A method and system for denoising magnetic leakage signals from rail surface damage

Abstract

The present invention discloses a denoising method for a magnetic leakage signal of damage on a rail surface. A given signal is processed by CEEMD to filter out a mode without damage components. In the reconstructed signal, small damage that was originally difficult to find can also be effectively identified, and suspected damage can be effectively distinguished. At the same time, a corresponding mode is obtained by differential processing of a mode containing damage components and an auxiliary noise signal processed by CEEMD. The corresponding modes are summed to obtain a filtered damage signal. The filtered damage signal is the denoised damage signal, which effectively suppresses vibration interference. The present invention correspondingly provides a denoising system for a magnetic leakage signal of damage on a rail surface.

Description

Technical Field

[0001] The invention relates to the technical field of nondestructive testing, and in particular to a method and system for denoising magnetic leakage signals of rail surface damage. Background Art

[0002] Nondestructive testing is an emerging discipline that uses the changes in the response to heat, sound, electricity, light, magnetism, etc. caused by the abnormalities and damages in the internal structure of the material to evaluate the abnormalities and damages of the structure, that is, to detect whether there are cracks, inclusions and other damages in the internal structure, physical properties or state without damaging the workpiece or material being tested. The magnetic flux leakage nondestructive testing method can detect surface and near-surface damage of ferromagnetic workpieces, and has the advantages of high detection sensitivity, fast speed, low requirements for workpiece surface cleanliness, low cost, and simple operation. It is widely used in nondestructive testing of surface damages of ferromagnetic materials such as rails, steel pipes, etc.

[0003] When the detection probe is patrolling the surface of the workpiece, it is affected by factors such as vibration, causing the output of the magnetic sensor to change. The vibration interference is superimposed on the detection signal, making it difficult to distinguish the damage signal, which is not conducive to damage measurement. Since the vibration interference is affected by factors such as the inspection speed, it is difficult to effectively filter out the existing digital filtering technology. In addition to vibration interference, factors such as background magnetic field and residual magnetism of rails can also interfere with the signal, collectively referred to as background noise. Vibration interference and background noise act together on the detection signal, making it more difficult to identify damage. In order to effectively detect, identify and reconstruct damage, it is necessary to denoise the leakage magnetic signal. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a denoising method for leakage magnetic signals of rail surface damage, which can effectively suppress the vibration and background noise interference during the inspection process, increase the signal-to-noise ratio of the damage signal, and is suitable for leakage magnetic detection of rail damage. The present invention correspondingly provides a denoising system for leakage magnetic signals of rail surface damage.

[0005] Technical solution: The method for denoising the rail surface damage magnetic leakage signal provided by the present invention comprises the following steps:

[0006] S1: A single magnetic sensor is used as a single channel, and multiple magnetic sensors are used as multiple channels to identify the detection signal to form a multi-channel detection signal. At the same time, the damaged signal and the non-damaged signal in the multi-channel detection signal are distinguished. The n-order complementary integrated empirical mode decomposition CEEMD is performed on any damaged signal that needs to be filtered to obtain the decomposed modes IMF1, IMF2, ..., IMF n and the residual component RES;

[0007] S2: Calculate the kurtosis values ​​of all modes, and take the mode with the largest kurtosis value as the reference mode IMF m;

[0008] S3: Perform Hilbert transform on all modes to obtain the corresponding envelope signal and envelope energy;

[0009] S4: Calculate the benchmark modal IMF m The envelope signal and other modal IMFs i The correlation coefficient of the envelope signal is used to calculate the reference modal IMF m The envelope energy of i The correlation coefficient of the envelope energy, where i = 1, 2, ..., n, if the two calculated correlation coefficients are greater than the threshold, then it is considered that other modal IMF i If there is a damage component, if one or two correlation coefficients are less than or equal to the threshold, then the other modal IMF is considered i Contains no damaging ingredients;

[0010] S5: Select any non-impaired signal as the auxiliary noise signal, perform n-order CEEMD on the auxiliary noise signal, and obtain the decomposed modal IMF'1, IMF'2, ..., IMF' n and the residual component RES';

[0011] S6: Subtract the modes after decomposition of the auxiliary noise signal in S5 from the modes with damage components identified in S4 to obtain filtered modes, and delete the modes without damage components identified in S4 and the residual component RES;

[0012] S7: summing the modes filtered in S6 to obtain a reconstructed filtered impairment signal.

