by a frequency analysis unit of the railway wheel flat detection system according to the fourth embodiment. [Figure 22] FIG. 10 is a diagram showing an example of a time signal limited to a second frequency band calculated by a frequency analysis unit of the railway wheel flat detection system according to the fourth embodiment. [Figure 23] FIG. 10 is a diagram for explaining an example of a threshold value set by an independent component decomposition unit of the railway wheel flat detection system according to the fourth embodiment. [Figure 24] FIG. 10 is a diagram showing an example of an independent component signal extracted from a time signal limited to a first frequency band by an independent component decomposition unit of the railway wheel flat detection system according to the fourth embodiment; [Figure 25] FIG. 10 is a diagram showing an example of an independent component signal extracted from a time signal limited to a second frequency band by an independent component decomposition unit of the railway wheel flat detection system according to the fourth embodiment; [Figure 26] FIG. 10 is a diagram showing an example of a time signal obtained by removing noise components from a time signal limited to a second frequency band by an independent component removal unit of the railway wheel flat detection system according to the fourth embodiment; [Figure 27] 10 is a flowchart showing an operation processing procedure of the railway wheel flat detection system according to the fourth embodiment. [Figure 28] FIG. 1 is a diagram showing an example of the configuration of a processing circuit when the processing circuit included in the railway wheel flat detection system according to the first and second embodiments is realized by a processor and a memory. [Figure 29] FIG. 1 is a diagram illustrating an example of a processing circuit when the processing circuit included in the railway wheel flat detection system according to the embodiment is configured with dedicated hardware. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, a railway wheel flat detection system, a railway wheel flat detection method, and a railway wheel flat detection program according to embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0011] Embodiment 1 1 is a diagram showing the configuration of a railway wheel flat detection system according to a first embodiment. The railway wheel flat detection system 1A includes a frequency analysis unit 10A, a normalization unit 20, a data extraction unit 30, a seam vibration elimination unit 40A, a data combination unit 50, a elimination width expansion unit 60, a similarity filter unit 70, a flat detection unit 80A, an output unit 90, and a display unit 91. Note that the display unit 91 and the railway wheel flat detection system 1A may be configured separately.

[0012] The seam vibration elimination unit 40A has an independent component decomposition unit 41A and an independent component elimination unit 42A. The frequency analysis unit 10A is connected to the vibration sensor 3 (described later) and receives a signal (a vibration sensor signal X1 (described later)) sent from the vibration sensor 3. The similarity filter unit 70 is connected to the wheel detection sensor 2 (described later) and receives a signal (a wheel detection sensor signal X2 (described later)) sent from the wheel detection sensor 2.

[0013] 2 is a diagram illustrating the positions of the vibration sensor and the wheel detection sensor connected to the railway wheel flat detection system according to the embodiment 1. The vibration sensor 3 and the wheel detection sensor 2 are installed near a rail 5 on which a vehicle 4, which is a railway vehicle, runs.

[0014] For example, one vibration sensor 3 is arranged for one rail 5. That is, for two rails 5, one vibration sensor 3 is attached to one rail 5 and one vibration sensor 3 is attached to the other rail 5. The two vibration sensors 3 are arranged, for example, at positions facing each other in a direction perpendicular to the rails 5.

[0015] Furthermore, one wheel detection sensor 2 is arranged for one rail 5. That is, for two rails 5, one wheel detection sensor 2 is arranged near one rail 5, and one wheel detection sensor 2 is arranged near the other rail 5. In order to calculate the speed of the vehicle 4 from the difference in detection times of the two wheel detection sensors 2 and the difference in arrangement distances, one wheel detection sensor 2 and the other wheel detection sensor 2 are arranged at a certain distance in the longitudinal direction of the rails.

[0016] The vibration sensor 3 is a sensor that measures vibrations that occur when the vehicle 4 is traveling. The vibration sensor 3 measures vibrations of the rail 5 that are excited by the traveling vehicle 4. The vibration sensor 3 is, for example, an acceleration sensor. The vibration sensor 3 sends vibration data indicating information about the measured vibrations to the frequency analysis unit 10A as a vibration sensor signal X1. In this way, the vibration sensor 3 detects vibrations of the rail 5 and outputs an output signal indicating the detection result to the frequency analysis unit 10A as the vibration sensor signal X1.

[0017] The wheel detection sensor 2 is a sensor that detects the wheels of a vehicle 4 traveling on the rails 5. The wheel detection sensor 2 detects wheels that pass the position where the wheel detection sensor 2 is located. Specifically, the wheel detection sensor 2 has a detection range in a direction perpendicular to the rails 5, and outputs an ON signal indicating that a wheel has been detected when a wheel enters this detection range. Furthermore, the wheel detection sensor 2 does not output an ON signal when a wheel does not enter the detection range. The wheel detection sensor 2 sends information indicating that a wheel has been detected to the similarity filter unit 70 as a wheel detection sensor signal X2. In this way, the wheel detection sensor 2 detects a wheel and sends an output signal indicating the detection result to the similarity filter unit 70 as the wheel detection sensor signal X2. The wheel detection sensor 2 is also capable of detecting a traveling vehicle 4.

[0018] FIG. 3 is a diagram for explaining an example of a vibration sensor signal input to the railway wheel flat detection system according to the first embodiment. The horizontal axis of the graph shown in FIG. 3 represents time, and the vertical axis represents vibration amplitude. The graph shown in FIG. 3 illustrates the vibration sensor signal Sa1 obtained during a specific period, rather than the entire vibration sensor signal X1 obtained for the entire train. In other words, the graph shown in FIG. 3 illustrates the vibration sensor signal Sa1 (part of the vibration sensor signal X1) obtained for part of the train.

[0019] 3 shows a vibration sensor signal Sa1 obtained for a portion of one train, but the frequency analysis unit 10A and the normalization unit 20 perform arithmetic processing on the entire vibration sensor signal X1 obtained for the entire train. Here, the frequency analysis unit 10A performs arithmetic processing on the vibration sensor signal Sa1, and the normalization unit 20 performs arithmetic processing on a frequency spectrum signal Sb1 (not shown), which is the result of the calculation by the frequency analysis unit 10A.

[0020] Peaks P1, P2, and P3 of vibration sensor signal Sa1 shown in Figure 3 are vibration data (vibration data of joint vibration) when vehicle 4 passes over a joint in rail 5, and peaks P4 and P5 are vibration data (vibration data of flat vibration) when the flat surface of the wheel comes into contact with rail 5. At this point, railway wheel flat detection system 1A cannot determine which signal is vibration data of joint vibration and which signal is vibration data of flat vibration, but in subsequent processing, it extracts only the vibration data of flat vibration and determines whether a wheel flat has occurred. Vibration sensor signal X1, which includes vibration sensor signal Sa1, is detected by vibration sensor 3, and vibration sensor 3 outputs it to frequency analysis unit 10A.

[0021] When a vibration sensor signal X1 including a vibration sensor signal Sa1 is input, the frequency analysis unit 10A performs a short-time Fourier transform on the vibration sensor signal X1 to calculate a frequency spectrum signal including a frequency spectrum signal (two-dimensional spectrum signal) Sb that represents the frequency spectrum at each time. The frequency spectrum signal for the entire train including the frequency spectrum signal Sb1 is a time signal that indicates the intensity for each frequency band of the vibration sensor signal X1, which is vibration data. The frequency spectrum signal Sb1 is a signal element included in the frequency spectrum signal Sb, and corresponds to the vibration sensor signal Sa1. In the first embodiment, the frequency spectrum signal for the entire train including the frequency spectrum signal Sb1 is a first time signal.

[0022] Although the case where the frequency analysis unit 10A uses a short-time Fourier transform for frequency analysis has been described here, the frequency analysis unit 10A may also use a wavelet transform for frequency analysis. The frequency analysis unit 10A outputs the frequency spectrum signal Sb1 to the normalization unit 20.

[0023] When the frequency spectrum signal Sb1 is input, the normalization unit 20 calculates the scale of noise other than joint vibration for the time signal of each frequency of the frequency spectrum signal Sb1. The normalization unit 20 calculates the noise scale of the time signal of each frequency by dividing the smallest m% values ​​of the time signal by the value at the m% point of the distribution to which noise other than vibration caused by the joint of the rail 5 (hereinafter referred to as the rail joint) follows, where m is a preset parameter. The normalization unit 20 calculates a frequency spectrum signal Sc in which the noise scale is normalized by dividing the time signal by the calculated scale.

[0024] The normalization performed by the normalization unit 20 makes the signal strength (scale) uniform for each frequency band. As the noise at each frequency differs, the normalization unit 20 averages small noise in the time signal by performing normalization, thereby making the signal strength uniform for each frequency band. In this case, the normalization unit 20 normalizes the noise scale while excluding the noise due to seam vibration, so that the signal strength for each frequency band of noise other than the seam vibration can be accurately made uniform.

[0025] The noise contained in the frequency spectrum signal Sb1 is obtained by performing a short-time Fourier transform on the Gaussian noise with an expectation value of 0 and a variance of σ^2 contained in the vibration sensor signal X1. Therefore, the noise contained in the frequency spectrum signal Sb1 follows a gamma distribution with k=1 / 2 and θ=2σ^2 for the DC component, and an exponential distribution with λ=1 / (2σ^2) for frequency bands other than the DC component. The normalization unit 20 outputs the normalized frequency spectrum signal Sc to the data extraction unit 30.

[0026] 4 is a diagram for explaining an example of a frequency spectrum signal calculated by the normalization unit of the railway wheel flat detection system according to the first embodiment. The horizontal axis of the graph shown in FIG. 4 represents time, and the vertical axis represents vibration frequency. FIG. 4 shows a color map in which frequency is represented by color.

[0027] 4 shows the frequency spectrum signal Sc1 obtained by performing arithmetic processing on the frequency spectrum signal Sb1 by the normalization unit 20. The frequency spectrum signal Sc contains non-periodic noise N1, which is noise obtained by performing a short-time Fourier transform on the Gaussian noise described above.

