Rail structure health monitoring method based on double-layer filtering and comprehensive health index

Abstract

The application discloses a kind of based on double-layer filtering and comprehensive health index steel rail structure health monitoring method, the method comprises the following steps: step one: according to the spectrum energy distribution of the unique electromagnetic acoustic emission signal, the original acoustic emission data set is filtered, and interference signal is excluded;Step two: the correlation index between each group of signals is calculated one by one, the correlation degree of different signals is measured, and noise signals with the same spectrum energy distribution as the electromagnetic acoustic emission signal are excluded;Step three: two features of barycenter frequency and Escort-Tsallis entropy are extracted from each group of electromagnetic acoustic emission signals, and a comprehensive health index is constructed accordingly to achieve accurate monitoring of the health of the steel rail structure.The application has fast operation rate, accurate monitoring and accurate crack classification.It has high social significance and economic value in the field of real-time monitoring of high-speed rail structure health.

Description

TECHNICAL FIELD

[0001] The present application relates to a rail crack signal processing and structural health monitoring method, in particular to a rail structural health monitoring method based on double-layer filtering and comprehensive health index. BACKGROUND

[0002] Rail plays an indispensable role in railway transportation, supporting the smooth operation of various trains. However, due to high-frequency, high-intensity load, and harsh operating environment, rail is prone to crack and even break in extreme cases, leading to railway safety accidents. According to the statistics of the United States Federal Railway Administration in 2016, about 30% of the recorded railway safety accidents in the past few decades are related to rail. Therefore, structural health monitoring of rail has great significance for ensuring safe, stable and reliable operation of railway.

[0003] At present, the main technologies for structural health monitoring of rail include eddy current detection, magnetic force detection, ultrasonic detection and vibration detection. The principles and characteristics of the above detection methods are different, but they can all achieve basic rail structural health monitoring. However, most of these methods can only be operated offline, and must be checked regularly, which cannot realize real-time structural health monitoring of rail. In addition, the above detection methods are often limited by the shape of the rail, and the monitoring speed and accuracy cannot meet the safety requirements of the actual railway system. Acoustic emission detection technology utilizes the sudden energy release triggered by the train passing through the rail crack, has the advantages of high detection sensitivity, strong real-time performance and no geometric shape limitation, and is gradually applied to real-time structural health monitoring of rail. However, in actual train operation, the wheel and rail are in rolling contact at all times, thereby generating a large amount of high-amplitude noise, and the crack information signal collected by the acoustic emission technology is drowned in the wheel-rail rolling noise, affecting the accuracy of rail structural health monitoring based on acoustic emission technology. Electromagnetic acoustic emission technology combines the advantages of electromagnetic detection and acoustic emission detection, the signal source comes from a high-power pulse emission device, realizes the transition from passive detection to active detection, and can adjust the amplitude of the crack information signal. In addition, electromagnetic acoustic emission technology allows the detection system to be loaded on demand, and can highly adapt to monitoring applications in various complex environments. However, most electromagnetic acoustic emission detection technologies are based on deep learning algorithms, which require a large amount of prior information, resulting in low monitoring speed and accuracy. Therefore, it is of great significance to propose a high-precision rail structural health monitoring method without prior information and capable of effectively eliminating complex noise interference. SUMMARY

[0004] In order to solve the problems of slow speed and low efficiency of the traditional rail crack signal real-time monitoring method, the application provides a rail structure health monitoring method based on double-layer filtering and comprehensive health index. The method is based on electromagnetic acoustic emission technology, combined with a double-layer filtering method, to exclude a large number of interference signals and noise signals generated during the fatigue crack propagation of the rail, calculate the barycenter frequency and Escort-Tsallis entropy of each group of electromagnetic acoustic emission signals, represent the crack change information contained in the signal, and construct a comprehensive health index according to the two normalized features, so as to accurately track the fatigue crack propagation process of the rail and realize accurate monitoring of the rail structure health.

[0005] The purpose of the application is realized by the following technical solutions:

[0006] A rail structure health monitoring method based on double-layer filtering and comprehensive health index, comprising the following steps:

[0007] Step one: according to the frequency spectrum energy distribution of the electromagnetic acoustic emission signal, the original acoustic emission data set is filtered to exclude interference signals;

[0008] Step two: calculate the comprehensive correlation index between each group of signals one by one to measure the correlation degree of different signals and exclude noise signals with the same frequency spectrum energy distribution as the electromagnetic acoustic emission signal;

[0009] Step three: extract the barycenter frequency and Escort-Tsallis entropy of each group of electromagnetic acoustic emission signals, and construct a comprehensive health index accordingly to realize accurate monitoring of the rail structure health.

[0010] Compared with the prior art, the application has the following advantages:

[0011] 1. The application uses four-layer wavelet packet decomposition to obtain the frequency band energy distribution information of the electromagnetic acoustic emission signal, and determines the signals generated during the fatigue crack propagation process of the rail according to the information, and excludes various interference noises. In addition, a comprehensive correlation index is constructed to measure the correlation between multiple groups of signals, and signals meeting certain threshold standards are selected as the final electromagnetic acoustic emission signals, and noise signals with the same frequency band energy distribution as the electromagnetic acoustic emission signals are filtered out, thereby increasing the filtering accuracy.