[0013] Furthermore, the envelope signal and envelope energy in S3 are calculated as follows: given a signal x(t), its envelope signal A(t) and envelope energy E(t) are respectively:

[0014]

[0015] E(t)=A 2 (t) = x 2 (t)+H 2 [x(t)]

[0016] H[] represents Hilbert transform.

[0017] Furthermore, the calculation method of the impairment mode of the impairment signal in S6 after filtering is as follows: the modes of the impairment component signal CEEMD in S1 are IMF1, IMF2, ..., IMF nand the residual component RES, and CEEMD is performed on the auxiliary noise signal. The decomposition order n is the same as the order of CEEMD of the damaged signal in S1, and the decomposed modes are IMF'1, IMF'2, ..., IMF' n and the residual component RES';

[0018] From IMF1 to IMF n Deal with each case individually, when dealing with IMF j If the IMF j The envelope signal and envelope energy of IMF m If the correlation coefficients of the envelope signal and envelope energy are both greater than the threshold, then IMF" j =IMF j -IMF' j , to the IMF” j As a filtered mode, if one or both correlation coefficients are less than or equal to the threshold, the modal IMF without the impairment component will be j delete.

[0019] The present invention provides a denoising system for rail surface damage magnetic leakage signals, comprising a damage signal processing module, a kurtosis value calculation module, a Hilbert transformation module, a judgment module, an auxiliary noise signal processing module, a filtering module, and a summing module;

[0020] The damage signal processing module is used to use a single magnetic sensor as a single channel, and multiple magnetic sensors as multiple channels to identify the detection signal to form a multi-channel detection signal, and distinguish the damage signal and the non-damage signal in the multi-channel detection signal. The n-order complementary integrated empirical mode decomposition CEEMD is performed on any damage signal that needs to be filtered to obtain the decomposed modes IMF1, IMF2, ..., IMF n and the residual component RES;

[0021] The kurtosis calculation module is used to calculate the kurtosis values ​​of all modes, and the mode with the largest kurtosis value is taken as the benchmark mode IMF. m ;

[0022] The Hilbert transform module is used to perform Hilbert transform on all modes to obtain the corresponding envelope signal and envelope energy;

[0023] The judgment module is used to calculate the reference modal IMF m The envelope signal and other modal IMFs i The correlation coefficient of the envelope signal is used to calculate the reference modal IMF m The envelope energy of iThe correlation coefficient of the envelope energy, where i = 1, 2, ..., n, if the two calculated correlation coefficients are greater than the threshold, then it is considered that other modal IMF i If there is a damage component, if one or two correlation coefficients are less than or equal to the threshold, then the other modal IMF is considered i Contains no damaging ingredients;

[0024] The auxiliary noise signal processing module is used to select any non-impaired signal as the auxiliary noise signal, perform n-order CEEMD on the auxiliary noise signal, and obtain the decomposed modal IMF'1, IMF'2, ..., IMF' n and the residual component RES';

[0025] The filtering module is used to subtract the modes after decomposition of the auxiliary noise signal in the auxiliary noise signal processing module from the modes with damage components identified in the judgment module to obtain the filtered modes, and delete the modes without damage components identified in the judgment module and the residual component RES;

[0026] The summing module is used to sum the modes filtered in the filtering module to obtain a reconstructed filtered damage signal.

[0027] Furthermore, the envelope signal and envelope energy in the Hilbert transform module are calculated as follows: given a signal x(t), its envelope signal A(t) and envelope energy E(t) are respectively:

[0028]

[0029] E(t)=A 2 (t) = x 2 (t)+H 2 [x(t)]

[0030] H[] represents Hilbert transform.

[0031] Furthermore, the calculation method of the impairment mode of the impairment signal after filtering in the filtering module is as follows: the modes of the impairment component signal CEEMD in S1 are IMF1, IMF2, ..., IMF n and the residual component RES, and CEEMD is performed on the auxiliary noise signal. The decomposition order n is the same as the order of CEEMD of the damaged signal in S1, and the decomposed modes are IMF'1, IMF'2, ..., IMF' n and the residual component RES';

[0032] From IMF1 to IMF n Deal with each case individually, when dealing with IMF j If the IMF j The envelope signal and envelope energy of IMFm If the correlation coefficients of the envelope signal and envelope energy are both greater than the threshold, then IMF" j =IMF j -IMF' j , to the IMF” j As a filtered mode, if one or both correlation coefficients are less than or equal to the threshold, the modal IMF without the impairment component will be j delete.