[0028] Peaks P1 to P5 shown in Fig. 4 correspond to peaks P1 to P5 shown in Fig. 3. The frequency spectrum signal Sc includes a signal of vibration caused by rail joints and a signal of vibration caused by flat vibration.

[0029] The spectra of peaks P1 to P3 are observed over a wide band, while the spectra of peaks P4 and P5 are observed in the low frequency band. That is, in the case of joint vibration, the spectrum is observed over a wide band from low to high frequency bands, and in the case of flat vibration, the spectrum is observed in the low frequency band.

[0030] When the normalized frequency spectrum signal Sc is input, the data extracting unit 30 calculates a frequency spectrum signal group Scn by extracting the frequency spectrum signal Sc within a specific time interval. That is, the data extracting unit 30 divides the normalized frequency spectrum signal Sc into specific time intervals to calculate frequency spectrum signals Sc1, Sc2, Sc3, . . . , ScN, which are signal elements of the frequency spectrum signal group (time signal group) Scn. In the first embodiment, the frequency spectrum signal group Scn calculated by the data extracting unit 30 is a corresponding signal corresponding to the frequency spectrum signal (first time signal) for the entire train, including the frequency spectrum signal Sb1.

[0031] The data extractor 30 extracts the signal by sliding an extraction window of width T1 by width T2. That is, frequency spectrum signal Sc1 is the signal of frequency spectrum signal Sc from time 0 to time T1, frequency spectrum signal Sc2 is the signal from time T2 to time (T1+T2), and frequency spectrum signal Sc3 is the signal from time 2×T2 to time (T1+2×T2).

[0032] The joint vibration elimination unit 40A, which performs a calculation process later than the calculation process performed by the data extraction unit 30, does not perform the calculation process all at once for the entire train, but performs the calculation process for each specific period. In other words, the joint vibration elimination unit 40A does not perform the calculation process all at once for the entire period of the vibration sensor signal X1 corresponding to the entire train, but performs the calculation process for each period divided by the data extraction unit 30. Therefore, the data extraction unit 30 calculates a frequency spectrum signal group Scn by extracting the frequency spectrum signal Sc from a specific time width. The data extraction unit 30 outputs the extracted frequency spectrum signal group Scn to the independent component decomposition unit 41A. The following describes the case where the frequency spectrum signal Sc1 corresponds to the frequency spectrum signal Sb1.

[0033] When the extracted frequency spectrum signal group Scn is input, the independent component decomposition unit 41A calculates an independent component signal group Sdn by decomposing each signal element of the frequency spectrum signal group Scn into independent components by independent component analysis. In the following, a frequency spectrum signal Sc1, which is a signal element of the frequency spectrum signal group Scn, will be described.

[0034] The independent component decomposition unit 41A, for example, calculates an independent component signal Sd1 from the frequency spectrum signal Sc1, and calculates an independent component signal Sd2 from the frequency spectrum signal Sc2. Each of the independent component signals Sd1 and Sd2 includes one or more independent component signals. The independent component signal Sd1 includes, for example, independent component signals Sd1a, Sd1b, Sd1c, and Sd1d, which will be described later.

[0035] Independent component analysis can decompose each signal element of the frequency spectrum signal group Scn into as many independent component signals as there are frequency bins. However, in practice, to speed up processing, the independent component decomposition unit 41A terminates the decomposition when sufficient decomposition is achieved. For example, when the independent component decomposition unit 41A decomposes the frequency spectrum signal Sc1 into M (M is a natural number) independent component signals, it calculates the AIC (Akaike's Information Criterion) of the difference between the frequency spectrum signal Sc1 and the sum of the M independent component signals. The independent component decomposition unit 41A determines that sufficient decomposition is achieved when the AIC calculated for the M independent component signals is greater than the AIC calculated when the previous independent component signals were decomposed into (M-1) independent component signals.

[0036] The independent component decomposition unit 41A sequentially decomposes the independent component signals of the frequency spectrum signal Sc from the first to the Mth independent component signals. For example, for the frequency spectrum signal Sc1, the independent component decomposition unit 41A decomposes the independent component signal Sd1a shown in FIG. 5 (described later) as the first independent component signal and the independent component signal Sd1b shown in FIG. 6 (described later) as the second independent component signal. After decomposing the second independent component signal, the independent component decomposition unit 41A calculates the AIC of the difference between the two independent component signals Sd1a and Sd1b and the frequency spectrum signal Sc1. That is, the independent component decomposition unit 41A calculates a signal by removing the independent component signals Sd1a and Sd1b corresponding to peaks P1 and P2 from the frequency spectrum signal Sc1 shown in FIG. 4. In this case, the independent component decomposition unit 41A determines that decomposition is sufficient when the calculated AIC is greater than the AIC obtained when the previous independent component signal was decomposed into (M-1) independent component signals (here, 1).

[0037] The independent component decomposition unit 41A calculates the independent component signal group Sdn (independent component signals Sd1, Sd2, Sd3, . . . , SdN) from the extracted frequency spectrum signal group Scn (frequency spectrum signals Sc1, Sc2, Sc3, . . . , ScN). That is, the independent component decomposition unit 41A calculates the independent component signal group Sdn from the frequency spectrum signal group Scn by independent component analysis.

[0038] The independent component decomposition unit 41A outputs the extracted frequency spectrum signal group Scn and independent component signal group Sdn to the independent component removal unit 42A.

[0039] Here, the independent component signals obtained when the independent component decomposition unit 41A performs arithmetic processing on each signal element of the normalized frequency spectrum signal group Scn will be described. Here, the independent component signal Sd1 obtained when the independent component decomposition unit 41A performs arithmetic processing on the frequency spectrum signal Sc1 will be described.

[0040] Fig. 5 is a diagram illustrating an example of a first independent component signal calculated by the independent component decomposition unit of the railway wheel flat detection system according to the first embodiment. Fig. 6 is a diagram illustrating an example of a second independent component signal calculated by the independent component decomposition unit of the railway wheel flat detection system according to the first embodiment. Fig. 7 is a diagram illustrating an example of a third independent component signal calculated by the independent component decomposition unit of the railway wheel flat detection system according to the first embodiment. Fig. 8 is a diagram illustrating an example of a fourth independent component signal calculated by the independent component decomposition unit of the railway wheel flat detection system according to the first embodiment.

[0041] The horizontal axis of each graph in FIGS. 5 to 8 represents time, and the vertical axis represents frequency. FIGS. 5 to 8 show color maps in which frequency is represented by color. The independent component signal Sd1 calculated by the independent component decomposition unit 41A from the frequency spectrum signal Sc1 includes independent component signals Sd1a to Sd1d as signal elements. FIG. 5 shows the independent component signal Sd1a, which is the first independent component signal, and FIG. 6 shows the independent component signal Sd1b, which is the second independent component signal. FIG. 7 shows the independent component signal Sd1c, which is the third independent component signal, and FIG. 8 shows the independent component signal Sd1d, which is the fourth independent component signal. The independent component signal Sd1d shown in FIG. 8 is the same signal as the frequency spectrum signal Se1, which will be described later. The independent component signals Sd1a to Sd1d include non-periodic noise N1.

[0042] The independent component signals shown in Fig. 5 to Fig. 8 correspond to peaks P1 to P5 of the frequency spectrum signal Sc1 shown in Fig. 4. The independent component signal shown in Fig. 5 corresponds to peak P1, the independent component signal shown in Fig. 6 corresponds to peak P2, and the independent component signal shown in Fig. 7 corresponds to peak P3. That is, Figs. 5 to 7 correspond to peaks P1 to P3 of the joint vibration. Note that the independent component signal shown in Fig. 7 does not include reverberation noise P3E, which will be described later, of peak P3. The independent component signal shown in Fig. 8 is a component other than the joint vibration. The independent component signal shown in Fig. 8 includes peaks P4 and P5, which are flat vibrations, as well as reverberation noise P3E of peak P3, which was not fully resolved.

[0043] When the independent component remover 42A receives the extracted frequency spectrum signal Sc1 and independent component signal Sd1, it removes the corresponding signal element of the independent component signal Sd1 from the frequency spectrum signal Sc1 until the noise component is sufficiently removed from each signal element of the frequency spectrum signal Sc1. In this way, the independent component remover 42A calculates a frequency spectrum signal Se1 from which the noise component has been removed from the frequency spectrum signal Sc1. In this case, as described above, the independent component remover 42A determines that the noise component has been sufficiently removed when, after removing Q (Q is a natural number) noise components from the frequency spectrum signal Sc1, the AIC of the first frequency band (high frequency band) that does not contain flat oscillations in the removed signal is greater than the AIC obtained when (Q-1) noise components are removed from the frequency spectrum signal Sc1. That is, the independent component remover 42A removes signal elements (noise components) from the independent component signal Sd1 until the AIC of the first frequency band (high frequency band) is minimized. The independent component removing section 42A here determines that the noise components have been sufficiently removed when the peaks P1 to P3 of the joint vibration shown in FIGS. 5 to 7, which are noise components, have been removed.

[0044] The independent component removal unit 42A removes noise components from the independent component signal Sd1 in descending order of the norm in the first frequency band. That is, the independent component removal unit 42A removes noise components from the independent component signal Sd1 in descending order of signal strength in the first frequency band. As a result, when the strength of a time signal in the first frequency band of the independent component signal Sd1 is at a level that should be removed based on the AIC, the independent component removal unit 42A removes this time signal from the frequency spectrum signal Sc1 as a time signal of a noise component. In this way, the independent component removal unit 42A removes signals (noise components) of vibrations caused by rail joints from the frequency spectrum signal Sc, which is a time signal.

[0045] The independent component removal unit 42A here calculates the frequency spectrum signal Se1 by removing the independent component signals Sd1a to Sd1c from the frequency spectrum signal Sc1. The independent component removal unit 42A also removes noise components from the frequency spectrum signals Sc2 to ScN, as with the frequency spectrum signal Sc1. In this way, the independent component removal unit 42A calculates the frequency spectrum signal set Sen (frequency spectrum signals Se1, Se2, Se3, . . . , SeN) from which the noise components have been removed from the frequency spectrum signal set Scn. The independent component removal unit 42A outputs the frequency spectrum signal set Sen from which the noise components have been removed to the data combination unit 50.