[0033] Beneficial effect: Compared with the prior art, the present invention has the remarkable feature of processing a given signal through CEEMD to filter out the mode without damage components. In the reconstructed signal, the originally difficult-to-find small damage can also be effectively identified, and the suspected damage can be effectively judged. At the same time, the corresponding mode is obtained by differential processing of the mode containing the damage component and the auxiliary noise signal after CEEMD processing. The corresponding modes are summed to obtain the filtered damage signal. The filtered damage signal is the denoised damage signal, which effectively suppresses vibration interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the structure of the magnetic flux leakage detection probe in the present invention;

[0035] Figure 2 It is a schematic diagram of multi-channel magnetic leakage signal in the present invention;

[0036] Figure 3 It is a schematic diagram of the process of the present invention;

[0037] Figure 4 It is a diagram of the original magnetic flux leakage signal and its CEEMD processing result in the present invention;

[0038] Figure 5 This is a denoising effect diagram of the present invention. DETAILED DESCRIPTION

[0039] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] Example 1

[0041] See also Figure 3 As shown, the present invention provides a method for denoising a rail surface damage magnetic leakage signal, comprising the following steps:

[0042] S1: A single magnetic sensor is used as a single channel, and multiple magnetic sensors are used as multiple channels to identify the detection signal to form a multi-channel detection signal. At the same time, the damaged signal and the non-damaged signal in the multi-channel detection signal are distinguished. The n-order complementary integrated empirical mode decomposition CEEMD is performed on any damaged signal that needs to be filtered to obtain the decomposed modes IMF1, IMF2, ..., IMFn and the remaining component RES.

[0043] See also Figure 1 As shown, the detection range of a single magnetic sensor is limited. Generally, a row of magnetic sensors are arranged along the y direction so that the detection range covers the entire surface of the rail head.

[0044] Since the leakage magnetic field of damage is relatively weak and has a small distribution range, when a row of magnetic sensors pass through the damaged part, some magnetic sensors can detect the leakage magnetic field of damage, and their output data will change due to the damage; some magnetic sensors cannot detect the leakage magnetic field of damage, and their data will not change due to the damage. The output of the magnetic sensor that detects the leakage magnetic field of damage is affected by the leakage magnetic field of damage, vibration interference and background noise, while the data of the magnetic sensor that does not detect the leakage magnetic field is only affected by vibration interference and background noise. A certain recognition algorithm can be used to distinguish between damage signals and non-damage signals, such as Figure 2 As shown, eight magnetic sensors are arranged along the y direction, and each magnetic sensor serves as a channel.

[0045] The steps of complementary integrated empirical mode decomposition (CEEMD) are as follows:

[0046] 1.1) Given a signal x(t), add a pair of Gaussian white noise signals, we have:

[0047] x i (t) = x(t) + (-1) i MOD 2 εv i (t)

[0048] In the formula, v i (t) represents the i-th white noise sequence (i=1,2,…,N); the higher the N, the higher the decomposition accuracy, but the lower the algorithm efficiency, and it is usually taken as 100; ε is the standard deviation of Gaussian white noise. The smaller the value, the higher the decomposition accuracy, but too small a value may cause the distribution of the original signal peak to remain unchanged and the algorithm to fail, and it is usually taken as 0.1~0.2; MOD represents the modulo operation.

[0049] x i (t)(i=1,2,…,N) are decomposed by EMD to obtain N first-order modal components imf1 i (t), that is:

[0050] x i (t) = imf1 i (t)+r1 i (t)

[0051] 1.2) After N times decomposition, imf1 i (t) After averaging, we get the final first-order That is:

[0052]

[0053] The residual component r1(t) of the mode after the first-order decomposition can be expressed as:

[0054]

[0055] 1.3) Take the above obtained r1(t) as the input x(t) of step 1), repeat steps 1.1) and 1.2) to obtain the second-order modal component imf2 i (t), that is:

[0056] r1(t)=imf2 i (t)+r2 i (t)

[0057] Second order modal component have:

[0058]

[0059] The residual component r2(t) of the second-order mode is:

[0060]

[0061] 1.4) Repeat the above steps until the residual component after a certain order of decomposition can no longer be decomposed by CEEMD, and the k+1th order modal component is obtained. and the k+1th order residual component r k+1 (t).