[0046] In the first embodiment, the frequency spectrum signal group Sen is the second time signal, and the independent component signal group Sdn is the third time signal.

[0047] When the frequency spectrum signal group Sen from which noise components have been removed is input, the data combiner 50 calculates a frequency spectrum signal Se by combining frequency spectrum signals Se1, Se2, Se3, ..., SeN, which are signal elements of the frequency spectrum signal group Sen. Specifically, the data combiner 50 calculates a frequency spectrum signal Se from which noise components have been removed and which has the same time width as the frequency spectrum signal Sc before being extracted. In other words, the data combiner 50 combines the frequency spectrum signals Se1, Se2, Se3, ..., SeN so that the frequency spectrum signal group Sen from which noise components have been removed becomes a frequency spectrum signal Se which has the same time width as the frequency spectrum signal Sc before being extracted.

[0048] When combining data, the data combiner 50 uses the values ​​of each signal element of the frequency spectrum signal set Sen, and for overlapping times, i.e., times when the extraction windows overlap, combines the signal elements by using the minimum value of the overlapping signal elements. This allows the independent component remover 42A to remove the seam vibration from a signal, even if the independent component remover 42A fails to remove the seam vibration from another signal at the same time. The data combiner 50 outputs the frequency spectrum signal Se from which the noise component has been removed to the removal width expander 60.

[0049] When the frequency spectrum signal Se from which the noise components have been removed is input, the removal width expansion unit 60 calculates a frequency spectrum signal Sf from which the reverberation noise P3E has been removed by also removing the signal from a specific time after the time when each noise component of the frequency spectrum signal Se was removed. That is, the removal width expansion unit 60 removes the reverberation noise P3E by removing the signal from a specific time after the time when the peaks P1 to P3, which are noise components of the frequency spectrum signal Se, were removed (the signal immediately after removal). The removal width expansion unit 60 outputs the frequency spectrum signal Sf from which the reverberation noise P3E has been removed to the similarity filter unit 70.

[0050] 9 is a diagram illustrating an example of a frequency spectrum signal calculated by the removal width expansion unit of the railway wheel flat detection system according to the first embodiment. The horizontal axis of the graph illustrated in FIG. 9 represents time, and the vertical axis represents vibration frequency. FIG. 9 illustrates a color map in which frequency is represented by color.

[0051] The graph shown in Fig. 9 represents the frequency spectrum signal Sf from which the reverberation noise P3E has been removed, obtained when calculations are performed by the removal width expanding unit 60. Peaks P4 and P5 shown in Fig. 9 correspond to the peaks P4 and P5 shown in Fig. 8. As shown in Fig. 9, the frequency spectrum signal Sf obtained when calculations are performed by the removal width expanding unit 60 has not only the peaks P1 to P3, which are joint vibrations, but also the reverberation noise P3E of the peak P3 that could not be resolved.

[0052] When the similarity filter unit 70 receives the wheel detection sensor signal X2 sent from the wheel detection sensor 2, it calculates the speed of the vehicle 4 from the wheel detection sensor signal X2 and calculates the time it takes for the wheel to make one revolution from the speed of the vehicle 4.

[0053] Furthermore, when the frequency spectrum signal Sf from which the reverberation noise P3E has been removed is input, the similarity filter unit 70 applies a filter to the frequency spectrum signal Sf at each time such that similar signals between signals with a time difference of one revolution of the front and rear wheels are emphasized and the intensity is reduced (removed) when similar signals are not present. That is, the similarity filter unit 70 emphasizes signals at each time if they are similar to at least one signal from the time before or after one revolution of the wheel. In other words, if a signal at a specific time in the frequency spectrum signal Sf is similar to a signal from before one revolution of the wheel or a signal from after one revolution of the wheel, the similarity filter unit 70 increases the intensity of both signals, and decreases the intensity if the signals are not similar.

[0054] The time difference between the signal before one rotation of the wheel and the signal after one rotation of the wheel is the time difference for one rotation of the wheel. The similarity filter unit 70 determines whether the signals are similar based on the difference in signal strength for each frequency, the difference in the period during which the signals appear, and the like. Specifically, the similarity filter unit 70 calculates the cosine similarity for one rotation of the front and rear wheels. That is, for a signal at a certain time, the similarity filter unit 70 calculates the cosine similarity with the signal at a time one rotation before the wheel and the cosine similarity with the signal at a time one rotation after the wheel, and multiplies the result by the maximum value. That is, for the signal at the Rth rotation of the wheel (R is a natural number), the similarity filter unit 70 calculates the cosine similarity with the signal at the (R-1)th rotation and the cosine similarity with the signal at the (R+1)th rotation, and multiplies the signal at the Rth rotation by the maximum value of these.

[0055] Cosine similarity is expressed as a value between 0 and 1. When signals for one rotation of the wheel are similar (when the similarity is equal to or greater than a specific value), the similarity filter unit 70 performs filtering by multiplying these signals by "1", which is the maximum value of cosine similarity. As a result, when there is a pair of a signal before one rotation of the wheel and a signal after one rotation of the wheel, the pair of signals is emphasized and remains. This filtering may also be performed by multiplying by cosine similarity.

[0056] In this way, the similarity filter unit 70 strengthens or weakens the signal strength based on the cosine similarity between a signal at a certain time and a signal at a time one revolution before the wheel and the cosine similarity between a signal at a time one revolution after the wheel, thereby making it possible to calculate a frequency spectrum signal Sg from which the non-periodic noise N1 that was not removed by the processing up to the removal width expansion unit 60 has been removed. In other words, the similarity filter unit 70 can calculate a frequency spectrum signal Sg from which the non-periodic noise N1, which is a signal other than the wheel rotation period, has been removed. The similarity filter unit 70 outputs the frequency spectrum signal Sg from which the non-periodic noise N1 has been removed to the flat detection unit 80A.

[0057] The similarity filter unit 70 also calculates a wheel detection output (wheel passing time data) indicating the time when the wheel passed the arrangement position of the wheel detection sensor 2 based on the wheel detection sensor signal X2, and outputs the calculated output to the output unit 90.

[0058] Fig. 10 is a diagram illustrating an example of a frequency spectrum signal calculated by the similarity filter unit of the railway wheel flat detection system according to the first embodiment. The horizontal axis of the graph illustrated in Fig. 10 represents time, and the vertical axis represents vibration frequency. Fig. 10 also illustrates a color map in which frequency is represented by color.

[0059] The graph shown in Fig. 10 represents the frequency spectrum signal Sg from which the non-periodic noise N1 has been removed, which is obtained when the similarity filter unit 70 performs the calculation. Peaks P4 and P5 shown in Fig. 10 correspond to the peaks P4 and P5 shown in Fig. 9. As shown in Fig. 10, the frequency spectrum signal Sg from which the similarity filter unit 70 performs the calculation has also had the non-periodic noise N1 removed.

[0060] When the frequency spectrum signal Sg from which the non-periodic noise N1 has been removed is input, the flat detection unit 80A calculates, for each time instant of the frequency spectrum signal Sg, the sum of the absolute values ​​of the signals in the second frequency band (a low frequency band having a frequency lower than the first frequency band) divided by the sum of the absolute values ​​of the signals in frequency bands other than the second frequency band, thereby obtaining a time signal Sh that indicates the presence or absence of a wheel flat at each time instant.

[0061] The flat detection unit 80A determines whether a wheel flat has been detected by comparing the time signal Sh with a preset threshold value. The flat detection unit 80A determines that a wheel flat has occurred if the time signal Sh is greater than the threshold value.

[0062] 11 is a diagram illustrating an example of a time signal calculated by the flat detection unit of the railway wheel flat detection system according to the embodiment 1. The horizontal axis of the graph illustrated in FIG. 11 represents time, and the vertical axis represents signal intensity.

[0063] The graph shown in Fig. 11 represents the time signal Sh, which indicates the presence or absence of a wheel flat, obtained when the flat detection unit 80A performs calculations on the frequency spectrum signal Sg from which the non-periodic noise N1 has been removed. Peaks P4 and P5 shown in Fig. 11 correspond to peaks P4 and P5 shown in Fig. 10.

[0064] The flat detection unit 80A detects the peaks P4 and P5, which are wheel flats, by comparing the time signal Sh of the peaks P4 and P5 with a threshold. The flat detection unit 80A outputs the flat detection judgment result to the output unit 90. The output unit 90 outputs the flat detection judgment result sent from the flat detection unit 80A and the wheel detection output sent from the similarity filter unit 70 to the display unit 91. The display unit 91 displays the flat detection judgment result and the wheel detection output.

[0065] 12 is a diagram showing an example of wheel detection output and flat detection determination result displayed by the display unit of the railway wheel flat detection system according to the first embodiment. The horizontal axis of the graph shown in the upper part of Fig. 12 represents time, and the vertical axis represents wheel detection output. The horizontal axis of the graph shown in the lower part of Fig. 12 represents time, and the vertical axis represents flat detection determination result.

[0066] In the graph shown in the upper part of Fig. 12, a waveform appears when a wheel passes, and in the graph shown in the lower part of Fig. 12, a waveform appears when a flat is detected. This makes it easy to identify the wheel on which a wheel flat has occurred.

[0067] Next, an operation processing procedure of the railway wheel flat detection system 1A will be described. Fig. 13 is a flowchart showing the operation processing procedure of the railway wheel flat detection system according to the first embodiment. Fig. 13 shows the wheel flat detection processing procedure by the railway wheel flat detection system 1A.

[0068] The frequency analysis unit 10A receives the vibration sensor signal X1 sent from the vibration sensor 3. The frequency analysis unit 10A calculates a frequency spectrum signal Sb from the vibration sensor signal X1 (step S10). Specifically, the frequency analysis unit 10A performs a short-time Fourier transform on the vibration sensor signal X1 to calculate a frequency spectrum signal Sb representing the frequency spectrum at each time. The frequency analysis unit 10A outputs the frequency spectrum signal Sb to the normalization unit 20.