[0062] r k (t) = imf k+1 i (t)+r k+1 i (t)

[0063]

[0064]

[0065] Then the signal x(t) can be expressed as:

[0066]

[0067] For a given signal x(t), the basic steps of the EMD algorithm are as follows:

[0068] 1.1.1) Determine all local extreme points in the original signal x(t), including maximum and minimum values;

[0069] 1.1.2) Use cubic spline interpolation method to construct the upper envelope u(t) and lower envelope u′(t) of x(t) for the determined extreme points;

[0070] 1.1.3) Calculate the average value of the upper envelope and the lower envelope m(t) = [u(t) + u′(t)] / 2;

[0071] 1.1.4) Find the difference between x(t) and m(t), h(t) = x(t) - m(t), and determine whether h(t) satisfies the two conditions defined by the intrinsic mode function: one is that in the entire time series, the number of extreme points, including maximum and minimum values, and the number of zero-crossing points must be equal or differ by at most one; the other is that at any moment in the time series, the mean of the upper envelope determined by the signal maximum and the lower envelope determined by the minimum is zero, and the signal has local characteristics that are symmetrical about the time axis.

[0072] If the condition is not met, h(t) is used as the original signal x(t) and the above steps are repeated. If the condition is met, the first intrinsic mode function imf1(t) = h(t), which represents the highest frequency component of the signal in the local time; 1.1.5) Subtract the imf decomposed in step 4 from x(t) j (t) get r j (t), r j (t) is used as the input of step 1.1.1) and the above steps are repeated until r is satisfied. j (t) EMD decomposition is no longer possible.

[0073] S2: Calculate the kurtosis values ​​of all modes, and take the mode with the largest kurtosis value as the reference mode IMF m ;

[0074] The mode containing the damage signal usually has a higher kurtosis value. Therefore, we can first determine that the mode with the largest kurtosis value is the modal IMF containing the damage signal component. m , except for the IMF m In addition, other modes may also contain damage components. In order to distinguish which modes contain damage components and which do not, the present invention uses IMF m As the reference signal, the IMF m The mode with stronger correlation contains the damage component, while the mode with weaker correlation does not contain the damage component.

[0075] The kurtosis value calculation formula is:

[0076]

[0077] In the formula, μ is the mean of the input signal x, σ is the standard deviation of the input signal x, and E() represents the calculation of its expected value.

[0078] S3: Perform Hilbert transform on all modes to obtain the corresponding envelope signal and envelope energy;

[0079] Since the frequency ranges of various modes are different, the signal change details may be inconsistent, and directly calculating the correlation coefficient of the original signal waveform is not accurate. The present invention adopts a method of calculating the modal envelope signal correlation and envelope energy correlation, and according to the Hilbert transform, the envelope signal and envelope energy of each mode are obtained.

[0080] The calculation method of envelope signal and envelope energy is: given a signal x(t), its envelope signal A(t) and envelope energy E(t) are respectively:

[0081]

[0082] E(t)=A 2 (t) = x 2 (t)+H 2 [x(t)]

[0083] H[] represents Hilbert transform.

[0084] S4: Calculate the benchmark modal IMF m The envelope signal and other modal IMFs i The correlation coefficient of the envelope signal is used to calculate the reference modal IMF m The envelope energy of i The correlation coefficient of the envelope energy, where i = 1, 2, ..., n, if the two calculated correlation coefficients are greater than the threshold, then it is considered that other modal IMF i If there is a damage component, if one or two correlation coefficients are less than or equal to the threshold, then the other modal IMF is considered i There is no damage component, and the threshold of the correlation coefficient is set to 0.5.

[0085] S5: Select any non-impaired signal as the auxiliary noise signal, perform n-order CEEMD on the auxiliary noise signal, and obtain the decomposed modal IMF'1, IMF'2, ..., IMF' n and the residual component RES';

[0086] S6: Subtract the modes after decomposition of the auxiliary noise signal in S5 from the modes with damage components identified in S4 to obtain filtered modes, and delete the modes without damage components identified in S4 and the residual component RES.

[0087] From IMF1 to IMF n Process them one by one, and when they reach IMF j If the IMF jThe envelope signal and envelope energy of IMF m If the correlation coefficients of the envelope signal and envelope energy are both greater than the threshold, then IMF" j =IMF j -IMF' j , to the IMF” j As a filtered mode, if there is a correlation coefficient less than or equal to the threshold, the modal IMF without the damage component will be j delete.

[0088] S7: summing the modes filtered in S6 to obtain a reconstructed filtered impairment signal.

[0089] See also Figure 2 As shown, this embodiment takes the 4th-order CEEMD decomposition as an example, and selects a leakage magnetic detection signal of 8 channels in the Z direction at a detection speed of 80 km / h. The signal includes damage signals and vibration signals, and there are 4 damage signals.