[0069] The normalization unit 20 calculates a frequency spectrum signal Sc in which the scale of noise has been normalized from the frequency spectrum signal Sb (step S20). Specifically, the normalization unit 20 calculates the scale of noise other than the joint vibration, and divides the frequency spectrum signal Sb by the calculated scale to calculate the frequency spectrum signal Sc in which the scale of noise has been normalized. The normalization unit 20 outputs the normalized frequency spectrum signal Sc to the data extraction unit 30.

[0070] The data extracting unit 30 calculates a set of frequency spectrum signals Scn (n=1, 2, 3, . . . , N) by extracting the frequency spectrum signal Sc for each specific time duration (step S30). The data extracting unit 30 outputs the extracted set of frequency spectrum signals Scn to the independent component decomposition unit 41A.

[0071] The independent component decomposition unit 41A calculates an independent component signal group Sdn by decomposing each signal element of the frequency spectrum signal group Scn into independent components by independent component analysis (step S41). The independent component decomposition unit 41A outputs the extracted frequency spectrum signal group Scn and independent component signal group Sdn to the independent component removal unit 42A.

[0072] The independent component remover 42A removes the signal elements of the independent component signal group Sdn from the frequency spectrum signal group Scn until the noise components are sufficiently removed from each signal element of the frequency spectrum signal group Scn. The independent component remover 42A then calculates the frequency spectrum signal group Sen from which the noise components of the independent component signal group Sdn have been removed (step S42). The independent component remover 42A outputs the frequency spectrum signal group Sen from which the noise components have been removed to the data combiner 50.

[0073] The data combining unit 50 calculates, from the group of frequency spectrum signals Sen from which the noise components have been removed, a frequency spectrum signal Se from which the noise components have been removed, which has the same time width as the frequency spectrum signal Sc before being extracted (step S50).The data combining unit 50 outputs the frequency spectrum signal Se from which the noise components have been removed to the removal width expanding unit 60.

[0074] The elimination width expansion unit 60 calculates a frequency spectrum signal Sf from the frequency spectrum signal Se, from which the reverberation noise P3E has been removed (step S60). Specifically, the elimination width expansion unit 60 calculates the frequency spectrum signal Sf from which the reverberation noise P3E has been removed by removing the signal from the frequency spectrum signal Se up to a specific time after the time at which each noise component has been removed. The elimination width expansion unit 60 outputs the frequency spectrum signal Sf from which the reverberation noise P3E has been removed to the similarity filter unit 70.

[0075] The similarity filter unit 70 receives the wheel detection sensor signal X2 sent from the wheel detection sensor 2. The similarity filter unit 70 calculates the speed of the vehicle 4 from the wheel detection sensor signal X2, and calculates the time it takes for the wheel to make one rotation from the speed of the vehicle 4.

[0076] The similarity filter unit 70 calculates the frequency spectrum signal Sg from which the non-periodic noise N1 has been removed, based on the time it takes for the wheel to make one revolution and the frequency spectrum signal Sf (step S70). Specifically, the similarity filter unit 70 applies a filter to the frequency spectrum signal Sf so that if there is a signal similar to the signal at a time with a time difference of one revolution of the front and rear wheels for each time, the filter emphasizes the signal, and if there is no similar signal, the filter reduces the signal strength.

[0077] The similarity filter unit 70 outputs the frequency spectrum signal Sg from which the non-periodic noise N1 has been removed to the flat detection unit 80A. Furthermore, the similarity filter unit 70 calculates a wheel detection output indicating the time when the wheel passed the arrangement position of the wheel detection sensor 2 based on the wheel detection sensor signal X2, and outputs the wheel detection output to the output unit 90.

[0078] The flat detection unit 80A calculates a time signal Sh from the frequency spectrum signal Sg, which indicates whether or not a wheel flat has occurred at each time, and determines whether or not a wheel flat has occurred based on the time signal Sh (Step S80). Specifically, the flat detection unit 80A obtains the time signal Sh, which indicates whether or not a wheel flat has occurred at each time, by dividing the sum of the absolute values ​​of the signals in the second frequency band by the sum of the absolute values ​​of the signals in frequency bands other than the second frequency band, for each time of the frequency spectrum signal Sg. The flat detection unit 80A then determines whether or not a wheel flat has occurred by comparing the time signal Sh with a preset threshold. The flat detection unit 80A determines whether or not a wheel flat has occurred if the time signal Sh is greater than the threshold. The flat detection unit 80A outputs the flat detection determination result to the output unit 90. The output unit 90 outputs the flat detection determination result and wheel detection output to the display unit 91. The display unit 91 displays the flat detection determination result and wheel detection output (Step S90).

[0079] As described above, in the first embodiment, the railway wheel flat detection system 1A detects a wheel flat based on the vibration sensor signal X1 sent from the vibration sensor 3 installed on the rail 5. In this case, the railway wheel flat detection system 1A can record flat vibration only at the time when the wheel on which the wheel flat occurred passes near the installation position of the vibration sensor 3. Even in this case, the railway wheel flat detection system 1A is equipped with an independent component decomposition unit 41A and an independent component removal unit 42A, and therefore can detect vibration caused by a wheel flat while preventing erroneous detection due to noise caused by rail joints, etc.

[0080] That is, even if the number of flat vibrations contained in the data of the vibration sensor signal X1 is small, the railway wheel flat detection system 1A can accurately detect wheel flats because the independent component decomposition unit 41A calculates independent component signals for the normalized frequency spectrum signal Sc and the independent component removal unit 42A removes noise based on the independent component signals. In this way, the railway wheel flat detection system 1A does not need to suppress false detections based on the agreement between the number of shock waves and the wheel rotation speed at a specific time, and can accurately detect wheel flats because the independent component decomposition unit 41A and the independent component removal unit 42A suppress false detections.

[0081] As described above, according to the first embodiment, the railway wheel flat detection system 1A detects a wheel flat based on the vibration sensor signal X1 sent from the vibration sensor 3 installed on the rail 5, which makes it possible to reduce the number of vibration sensors 3 required to detect a wheel flat. In other words, if the vibration sensors 3 are installed on the vehicle 4, the same number of vibration sensors 3 as the number of wheels for which wheel flat detection is desired is required, but the railway wheel flat detection system 1A of the first embodiment makes it possible to reduce the number of vibration sensors 3 installed.

[0082] In addition, the railway wheel flat detection system 1A can accurately detect wheel flats because the independent component decomposition unit 41A calculates independent component signals for the normalized frequency spectrum signal Sc, and the independent component removal unit 42A removes noise based on the independent component signals.

[0083] Therefore, the railway wheel flat detection system 1A can accurately detect wheel flats at low cost even when detecting wheel flats on a large number of wheels.

[0084] Embodiment 2 Next, a second embodiment will be described with reference to FIG. 14. In the second embodiment, another form of the independent component decomposition unit 41A will be described. In the first embodiment, the independent component decomposition unit 41A decomposed the signal using independent component analysis, but in the second embodiment, the independent component decomposition unit 41A performs threshold processing on the norm of the signal at each time to determine whether or not it is a vibration, and decomposes the vibration signals that are occurring independently. Note that, among the components included in the railway wheel flat detection system 1A, the components other than the independent component decomposition unit 41A perform the same processing as in the first embodiment, and therefore their description will be omitted.

[0085] 14 is a diagram illustrating an example of a norm calculated by the independent component decomposition unit of the railway wheel flat detection system according to the second embodiment. The horizontal axis of the graph illustrated in FIG. 14 represents time, and the vertical axis represents the norm.

[0086] 14 shows the norms obtained when the independent component decomposition unit 41A calculates the signal elements of the normalized frequency spectrum signal group Scn. The independent component decomposition unit 41A calculates the norm at each time by calculating the sum of the absolute values ​​of the signal intensity for each frequency. The independent component decomposition unit 41A decomposes the frequency spectrum signal group Scn into independent component signals by performing threshold processing on each signal element for which the norm has been calculated. For example, the independent component decomposition unit 41A decomposes the frequency spectrum signal Sc1 into the independent component signals Sd1a to Sd1d shown in FIGS. 5 to 8 by extracting signal elements for which the norm has been calculated that are greater than a threshold.

[0087] Specifically, when the extracted frequency spectrum signal group Scn is input, the independent component decomposition unit 41A of the second embodiment calculates the independent component signal group Sdn by extracting only the time when the norm of all frequencies for each signal element (frequency spectrum signals Sc1, Sc2, Sc3, . . . , ScN) exceeds a first threshold value Th1. The independent component decomposition unit 41A outputs the extracted frequency spectrum signal group Scn and the independent component signal group Sdn to the independent component removal unit 42A.

[0088] In this way, in the second embodiment, the independent component decomposition unit 41A determines whether or not a vibration occurs by thresholding the norm of the frequency spectrum signal group Scn at each time, and decomposes the independently occurring vibration signals into independent component signals. This enables the railway wheel flat detection system 1A to accurately detect wheel flats using a small number of vibration sensors 3, as in the first embodiment.

[0089] Embodiment 3 Next, a third embodiment will be described with reference to Figs. 15 to 19. In the third embodiment, another form of the independent component removal unit 42A will be described. In the first embodiment, the independent component removal unit 42A removed components other than flat vibration using AIC, but in the third embodiment, the independent component removal unit 42A removes components other than flat vibration using the difference in distribution of the frequency spectrum between the wheel flat vibration signal and vibration signals other than this signal. Note that, of the components included in the railway wheel flat detection system 1A, the components other than the independent component removal unit 42A perform the same processing as in the first embodiment, and therefore description thereof will be omitted.

[0090] 15 is a diagram illustrating an example of a first frequency set by the independent component remover of the railway wheel flat detection system according to the third embodiment when calculating signal elements of a frequency spectrum signal group. The horizontal axis of the graph illustrated in FIG. 15 represents time, and the vertical axis represents frequency.