[0090] The decomposed modes IMF1, IMF2, IMF3, IMF4 and the residual component RES are obtained. According to the maximum kurtosis criterion, the mode with the largest kurtosis value is selected. In this embodiment, IMF1 is used as the reference mode IMF. m , calculate IMF m Correlation coefficient with the envelope signals of other modes, calculation of IMF m The correlation coefficient between the envelope energy of the two modes and the other modes is found, and the modes with two correlation coefficients greater than 0.5 are found.

[0091] Figure 4 for Figure 2 The effect of the signal of middle channel 4 after 4th-order CEEMD decomposition. IMF1 has the largest kurtosis and is used as the benchmark modal IMF. m Through correlation calculation, it is found that IMF1, IMF2 and IMF3 contain damage components, while IMF4 and the remaining component RES do not contain damage components.

[0092] Take the non-impaired signal channel 2 as the auxiliary noise channel to obtain IMF'1, IMF'2, IMF'3, IMF'4 and RES', calculate IMF"1 = IMF1-IMF'1, IMF"2 = IMF2-IMF'2, IMF"3 = IMF3-IMF'3, and then sum IMF"1, IMF"2 and IMF"3.

[0093] The first sampling points of IMF”1, IMF”2 and IMF”3 are added together, and the sum is used as the first sampling point of the filtered signal; similarly, the second sampling points of IMF”1, IMF”2 and IMF”3 are added together, and the sum is used as the second sampling point of the filtered signal; and so on, the result of arranging the summed sampling points in sequence is used as the final filtered damage signal, that is, the denoised signal, and its denoising effect is as follows: Figure 5 shown.

[0094] Example 2

[0095] See also Figure 3 As shown, corresponding to the denoising method of rail surface damage leakage magnetic signal of embodiment 1, this embodiment 2 provides a denoising system for rail surface damage leakage magnetic signal, including a damage signal processing module, a kurtosis value calculation module, a Hilbert transformation module, a judgment module, an auxiliary noise signal processing module, a filtering module, and a summing module;

[0096] The damage signal processing module is used to use a single magnetic sensor as a single channel, and multiple magnetic sensors as multiple channels to identify the detection signal to form a multi-channel detection signal, and distinguish the damage signal and the non-damage signal in the multi-channel detection signal. The n-order complementary integrated empirical mode decomposition CEEMD is performed on any damage signal that needs to be filtered to obtain the decomposed modes IMF1, IMF2, ..., IMF n and the residual component RES;

[0097] See also Figure 1 As shown, the detection range of a single magnetic sensor is limited. Generally, a row of magnetic sensors are arranged along the y direction so that the detection range covers the entire surface of the rail head.

[0098] Since the leakage magnetic field of damage is relatively weak and has a small distribution range, when a row of magnetic sensors pass through the damaged part, some magnetic sensors can detect the leakage magnetic field of damage, and their output data will change due to the damage; some magnetic sensors cannot detect the leakage magnetic field of damage, and their data will not change due to the damage. The output of the magnetic sensor that detects the leakage magnetic field of damage is affected by the leakage magnetic field of damage, vibration interference and background noise, while the data of the magnetic sensor that does not detect the leakage magnetic field is only affected by vibration interference and background noise. A certain recognition algorithm can be used to distinguish between damage signals and non-damage signals, such as Figure 2 As shown, eight magnetic sensors are arranged along the y direction, and each magnetic sensor serves as a channel.

[0099] The steps of complementary integrated empirical mode decomposition (CEEMD) are as follows:

[0100] 1.1) Given a signal x(t), add a pair of Gaussian white noise signals, we have:

[0101] x i (t) = x(t) + (-1) i MOD 2 εv i (t)

[0102] In the formula, v i (t) represents the i-th white noise sequence (i=1,2,…,N); the higher the N, the higher the decomposition accuracy, but the lower the algorithm efficiency, and it is usually taken as 100; ε is the standard deviation of Gaussian white noise. The smaller the value, the higher the decomposition accuracy, but too small a value may cause the distribution of the original signal peak to remain unchanged and the algorithm to fail, and it is usually taken as 0.1~0.2; MOD represents the modulo operation.