[0091] As explained in Figure 4 of embodiment 1, the frequency spectrum of the wheel flat vibration signal is distributed in the low frequency band, whereas the frequency spectrum of vibration signals other than this vibration signal is widely distributed from the low frequency band to the high frequency band.

[0092] The independent component remover 42A receives the extracted frequency spectrum signal group Scn and independent component signal group Sdn from the independent component decomposition unit 41A. The independent component remover 42A calculates a frequency spectrum signal group Sen by removing noise components from the frequency spectrum signal group Scn based on a first frequency band F1 defined by a preset threshold. The first frequency band F1 is the entire frequency range (frequency region) equal to or greater than a preset threshold. This threshold is set in advance by the user to be a value greater than the frequencies at which wheel flat vibration signals are distributed.

[0093] The independent component remover 42A selects a high frequency band that does not include peaks P4 and P5, which are vibration signals of the wheel flat, as the first frequency band F1 based on a preset threshold value. The independent component remover 42A calculates the norm in the first frequency band F1 of the signal elements of the independent component signal group Sdn corresponding to each signal element of the frequency spectrum signal group Scn.

[0094] Here, the norm calculated by the independent component remover 42A based on the frequency spectrum signal group Scn and the independent component signal group Sdn will be described. Here, the case where the independent component remover 42A calculates the norm in the first frequency band F1 of the signal elements of the independent component signal Sd1 corresponding to the frequency spectrum signal Sc1 shown in FIG.

[0095] Fig. 16 is a diagram illustrating an example of the norm of a first peak calculated by the independent component removal unit of the railway wheel flat detection system according to the third embodiment. Fig. 17 is a diagram illustrating an example of the norm of a second peak calculated by the independent component removal unit of the railway wheel flat detection system according to the third embodiment. Fig. 18 is a diagram illustrating an example of the norm of a third peak calculated by the independent component removal unit of the railway wheel flat detection system according to the third embodiment. Fig. 19 is a diagram illustrating an example of the norm of reverberation noise of the third peak calculated by the independent component removal unit of the railway wheel flat detection system according to the third embodiment.

[0096] The horizontal axis of each graph in Figures 16 to 19 represents time, and the vertical axis represents norm. Independent component remover 42A calculates the norms shown in Figures 16 to 19 for independent component signals Sd1a, Sd1b, Sd1c, and Sd1d shown in Figures 5 to 8. In Figures 16 to 18, the first peak is peak P1, the second peak is peak P2, and the third peak is peak P3.

[0097] The norms of peaks P1 to P3 shown in Figures 16 to 19 correspond to peaks P1 to P3 shown in Figure 15. The norm shown in Figure 16 is the norm of peak P1 in the first frequency band F1, the norm shown in Figure 17 is the norm of peak P2 in the first frequency band F1, and the norm shown in Figure 18 is the norm of peak P3 in the first frequency band F1. That is, Figures 16 to 18 correspond to the norms of joint vibration peaks P1 to P3 in the first frequency band F1. Note that the norm in the first frequency band F1 shown in Figure 18 does not include the norm of reverberation noise P3E of peak P3. The norm in the first frequency band F1 shown in Figure 19 is the norm of reverberation noise P3E of peak P3.

[0098] The independent component removal unit 42A extracts only the peaks P1 to P3 by performing threshold processing on the norms shown in Figures 16 to 19. Specifically, the independent component removal unit 42A calculates the frequency spectrum signal Se1 shown in Figure 8 by removing frequency spectrum signals having a norm greater than a second threshold Th2 from the frequency spectrum signal Sc1. The threshold Th2 is a value smaller than the norm of the peaks P1 to P3 and greater than the norm of the reverberation noise P3E.

[0099] In this way, the independent component remover 42A removes components whose norm in the first frequency band F1 exceeds the threshold value Th2 from the signal elements of the independent component signal group Sdn corresponding to each signal element of the frequency spectrum signal group Scn, thereby obtaining the frequency spectrum signal group Sen from which the noise components have been removed. The independent component remover 42A outputs the frequency spectrum signal group Sen from which the noise components have been removed to the data combiner 50.

[0100] As described above, in the third embodiment, the independent component remover 42A removes from the frequency spectrum signal Sc1 signals whose norm in the first frequency band F1 is greater than the threshold value Th2. As a result, the railway wheel flat detection system 1A of the third embodiment can obtain the frequency spectrum signal Se1 from which the noise components shown in Fig. 8 have been removed, similar to the railway wheel flat detection system 1A of the first embodiment. Therefore, the railway wheel flat detection system 1A can accurately detect a wheel flat using a small number of vibration sensors 3, similar to the first embodiment.

[0101] Embodiment 4 Next, a fourth embodiment will be described with reference to Figures 20 to 27. In the fourth embodiment, a railway wheel flat detection system calculates a time signal limited to a first frequency band F1 that can detect vibrations caused by a rail joint, and a time signal limited to a second frequency band F2 that can detect both vibrations caused by the rail joint and vibrations caused by a wheel flat. The railway wheel flat detection system then calculates a time signal from which noise components have been removed for the second frequency band F2 from the time signal limited to the first frequency band F1 and the time signal limited to the second frequency band F2, and determines whether a flat has been detected by comparing this time signal with a threshold value.

[0102] Fig. 20 is a diagram showing the configuration of a railway wheel flat detection system according to embodiment 4. Among the components in Fig. 20, components that achieve the same functions as those in the railway wheel flat detection system 1A according to embodiment 1 shown in Fig. 1 are assigned the same reference numerals, and duplicated explanations will be omitted.

[0103] The railway wheel flat detection system 1B includes a frequency analysis unit 10B, a seam vibration elimination unit 40B, a elimination width expansion unit 60, a similarity filter unit 70, a flat detection unit 80B, an output unit 90, and a display unit 91. The display unit 91 and the railway wheel flat detection system 1B may be configured separately.

[0104] The joint vibration elimination unit 40B has an independent component decomposition unit 41B and an independent component elimination unit 42B. The frequency analysis unit 10B is connected to the vibration sensor 3 and receives the vibration sensor signal X1 sent from the vibration sensor 3.

[0105] The frequency analysis unit 10B calculates a time signal Sbb1 limited to a first frequency band F1 that can detect vibrations caused by rail joints, based on the vibration sensor signal X1. The frequency analysis unit 10B also calculates a time signal Sbb2 limited to a second frequency band F2 that can detect both vibrations caused by rail joints and vibrations caused by wheel flats, based on the vibration sensor signal X1. The first frequency band F1 is a high-frequency band that includes the frequencies of vibrations caused by rail joints. The second frequency band F2 is a low-frequency band lower than the first frequency band F1 that includes the frequencies of vibrations caused by rail joints and vibrations caused by wheel flats. In the fourth embodiment, the time signals Sbb1 and Sbb2 are first time signals.

[0106] The frequency analysis unit 10B applies a high-pass filter (HPF) or a band-pass filter (BPF) to the vibration sensor signal X1 to calculate a time signal Sbb1 limited to a first frequency band F1.

[0107] Furthermore, the frequency analysis unit 10B calculates a time signal Sbb2 limited to a second frequency band F2 by applying a low-pass filter (LPF) or a band-pass filter to the vibration sensor signal X1.

[0108] The frequency analysis unit 10B outputs the time signal Sbb1 limited to the first frequency band F1 and the time signal Sbb2 limited to the second frequency band F2 to the independent component decomposition unit 41B.

[0109] Fig. 21 is a diagram illustrating an example of a time signal limited to a first frequency band calculated by a frequency analysis unit of the railway wheel flat detection system according to the fourth embodiment. Fig. 22 is a diagram illustrating an example of a time signal limited to a second frequency band calculated by a frequency analysis unit of the railway wheel flat detection system according to the fourth embodiment.

[0110] Fig. 21 shows the time signal Sbb1 calculated by the frequency analysis unit 10B, and Fig. 22 shows the time signal Sbb2 calculated by the frequency analysis unit 10B. The horizontal axis of each graph in Fig. 21 and Fig. 22 represents time, and the vertical axis represents the amplitude of vibration.

[0111] The time signal Sbb1 shown in FIG. 21 includes components of the first frequency band F1 of peaks P1 to P3, which are joint vibrations, and the time signal Sbb2 shown in FIG. 22 includes components of the second frequency band F2 of peaks P1 to P3 in addition to peaks P4 and P5, which are flat vibrations.

[0112] In Figure 21, the component of the first frequency band F1 of peak P1 is indicated by the first frequency component FH1, the component of the first frequency band F1 of peak P2 is indicated by the first frequency component FH2, and the component of the first frequency band F1 of peak P3 is indicated by the first frequency component FH3.

[0113] 22, the component of the second frequency band F2 of peak P1 is indicated by second frequency component FL1, the component of the second frequency band F2 of peak P2 is indicated by second frequency component FL2, and the component of the second frequency band F2 of peak P3 is indicated by second frequency component FL3. Also, in Fig. 22, the component of the second frequency band F2 of peak P4 is indicated by second frequency component FL4, and the component of the second frequency band F2 of peak P5 is indicated by second frequency component FL5.

[0114] When a time signal Sbb1 limited to the first frequency band F1 is input, the independent component decomposition unit 41B calculates an independent component signal Sdd1 by extracting only the times at which the amplitude (signal strength) exceeds a specific threshold (specific value) from the time signal Sbb1. When a time signal Sbb2 limited to the second frequency band F2 is input, the independent component decomposition unit 41B calculates an independent component signal Sdd2 by extracting only the signals from the time signal Sbb2 that are the same as the times extracted when the independent component signal Sdd1 was calculated.

[0115] Here, the processing executed by the independent component decomposition unit 41B will be described. Fig. 23 is a diagram for explaining an example of thresholds set by the independent component decomposition unit of the railway wheel flat detection system according to the fourth embodiment. Fig. 24 is a diagram showing an example of independent component signals extracted from a time signal limited to a first frequency band by the independent component decomposition unit of the railway wheel flat detection system according to the fourth embodiment. Fig. 25 is a diagram showing an example of independent component signals extracted from a time signal limited to a second frequency band by the independent component decomposition unit of the railway wheel flat detection system according to the fourth embodiment.