[0103] x i (t)(i=1,2,…,N) are decomposed by EMD to obtain N first-order modal components imf1 i (t), that is:

[0104] x i (t) = imf1 i (t)+r1 i (t)

[0105] 1.2) After N times decomposition, imf1 i (t) After averaging, we get the final first-order That is:

[0106]

[0107] The residual component r1(t) of the mode after the first-order decomposition can be expressed as:

[0108]

[0109] 1.3) Take the above obtained r1(t) as the input x(t) of step 1), repeat steps 1.1) and 1.2) to obtain the second-order modal component imf2 i (t), that is:

[0110] r1(t)=imf2 i (t)+r2 i (t)

[0111] Second order modal component have:

[0112]

[0113] The residual component r2(t) of the second-order mode is:

[0114]

[0115] 1.4) Repeat the above steps until the residual component after a certain order of decomposition can no longer be decomposed by CEEMD, and the k+1th order modal component is obtained. and the k+1th order residual component r k+1 (t).

[0116] r k (t) = imf k+1 i (t)+r k+1 i (t)

[0117]

[0118]

[0119] Then the signal x(t) can be expressed as:

[0120]

[0121] For a given signal x(t), the basic steps of the EMD algorithm are as follows:

[0122] 1.1.1) Determine all local extreme points in the original signal x(t), including maximum and minimum values;

[0123] 1.1.2) Use cubic spline interpolation method to construct the upper envelope u(t) and lower envelope u′(t) of x(t) for the determined extreme points;

[0124] 1.1.3) Calculate the average value of the upper envelope and the lower envelope m(t) = [u(t) + u′(t)] / 2;

[0125] 1.1.4) Find the difference between x(t) and m(t), h(t) = x(t) - m(t), and determine whether h(t) satisfies the two conditions defined by the intrinsic mode function: one is that in the entire time series, the number of extreme points, including maximum and minimum values, and the number of zero-crossing points must be equal or differ by at most one; the other is that at any moment in the time series, the mean of the upper envelope determined by the signal maximum and the lower envelope determined by the minimum is zero, and the signal has local characteristics that are symmetrical about the time axis.

[0126] If the condition is not met, h(t) is used as the original signal x(t) and the above steps are repeated. If the condition is met, the first intrinsic mode function imf1(t) = h(t), which represents the highest frequency component of the signal in the local time; 1.1.5) Subtract the imf decomposed in step 4 from x(t) j (t) get r j (t), r j(t) is used as the input of step 1.1.1) and the above steps are repeated until r is satisfied. j (t) EMD decomposition is no longer possible.

[0127] The kurtosis calculation module is used to calculate the kurtosis values ​​of all modes, and the mode with the largest kurtosis value is taken as the benchmark mode IMF. m ;

[0128] The mode containing the damage signal usually has a higher kurtosis value. Therefore, we can first determine that the mode with the largest kurtosis value is the modal IMF containing the damage signal component. m , except for the IMF m In addition, other modes may also contain damage components. In order to distinguish which modes contain damage components and which do not, the present invention uses IMF m As the reference signal, the IMF m The mode with stronger correlation contains the damage component, while the mode with weaker correlation does not contain the damage component.

[0129] The kurtosis value calculation formula is:

[0130]

[0131] In the formula, μ is the mean of the input signal x, σ is the standard deviation of the input signal x, and E() represents the calculation of its expected value.

[0132] The Hilbert transform module is used to perform Hilbert transform on all modes to obtain the corresponding envelope signal and envelope energy;

[0133] Since the frequency ranges of various modes are different, the signal change details may be inconsistent, and directly calculating the correlation coefficient of the original signal waveform is not accurate. The present invention adopts a method of calculating the modal envelope signal correlation and envelope energy correlation, and according to the Hilbert transform, the envelope signal and envelope energy of each mode are obtained.

[0134] The calculation method of envelope signal and envelope energy is: given a signal x(t), its envelope signal A(t) and envelope energy E(t) are respectively:

[0135]

[0136] E(t)=A 2 (t) = x 2 (t)+H 2 [x(t)]

[0137] H[] represents Hilbert transform.

[0138] The judgment module is used to calculate the reference modal IMF mThe envelope signal and other modal IMFs i The correlation coefficient of the envelope signal is used to calculate the reference modal IMF m The envelope energy of i The correlation coefficient of the envelope energy, where i = 1, 2, ..., n, if the two calculated correlation coefficients are greater than the threshold, then it is considered that other modal IMF i If there is a damage component, if one or two correlation coefficients are less than or equal to the threshold, then the other modal IMF is considered i There is no damage component, and the threshold of the correlation coefficient is set to 0.5.

[0139] The auxiliary noise signal processing module is used to select any non-impaired signal as the auxiliary noise signal, perform n-order CEEMD on the auxiliary noise signal, and obtain the decomposed modal IMF'1, IMF'2, ..., IMF' n and the residual component RES';

[0140] The filtering module is used to subtract the decomposed modes of the auxiliary noise signal in the auxiliary noise signal processing module from the modes with damage components identified in the judgment module to obtain filtered modes, and delete the modes without damage components identified in the judgment module and the residual component RES.

[0141] From IMF1 to IMF n Process them one by one, and when they reach IMF j If the IMF j The envelope signal and envelope energy of IMF m If the correlation coefficients of the envelope signal and envelope energy are both greater than the threshold, then IMF" j =IMF j -IMF' j , to the IMF” j As a filtered mode, if there is a correlation coefficient less than or equal to the threshold, the modal IMF without the damage component will be j delete.

[0142] The summing module is used to sum the modes filtered in the filtering module to obtain a reconstructed filtered damage signal.

[0143] See also Figure 2 As shown, this embodiment takes the 4th-order CEEMD decomposition as an example, and selects a leakage magnetic detection signal of 8 channels in the Z direction at a detection speed of 80 km / h. The signal includes damage signals and vibration signals, and there are 4 damage signals.

[0144] The decomposed modes IMF1, IMF2, IMF3, IMF4 and the residual component RES are obtained. According to the maximum kurtosis criterion, the mode with the largest kurtosis value is selected. In this embodiment, IMF1 is used as the reference mode IMF. m , calculate IMF m Correlation coefficient with the envelope signals of other modes, calculation of IMF m The correlation coefficient between the envelope energy of the two modes and the other modes is found, and the modes with two correlation coefficients greater than 0.5 are found.

[0145] Figure 4 for Figure 2 The effect of the signal of middle channel 4 after 4th-order CEEMD decomposition. IMF1 has the largest kurtosis and is used as the benchmark modal IMF. m Through correlation calculation, it is found that IMF1, IMF2 and IMF3 contain damage components, while IMF4 and the remaining component RES do not contain damage components.

[0146] Take the non-impaired signal channel 2 as the auxiliary noise channel to obtain IMF'1, IMF'2, IMF'3, IMF'4 and RES', calculate IMF"1 = IMF1-IMF'1, IMF"2 = IMF2-IMF'2, IMF"3 = IMF3-IMF'3, and then sum IMF"1, IMF"2 and IMF"3.

[0147] The first sampling points of IMF”1, IMF”2 and IMF”3 are added together, and the sum is used as the first sampling point of the filtered signal; similarly, the second sampling points of IMF”1, IMF”2 and IMF”3 are added together, and the sum is used as the second sampling point of the filtered signal; and so on, the result of arranging the summed sampling points in sequence is used as the final filtered damage signal, that is, the denoised signal, and its denoising effect is as follows: Figure 5 shown.

Claims

1. A method for denoising a rail surface damage magnetic leakage signal, characterized in that: The following steps are involved: S1: A single magnetic sensor is used as a single channel, and multiple magnetic sensors are used as multiple channels to identify the detection signal to form a multi-channel detection signal. At the same time, the damaged signal and the non-damaged signal in the multi-channel detection signal are distinguished. The n-order complementary integrated empirical mode decomposition CEEMD is performed on any damaged signal that needs to be filtered to obtain the decomposed modes IMF1, IMF2, ..., IMF n and the residual component RES; S2: Calculate the kurtosis values ​​of all modes, and take the mode with the largest kurtosis value as the reference mode IMF m ; S3: Perform Hilbert transform on all modes to obtain the corresponding envelope signal and envelope energy; S4: Calculate the benchmark modal IMF m The envelope signal and other modal IMFs i The correlation coefficient of the envelope signal is used to calculate the reference modal IMF m The envelope energy of and other modal IMFs i The correlation coefficient of the envelope energy, where i = 1, 2, ..., n, if the two calculated correlation coefficients are greater than the threshold, then it is considered that other modal IMF i If there is a damage component, if one or two correlation coefficients are less than or equal to the threshold, then the other modal IMF is considered i Contains no damaging ingredients; S5: Select any non-impaired signal as the auxiliary noise signal, perform n-order CEEMD on the auxiliary noise signal, and obtain the decomposed modal IMF'1, IMF'2, ..., IMF' n and the residual component RES'; S6: Subtract the modes after decomposition of the auxiliary noise signal in S5 from the modes with damage components identified in S4 to obtain filtered modes, and delete the modes without damage components identified in S4 and the residual component RES; S7: summing the modes filtered in S6 to obtain a reconstructed filtered impairment signal.