[0116] The horizontal axis of each graph in Fig. 23 to Fig. 25 represents time, and the vertical axis represents vibration amplitude. The graph shown in Fig. 23 corresponds to the graph shown in Fig. 21. Fig. 23 shows the time signal Sbb1 calculated by the frequency analysis unit 10B and the thresholds Th3 and Th4 used by the independent component decomposition unit 41B. The graph shown in Fig. 24 corresponds to the graph shown in Fig. 23, and the graph shown in Fig. 25 corresponds to the graph shown in Fig. 22.

[0117] The threshold value Th3 is a positive value, and the threshold value Th4 is a negative value. The threshold values ​​Th3 and Th4 have the same absolute value but different signs, positive and negative.

[0118] The independent component decomposition unit 41B compares the time signal Sbb1 limited to the first frequency band F1 with thresholds Th3 and Th4, and extracts an independent component signal Sdd1, which is a time signal greater than the threshold Th3 or less than the threshold Th4, from the time signal Sbb1. The independent component decomposition unit 41B extracts a time signal of a specific time width from the time signal Sbb1 greater than the threshold Th3 or less than the threshold Th4.

[0119] FIG. 24 shows the first frequency component FH1 of peak P1, the first frequency component FH2 of peak P2, and the first frequency component FH3 of peak P3 as the independent component signal Sdd1 extracted by the independent component decomposition section 41B.

[0120] The independent component decomposition unit 41B extracts time signals having the same times as the first frequency components FH1 to FH3 from the time signal Sbb2 limited to the second frequency band F2. In Fig. 25, the independent component signal Sdd2 extracted by the independent component decomposition unit 41B includes the second frequency component FL1 of peak P1, the second frequency component FL2 of peak P2, and the second frequency component FL3 of peak P3. In the fourth embodiment, the independent component signal Sdd1 is the first limited signal, and the independent component signal Sdd2 is the second limited signal.

[0121] The independent component decomposition unit 41B outputs a time signal Sbb1 limited to the first frequency band F1, a time signal Sbb2 limited to the second frequency band F2, and the independent component signals Sdd1 and Sdd2 to the independent component removal unit 42B.

[0122] When the independent component remover 42B receives a time signal Sbb1 limited to the first frequency band F1, a time signal Sbb2 limited to the second frequency band F2, and the independent component signals Sdd1 and Sdd2, it removes signal elements from the independent component signal Sdd1 until the noise components are sufficiently removed from the time signal Sbb1. The independent component remover 42B then calculates a time signal See1 from which the noise components have been removed for the first frequency band F1. The noise components are sufficiently removed when, after L (L is a natural number) noise components have been removed from the time signal Sbb1, the AIC of the removed time signal is greater than the AIC when (L-1) noise components have been removed from the time signal Sbb1.

[0123] Similar to the independent component remover 42A, the independent component remover 42B removes noise components from the independent component signal Sdd1 in descending order of the norm in the first frequency band F1.

[0124] The independent component remover 42B removes, from the time signal Sbb2, an independent component signal Sdd2 corresponding to the time of the independent component signal Sdd1 removed from the time signal Sbb1, thereby calculating a time signal See2 from which the noise component has been removed for the second frequency band F2. In the fourth embodiment, the time signal See2 is the second time signal.

[0125] Although the method in which the independent component remover 42B uses AIC as a method for removing noise components has been described here, the independent component remover 42B may also use a method in which all components of the independent component signal Sdd2 from the time signal Sbb2 correspond to the time points at which the independent component signal Sdd1 exceeds a specific threshold. The independent component remover 42B outputs the time signal See2 from which the noise components have been removed for the second frequency band F2 to the removal width expander 60.

[0126] Fig. 26 is a diagram showing an example of a time signal obtained by removing noise components from a time signal limited to the second frequency band by the independent component remover of the railway wheel flat detection system according to the fourth embodiment. The horizontal axis of the graph in Fig. 26 represents time, and the vertical axis represents vibration amplitude. The graph shown in Fig. 26 corresponds to the graphs shown in Figs. 21 to 25.

[0127] 26 shows the time signal See2 from which the noise component has been removed, which is obtained when the independent component remover 42B performs arithmetic processing on the time signal Sbb2 and the independent component signals Sdd1 and Sdd2. In the time signal See2, the time signals of peaks P1 to P3, which are joint vibrations, have been removed.

[0128] When the time signal See2 from which the noise components for the second frequency band F2 have been removed is input, the removal width expansion unit 60 calculates a time signal Sff (not shown) from which the reverberation noise P3E has been removed by also removing the signal from the time signal See2 a specific time after the time at which each noise component was removed. The removal width expansion unit 60 outputs the time signal Sff from which the reverberation noise P3E has been removed to the similarity filter unit 70.

[0129] When the similarity filter unit 70 receives the wheel detection sensor signal X2 sent from the wheel detection sensor 2, it calculates the speed of the vehicle 4 from the wheel detection sensor signal X2 and calculates the time it takes for the wheel to make one revolution from the speed of the vehicle 4.

[0130] Furthermore, when the time signal Sff from which the reverberation noise P3E has been removed is input, the similarity filter unit 70 applies a filter to the time signal Sff such that if there are similar signals between signals with a time difference of one revolution of the front and rear wheels, the filter emphasizes the signals and if there are no similar signals, the filter reduces (removes) the intensity of the signals. That is, the similarity filter unit 70 increases the intensity of the signals before one revolution of the wheel and after one revolution of the wheel if they are similar, and reduces the intensity of the signals if they are not similar. The similarity filter unit 70 of the fourth embodiment, like the similarity filter unit 70 of the first embodiment, emphasizes similar signals based on the cosine similarity of one revolution of the front and rear wheels.

[0131] In this way, the similarity filter unit 70 strengthens or weakens the signal strength based on whether the signal before the wheel makes one rotation and the signal after the wheel makes one rotation are similar, and can therefore calculate a time signal Sgg (not shown) from which the non-periodic noise N1 that could not be removed by the processing up to the removal width expansion unit 60 has been removed. The similarity filter unit 70 outputs the time signal Sgg from which the non-periodic noise N1 has been removed to the flat detection unit 80B. Furthermore, the similarity filter unit 70 calculates a wheel detection output that indicates the time when the wheel passed the arrangement position of the wheel detection sensor 2 based on the wheel detection sensor signal X2, and outputs the wheel detection output to the output unit 90.

[0132] When the time signal Sgg from which the non-periodic noise N1 has been removed is input, the flat detection unit 80B compares the time signal Sgg at each time with a preset threshold to determine whether a flat has been detected. The flat detection unit 80B determines that a wheel flat has occurred if the time signal Sgg is greater than the threshold. The flat detection unit 80B outputs the flat detection determination result to the output unit 90. The output unit 90 outputs the flat detection determination result sent from the flat detection unit 80B and the wheel detection output sent from the similarity filter unit 70 to the display unit 91. The display unit 91 displays the flat detection determination result and the wheel detection output.

[0133] Next, an operation processing procedure of the railway wheel flat detection system 1B will be described. Fig. 27 is a flowchart showing the operation processing procedure of the railway wheel flat detection system according to the fourth embodiment. Fig. 27 shows the wheel flat detection processing procedure by the railway wheel flat detection system 1B.

[0134] The frequency analysis unit 10B receives the vibration sensor signal X1 sent from the vibration sensor 3. The frequency analysis unit 10B calculates, from the vibration sensor signal X1, a time signal Sbb1 limited to a first frequency band F1 and a time signal Sbb2 limited to a second frequency band F2 (step S110). Specifically, the frequency analysis unit 10B calculates the time signal Sbb1 by applying a high-pass filter or a band-pass filter to the vibration sensor signal X1. The frequency analysis unit 10B also calculates the time signal Sbb2 by applying a low-pass filter or a band-pass filter to the vibration sensor signal X1. The frequency analysis unit 10B outputs the time signals Sbb1 and Sbb2 to the independent component decomposition unit 41B.

[0135] The independent component decomposition unit 41B extracts an independent component signal Sdd1 from the time signal Sbb1 limited to the first frequency band F1, and extracts an independent component signal Sdd2 from the time signal Sbb2 limited to the second frequency band F2 (step S141).The independent component decomposition unit 41B outputs the time signal Sbb1 limited to the first frequency band F1, the time signal Sbb2 limited to the second frequency band F2, and the independent component signals Sdd1 and Sdd2 to the independent component removal unit 42B.

[0136] The independent component remover 42B calculates a time signal See2 from which the noise component has been removed for the second frequency band F2, based on the time signal Sbb1 limited to the first frequency band F1, the time signal Sbb2 limited to the second frequency band F2, and the independent component signals Sdd1 and Sdd2 (step S142).The independent component remover 42B outputs the time signal See2 to the removal width expander 60.

[0137] The removal width expansion unit 60 calculates a time signal Sff by removing the reverberation noise P3E from the time signal See2 (step S160). The removal width expansion unit 60 outputs the time signal Sff to the similarity filter unit .

[0138] The similarity filter unit 70 calculates a time signal Sgg from which the non-periodic noise N1 has been removed, based on the time required for the wheel to make one rotation and the time signal Sff (step S170). The similarity filter unit 70 outputs the time signal Sgg from which the non-periodic noise N1 has been removed to the flat detection unit 80B. The similarity filter unit 70 also calculates a wheel detection output indicating the time at which the wheel passed the arrangement position of the wheel detection sensor 2, based on the wheel detection sensor signal X2, and outputs the wheel detection output to the output unit 90.

[0139] The flat detection unit 80B compares the time signal Sgg at each time with a preset threshold value to determine whether a flat has been detected (step S180). Specifically, the flat detection unit 80B determines that a wheel flat has occurred when the time signal Sgg is greater than the threshold value. The flat detection unit 80B outputs the flat detection determination result to the output unit 90.