2. The method for denoising the rail surface damage magnetic leakage signal according to claim 1, characterized in that: The calculation method of envelope signal and envelope energy in S3 is: given a signal x(t), its envelope signal A(t) and envelope energy E(t) are respectively: E(t)=A 2 (t)=x 2 (t)+H 2 [x(t)] H[] represents Hilbert transform.

3. The method for denoising the rail surface damage magnetic leakage signal according to claim 1, characterized in that: The calculation method of the impairment mode of the impairment signal after filtering in S6 is: the modes of the impairment component signal CEEMD in S1 are IMF1, IMF2, ..., IMF n and the residual component RES, and CEEMD is performed on the auxiliary noise signal. The decomposition order n is the same as the order of CEEMD of the damaged signal in S1, and the decomposed modes are IMF'1, IMF'2, ..., IMF' n and the residual component RES'; From IMF1 to IMF n Deal with each case individually, when dealing with IMF j If the IMF j The envelope signal and envelope energy of IMF m If the correlation coefficients of the envelope signal and envelope energy are both greater than the threshold, then IMF" j =IMF j -IMF' j , to the IMF” j As a filtered mode, if one or both correlation coefficients are less than or equal to the threshold, the modal IMF without the impairment component will be j delete.

4. A denoising system for rail surface damage magnetic leakage signal, characterized in that: It includes a damage signal processing module, a kurtosis value calculation module, a Hilbert transform module, a judgment module, an auxiliary noise signal processing module, a filtering module, and a summing module; The damage signal processing module is used to use a single magnetic sensor as a single channel, and multiple magnetic sensors as multiple channels to identify the detection signal to form a multi-channel detection signal, and distinguish the damage signal and the non-damage signal in the multi-channel detection signal. The n-order complementary integrated empirical mode decomposition CEEMD is performed on any damage signal that needs to be filtered to obtain the decomposed modes IMF1, IMF2, ..., IMF n and the residual component RES; The kurtosis calculation module is used to calculate the kurtosis values ​​of all modes, and the mode with the largest kurtosis value is taken as the benchmark mode IMF. m ; The Hilbert transform module is used to perform Hilbert transform on all modes to obtain the corresponding envelope signal and envelope energy; The judgment module is used to calculate the reference modal IMF m The envelope signal and other modal IMFs i The correlation coefficient of the envelope signal is used to calculate the reference modal IMF m The envelope energy of and other modal IMFs i The correlation coefficient of the envelope energy, where i = 1, 2, ..., n, if the two calculated correlation coefficients are greater than the threshold, then it is considered that other modal IMF i If there is a damage component, if one or two correlation coefficients are less than or equal to the threshold, then the other modal IMF is considered i Contains no damaging ingredients; The auxiliary noise signal processing module is used to select any non-impaired signal as the auxiliary noise signal, perform n-order CEEMD on the auxiliary noise signal, and obtain the decomposed modal IMF'1, IMF'2, ..., IMF' n and the residual component RES'; The filtering module is used to subtract the modes after decomposition of the auxiliary noise signal in the auxiliary noise signal processing module from the modes with damage components identified in the judgment module to obtain the filtered modes, and delete the modes without damage components identified in the judgment module and the residual component RES; The summing module is used to sum the modes filtered in the filtering module to obtain a reconstructed filtered damage signal.

5. The denoising system for rail surface damage magnetic leakage signal according to claim 4, characterized in that: The calculation method of envelope signal and envelope energy in Hilbert transform module is: given signal x(t), its envelope signal A(t) and envelope energy E(t) are respectively: E(t)=A 2 (t)=x 2 (t)+H 2 [x(t)] H[] represents Hilbert transform.

6. The denoising system for rail surface damage magnetic leakage signal according to claim 4, characterized in that: The calculation method of the impairment mode of the impairment signal after filtering in the filtering module is as follows: the modes of the impairment component signal CEEMD in the impairment signal processing module are IMF1, IMF2, ..., IMF n and the residual component RES, and CEEMD is performed on the auxiliary noise signal. The decomposition order n is the same as the order of CEEMD of the impairment signal in the impairment signal processing module. The decomposed modes are IMF'1, IMF'2, ..., IMF' n and the residual component RES'; From IMF1 to IMF n Deal with each case individually, when dealing with IMF j If the IMF j The envelope signal and envelope energy of IMF m If the correlation coefficients of the envelope signal and envelope energy are both greater than the threshold, then IMF" j =IMF j -IMF' j , to the IMF” j As a filtered mode, if one or both correlation coefficients are less than or equal to the threshold, the modal IMF without the impairment component will be j delete.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to claim 1 to claim 3 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to claim 1 to claim 3 are implemented.

Images