[0140] The output unit 90 outputs the flat detection determination result and the wheel detection output to the display unit 91. The display unit 91 displays the flat detection determination result and the wheel detection output (step S190).

[0141] Next, the hardware configuration of the railway wheel flat detection systems 1A and 1B will be described. The railway wheel flat detection systems 1A and 1B are realized by a processing circuit. This processing circuit may be a processor and memory that executes a program stored in memory, or may be dedicated hardware. The processing circuit is also called a control circuit.

[0142] 28 is a diagram showing an example of the configuration of a processing circuit when the processing circuit included in the railway wheel flat detection systems according to the first and second embodiments is realized by a processor and a memory. Note that the railway wheel flat detection systems 1A and 1B have similar hardware configurations, and therefore, the hardware configuration of the railway wheel flat detection system 1A will be described below.

[0143] The processing circuit 150 shown in FIG. 28 is a control circuit and includes a processor 151 and a memory 152. When the processing circuit 150 is configured with the processor 151 and the memory 152, each function of the processing circuit 150 is realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory 152. The processor 151 reads and executes the program stored in the memory 152 to realize each function of the processing circuit 150. That is, the processing circuit 150 includes the memory 152 for storing a railway wheel flat detection program that results in the processing of the railway wheel flat detection system 1A. This railway wheel flat detection program can also be said to be a program that causes the railway wheel flat detection system 1A to execute each function realized by the processing circuit 150. This railway wheel flat detection program may be provided by a storage medium on which the program is stored, or by other means such as a communication medium.

[0144] The railway wheel flat detection program executed by the railway wheel flat detection system 1A has a modular configuration including a frequency analysis unit 10A, a normalization unit 20, a data extraction unit 30, a joint vibration removal unit 40A, a data combination unit 50, a removal width expansion unit 60, a similarity filter unit 70, a flat detection unit 80A, and an output unit 90, which are loaded onto the main memory device and generated on the main memory device.

[0145] The railway wheel flat detection program executed by the railway wheel flat detection system 1B has a modular configuration including a frequency analysis unit 10B, a joint vibration removal unit 40B, a removal width expansion unit 60, a similarity filter unit 70, a flat detection unit 80B, and an output unit 90, which are loaded onto the main memory device and generated on the main memory device.

[0146] Here, the processor 151 is, for example, a CPU (Central Processing Unit), a processing device, an arithmetic device, a microprocessor, a microcomputer, or a DSP (Digital Signal Processor), etc. Furthermore, the memory 152 is, for example, a non-volatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable ROM), or an EEPROM (registered trademark) (Electrically EPROM), a magnetic disk, a flexible disk, an optical disk, a compact disk, a minidisk, or a DVD (Digital Versatile Disc).

[0147] 29 is a diagram illustrating an example of a processing circuit when the processing circuit included in the railway wheel flat detection system according to the embodiment is configured with dedicated hardware. The processing circuit 153 illustrated in FIG. 29 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof.

[0148] The processing circuits 150, 153 may be partially implemented by dedicated hardware and partially implemented by software or firmware. In this way, the processing circuits 150, 153 can implement the above-described functions by dedicated hardware, software, firmware, or a combination of these. The railway wheel flat detection system 1A may be implemented by one processing circuit or multiple processing circuits.

[0149] As described above, in the fourth embodiment, the independent component decomposition unit 41B calculates a time signal Sbb1 limited to a first frequency band F1 capable of detecting vibrations due to a rail joint, and a time signal Sbb2 limited to a second frequency band F2 capable of detecting both vibrations due to a rail joint and vibrations due to a wheel flat. The independent component removal unit 42B then removes, from the time signal Sbb2, an independent component signal Sdd2 corresponding to the time of the independent component signal Sdd1 removed from the time signal Sbb1, thereby calculating a time signal See2 from which noise components have been removed for the second frequency band F2. Furthermore, the independent component removal unit 42B compares the time signal See2 with a threshold value to determine whether a wheel flat has been detected. Therefore, similar to the first embodiment, the railway wheel flat detection system 1B is capable of accurately detecting a wheel flat using a small number of vibration sensors 3.

[0150] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, or different embodiments may be combined with each other. It is also possible to omit or modify parts of the configurations as long as they do not deviate from the gist of the invention.

[0151] Various aspects of the present disclosure are summarized below as appendices.

[0152] (Appendix 1) a frequency analysis unit that converts a vibration sensor signal that indicates information about vibrations measured by a vibration sensor installed on the rail into a first time signal that indicates signal strength for each frequency band; a joint vibration elimination unit that calculates a second time signal by removing a signal of vibration caused by a rail joint, which is a joint of the rail, from a corresponding signal corresponding to the first time signal; a flat detection unit that determines whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail based on the second time signal; an output unit that outputs a determination result of the flat detection unit; Equipped with The joint vibration elimination unit is extracting a third time signal including a frequency of the vibration caused by the rail joint from the corresponding signal, and calculating the second time signal by removing the third time signal from the corresponding signal when the signal strength of the third time signal exceeds a specific value; A railway wheel flat detection system. (Appendix 2) a normalization unit that normalizes a scale of noise other than vibration caused by the rail joint based on the first time signal, The joint vibration elimination unit is calculating a second time signal obtained by removing a signal of vibration caused by the rail joint from the corresponding signal corresponding to the first time signal whose scale of the noise has been normalized; 2. The railway wheel flat detection system according to claim 1, (Appendix 3) The normalization unit where m is a preset parameter, calculating a scale of the noise by dividing m% of the smallest signal values ​​of the first time signal by the value at the m% point of a distribution to which noise other than vibration caused by the rail joint follows, and then dividing the first time signal by the calculated scale of the noise to calculate the first time signal with a normalized scale of the noise. 3. The railway wheel flat detection system according to claim 2, (Appendix 4) a removal width expansion unit that removes a signal a specific time after the time when the third time signal is removed from the second time signal, thereby removing reverberation noise caused by vibrations due to the rail joint; and a similarity filter unit that, upon receiving a wheel detection sensor signal indicating that the wheel has been detected from the wheel detection sensor that detects the wheel, removes non-periodic noise, which is a signal other than the rotation period of the wheel, from the second time signal from which the reverberation noise has been removed, based on the wheel detection sensor signal; Furthermore, The flat detection unit determining whether or not the wheel flat exists from the second time signal from which the reverberation noise and the non-periodic noise have been removed; 4. A railway wheel flat detection system according to any one of claims 1 to 3, (Appendix 5) The similarity filter unit calculating a speed of the vehicle from the wheel detection sensor signal and calculating a time taken for the wheel to make one revolution from the speed; calculating a similarity between a signal at each time point of the second time signal from which the reverberation noise has been removed and a signal with a time difference corresponding to one revolution of the wheel; and removing the non-periodic noise by emphasizing signals for which the similarity is equal to or greater than a specific value; 5. The railway wheel flat detection system according to claim 4, (Appendix 6) a data extracting unit that extracts the corresponding signal at specific time intervals to calculate a time signal group including a plurality of signal elements; the seam vibration elimination unit calculates the second time signal for each of the time signal groups. 6. A railway wheel flat detection system according to any one of claims 1 to 5, (Appendix 7) The joint vibration elimination unit is an independent component decomposition unit that decomposes the corresponding signal into independent component signals that are signals of independent vibration components; an independent component remover that calculates the second time signal by removing the independent component signal from the corresponding signal; having 7. A railway wheel flat detection system according to any one of claims 1 to 6, (Appendix 8) The independent component decomposition unit decomposing the corresponding signals into the independent component signals using independent component analysis; 8. The railway wheel flat detection system according to claim 7, (Appendix 9) The independent component decomposition unit extracting, as the independent component signal, a signal whose norm is greater than a first threshold, thereby decomposing the corresponding signal into the independent component signals; 8. The railway wheel flat detection system according to claim 7, (Appendix 10) The independent component decomposition unit decomposing the independent component signals until an Akaike information criterion for the difference between a component obtained by adding up the M independent component signals and the corresponding signal is minimized, where M is a natural number; 10. The railway wheel flat detection system according to any one of appendices 7 to 9, (Appendix 11) The independent component removal unit removing the independent component signals until the Akaike information criterion of the L independent component signals is minimized, where L is a natural number; 11. The railway wheel flat detection system according to any one of appendices 7 to 10, (Appendix 12) The independent component removal unit removing the independent component signals whose norms are greater than a second threshold; 11. The railway wheel flat detection system according to any one of appendices 7 to 10, (Appendix 13) the signal strength is a magnitude of the norm of the third time signal; 13. The railway wheel flat detection system according to any one of claims 1 to 12, (Appendix 14) a frequency analysis unit that converts a vibration sensor signal that indicates information about vibrations measured by a vibration sensor installed on the rail into a first time signal that indicates signal strength for each frequency band; a joint vibration elimination unit that calculates a second time signal by eliminating a signal of vibration caused by a rail joint, which is a joint of the rail, from the first time signal; a flat detection unit that determines whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail based on the second time signal; an output unit that outputs a determination result of the flat detection unit; Equipped with The frequency analysis unit calculating, as the first time signal, a first limited signal limited to a first frequency band including the frequency of the vibration caused by the rail joint, and a second limited signal limited to a second frequency band capable of detecting both the frequency of the vibration caused by the rail joint and the frequency of the vibration caused by a wheel flat; The joint vibration elimination unit is calculating the second time signal by removing from the second limited signal a signal having a signal intensity that is equal to both a signal corresponding to the first limited signal and a signal corresponding to the second limited signal; A railway wheel flat detection system. (Appendix 15) The joint vibration elimination unit is extracting a first independent component signal from the first limited signal, which is a signal at a time when the signal strength exceeds a specific value, and extracting a second independent component signal from the second limited signal, which is a signal at the same time as the first limited signal, and removing the second independent component signal from the second limited signal, thereby calculating the second time signal. 15. The railway wheel flat detection system according to claim 14, (Appendix 16) a frequency analysis step in which the railway wheel flat detection system converts a vibration sensor signal, which indicates information on vibrations measured by a vibration sensor installed on the rail, into a first time signal, which indicates signal strength for each frequency band; a joint vibration elimination step in which the railway wheel flat detection system calculates a second time signal by eliminating a signal of vibration caused by a rail joint, which is the joint of the rail, from the corresponding signal corresponding to the first time signal; a flat detection step in which the railway wheel flat detection system determines whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail based on the second time signal; an output step in which the railway wheel flat detection system outputs a determination result of the presence or absence of the wheel flat; Including, In the step of eliminating seam vibration, the railway wheel flat detection system extracts a third time signal, which includes a frequency of vibration caused by the rail joint, from the corresponding signal, and calculates the second time signal by removing the third time signal from the corresponding signal when the signal strength of the third time signal exceeds a specific value; A railway wheel flat detection method. (Appendix 17) a frequency analysis step of converting a vibration sensor signal indicating information on vibrations measured by a vibration sensor installed on the rail into a first time signal indicating signal strength for each frequency band; a joint vibration elimination step of calculating a second time signal by eliminating a signal of vibration caused by a rail joint, which is a joint of the rail, from a corresponding signal corresponding to the first time signal; a flat detection step of determining whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail from the second time signal; an output step of outputting the determination result of the presence or absence of the wheel flat; on the computer, In the step of eliminating seam vibration, extracting a third time signal including a frequency of the vibration caused by the rail joint from the corresponding signal, and calculating the second time signal by removing the third time signal from the corresponding signal when the signal strength of the third time signal exceeds a specific value; A railway wheel flat detection program. [Explanation of symbols]

[0153] 1A, 1B Railway wheel flat detection system, 2 Wheel detection sensor, 3 Vibration sensor, 4 Vehicle, 5 Rail, 10A, 10B Frequency analysis unit, 20 Normalization unit, 30 Data extraction unit, 40A, 40B Joint vibration removal unit, 41A, 41B Independent component decomposition unit, 42A, 42B Independent component removal unit, 50 Data combination unit, 60 Removal width expansion unit, 70 Similarity filter unit, 80A, 80B Flat detection unit, 90 Output unit, 91 Display unit, 150, 153 Processing circuit, 151 Processor, 152 Memory, F1 First frequency band, F2 Second frequency band, FH1 to FH3 First frequency component, FL1 to FL5 Second frequency component, N1 Non-periodic noise, P1 to P5 Peak, P3E Reverberation noise, Sa1, X1 Vibration sensor signal, Sb, Sb1, Sc, Sc1 to ScN, Se, Se1 to SeN, Sf, Sg frequency spectrum signals, Sbb1, Sbb2, See1, See2, Sff, Sgg, Sh time signal, Scn, Sen frequency spectrum signal group, Sd1 to SdN, Sd1a to Sd1d, Sdd1, Sdd2 independent component signals, Sdn independent component signal group, Th1 to Th4 threshold values, X2 wheel detection sensor signal.

Claims

1. a frequency analysis unit that converts a vibration sensor signal indicating information about vibrations measured by a vibration sensor installed on the rail into a first time signal indicating signal strength for each frequency band; a joint vibration elimination unit that calculates a second time signal by eliminating a signal of vibration caused by a rail joint, which is a joint of the rail, from a corresponding signal corresponding to the first time signal; a flat detection unit that determines whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail, based on the second time signal; an output unit that outputs a determination result of the flat detection unit; Equipped with The joint vibration elimination unit is extracting a third time signal including a frequency of the vibration caused by the rail joint from the corresponding signal, and calculating the second time signal by removing the third time signal from the corresponding signal when the signal strength of the third time signal exceeds a specific value; A railway wheel flat detection system.

2. a normalization unit that normalizes a scale of noise other than the vibration caused by the rail joint based on the first time signal, The joint vibration elimination unit is calculating a second time signal obtained by removing a signal of vibration caused by the rail joint from the corresponding signal corresponding to the first time signal whose scale of the noise has been normalized; 2. The railway wheel flat detection system of claim 1.

3. The normalization unit where m is a preset parameter, calculating a scale of the noise by dividing m% of the smallest signal values ​​of the first time signal by the value at the m% point of a distribution to which noise other than vibration caused by the rail joint follows, and then dividing the first time signal by the calculated scale of the noise to calculate the first time signal with a normalized scale of the noise.

3. The railway wheel flat detection system of claim 2.

4. a removal width expansion unit that removes a signal that is a specific time after the time when the third time signal is removed from the second time signal, thereby removing reverberation noise caused by vibrations due to the rail joint; and a similarity filter unit that, upon receiving a wheel detection sensor signal indicating that the wheel has been detected from the wheel detection sensor that detects the wheel, removes non-periodic noise, which is a signal other than the rotation period of the wheel, from the second time signal from which the reverberation noise has been removed, based on the wheel detection sensor signal; Furthermore, The flat detection unit determining whether or not the wheel flat exists from the second time signal from which the reverberation noise and the non-periodic noise have been removed; 2. The railway wheel flat detection system of claim 1.

5. The similarity filter unit calculating a speed of the vehicle from the wheel detection sensor signal and calculating a time taken for the wheel to make one revolution from the speed; calculating a similarity between a signal at each time point of the second time signal from which the reverberation noise has been removed and a signal at a time difference corresponding to one revolution of the wheel; and removing the non-periodic noise by emphasizing signals for which the similarity is equal to or greater than a specific value; 5. The railway wheel flat detection system of claim 4.

6. a data extracting unit that extracts the corresponding signal at specific time intervals to calculate a time signal group including a plurality of signal elements; the seam vibration elimination unit calculates the second time signal for each of the time signal groups.

2. The railway wheel flat detection system of claim 1.

7. The joint vibration elimination unit is an independent component decomposition unit that decomposes the corresponding signal into independent component signals that are signals of independent vibration components; an independent component remover that calculates the second time signal by removing the independent component signal from the corresponding signal; having 2. The railway wheel flat detection system of claim 1.

8. The independent component decomposition unit decomposing the corresponding signals into the independent component signals using independent component analysis; 8. The railway wheel flat detection system of claim 7.

9. The independent component decomposition unit extracting, as the independent component signal, a signal whose norm is greater than a first threshold, thereby decomposing the corresponding signal into the independent component signals; 8. The railway wheel flat detection system of claim 7.

10. The independent component decomposition unit decomposing the independent component signals until an Akaike information criterion of the difference between a component obtained by adding up M independent component signals and the corresponding signal, where M is a natural number, is minimized; 8. The railway wheel flat detection system of claim 7.

11. The independent component removal unit removing the independent component signals until the Akaike information criterion of the L independent component signals is minimized, where L is a natural number; 8. The railway wheel flat detection system of claim 7.

12. The independent component removal unit removing the independent component signals whose norms are greater than a second threshold; 8. The railway wheel flat detection system of claim 7.

13. the signal strength is a magnitude of the norm of the third time signal; Railway wheel flat detection system according to any one of claims 1 to 12.

14. a frequency analysis unit that converts a vibration sensor signal indicating information about vibrations measured by a vibration sensor installed on the rail into a first time signal indicating signal strength for each frequency band; a joint vibration elimination unit that calculates a second time signal by eliminating a signal of vibration caused by a rail joint, which is a joint of the rail, from the first time signal; a flat detection unit that determines whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail, based on the second time signal; an output unit that outputs a determination result of the flat detection unit; Equipped with The frequency analysis unit calculating, as the first time signal, a first limited signal limited to a first frequency band including the frequency of the vibration caused by the rail joint, and a second limited signal limited to a second frequency band capable of detecting both the frequency of the vibration caused by the rail joint and the frequency of the vibration caused by a wheel flat; The joint vibration elimination unit is calculating the second time signal by removing from the second limited signal a signal having a signal intensity that is equal to both a signal corresponding to the first limited signal and a signal corresponding to the second limited signal; A railway wheel flat detection system.

15. The joint vibration elimination unit is extracting a first independent component signal from the first limited signal, the first independent component signal being a signal at a time when the signal strength exceeds a specific value, and extracting a second independent component signal from the second limited signal, the second independent component signal being a signal at the same time as the first limited signal, and removing the second independent component signal from the second limited signal, thereby calculating the second time signal.

15. The railway wheel flat detection system of claim 14.

16. a frequency analysis step in which the railway wheel flat detection system converts a vibration sensor signal, which indicates information on vibrations measured by a vibration sensor installed on the rail, into a first time signal, which indicates signal strength for each frequency band; a joint vibration elimination step in which the railway wheel flat detection system calculates a second time signal by eliminating a signal of vibration caused by a rail joint, which is a joint of the rail, from the corresponding signal corresponding to the first time signal; a flat detection step in which the railway wheel flat detection system determines whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail based on the second time signal; an output step in which the railway wheel flat detection system outputs a determination result of the presence or absence of the wheel flat; Including, In the step of eliminating seam vibration, the railway wheel flat detection system extracts a third time signal, which includes a frequency of vibration caused by the rail joint, from the corresponding signal, and calculates the second time signal by removing the third time signal from the corresponding signal when the signal strength of the third time signal exceeds a specific value. A railway wheel flat detection method.

17. a frequency analysis step of converting a vibration sensor signal indicating information on vibrations measured by a vibration sensor installed on the rail into a first time signal indicating signal strength for each frequency band; a joint vibration removing step of calculating a second time signal by removing a signal of vibration caused by a rail joint, which is a joint of the rail, from a corresponding signal corresponding to the first time signal; a flat detection step of determining whether or not a wheel flat has occurred on a wheel of a vehicle running on the rail from the second time signal; an output step of outputting the determination result of the presence or absence of the wheel flat; on the computer, In the step of eliminating seam vibration, extracting a third time signal including a frequency of the vibration caused by the rail joint from the corresponding signal, and calculating the second time signal by removing the third time signal from the corresponding signal when the signal strength of the third time signal exceeds a specific value; A railway wheel flat detection program.

Images