[0012] 2. The application proposes a comprehensive health index, which adaptively fuses the barycenter frequency and Escort-Tsallis entropy extracted from the electromagnetic acoustic emission signal to obtain a feature that can accurately represent the fatigue crack propagation information of the rail. In addition, according to the change of the comprehensive health index, the rail structure health change process can be divided into three different stages, and real-time monitoring of the rail structure health is realized.

[0013] 3、The existing rail health monitoring method is often based on deep learning algorithm, which needs a large amount of prior information as data label, resulting in low health monitoring efficiency and huge time consumption. At the same time, a large amount of interference noise will be generated in the actual expansion process of rail fatigue crack, which will annihilate the effective damage information, so that the monitoring accuracy cannot meet the requirements of railway safe transportation. The improved double-layer filtering and comprehensive health index algorithm can effectively distinguish the electromagnetic acoustic emission signals containing damage information from other interference noise signals, and can accurately track the rail fatigue crack expansion process, improve the monitoring accuracy, and meet the requirements of real-time health monitoring of rails in the actual railway transportation process.

[0014] 4、The operation rate of the present application is fast, the monitoring is accurate, and the crack classification is accurate. In the field of real-time monitoring of high-speed rail structure health, it has high social significance and economic value. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The flow chart of the rail structure health monitoring method based on double-layer filtering and comprehensive health index of the present application.

[0016] Figure 2 The time-frequency diagram of the electromagnetic acoustic emission signal.

[0017] Figure 3 The energy distribution diagram of each frequency band of the electromagnetic acoustic emission signal.

[0018] Figure 4 The comprehensive correlation index and threshold value diagram of the electromagnetic acoustic emission signal.

[0019] Figure 5 The second layer filtering determination result diagram.

[0020] Figure 6 The rail structure health monitoring result diagram. DETAILED DESCRIPTION

[0021] The technical solutions of the present application will be further described below in combination with the drawings, but are not limited thereto. Any modification or equivalent replacement of the technical solutions of the present application without departing from the spirit and scope of the present application shall be covered in the protection scope of the present application.

[0022] The present application provides a rail structure health monitoring method based on double-layer filtering and comprehensive health index. First, according to the specific frequency spectrum energy distribution of the electromagnetic acoustic emission signal, a plurality of acoustic emission signals generated during the rail fatigue crack expansion are subjected to first layer filtering, most of the interference signals are excluded, and an initial signal set is obtained Then, the Pearson correlation coefficients between each group of signals in the formula are calculated one by one On this basis ​Each signal comprehensive correlation index Measure The correlation degree of different signals, set the correlation coefficient threshold θ, perform the second layer filtering, exclude the noise signals with the same frequency spectrum energy distribution as the electromagnetic acoustic emission signals, and obtain the electromagnetic acoustic emission signal set Finally, from The barycenter frequency and Escort-Tsallis entropy of each group of electromagnetic acoustic emission signals are extracted to represent the crack change information contained in the signals, and a comprehensive health index is constructed according to the two normalized features According to The fatigue crack propagation process of the steel rail can be accurately tracked, and the steel rail structure health can be accurately monitored. As shown in Figure 1 The specific steps are as follows:

[0023] Step one: according to the specific frequency spectrum energy distribution of the electromagnetic acoustic emission signal, the original acoustic emission data set is filtered to exclude interference signals, and the specific steps are as follows:

[0024] Step one: load a group of electromagnetic acoustic emission signals S, and use four-layer wavelet packet decomposition method to divide S spectrum into 16 sub-bands:

[0025]

[0026] Among them, represents the j0th node of the i0th layer, i0∈{1,..,4}, j0∈{0,..,15};

[0027] Step two: calculate the energy proportion of each sub-band of S spectrum

[0028]

[0029]

[0030] Among them, represents the energy of the j0th node of the 4th layer, represents the j0th node of the 4th layer, represents wavelet coefficient, n is the length of the electromagnetic acoustic emission signal, represents the energy proportion of the j0th sub-band;

[0031] Step three, select the sub-band with energy proportion greater than 5%, and mark the sub-band number that meets the condition as J S :

[0032]

[0033] Step 14: Load all acoustic emission signals obtained during the rail fatigue fracture process N0 represents the total number of N0 groups of acoustic emission signals obtained during the fatigue fracture of the rail. Repeat steps 1 to 13 to obtain The sub-band number of each signal energy greater than 5% is J Z , according to the obtained electromagnetic acoustic emission signal spectrum energy distribution, determine whether each group of signals belongs to the electromagnetic acoustic emission signal:

[0034]

[0035] If the calculation result R is 1, the signal is preliminarily judged to be a suspected electromagnetic acoustic emission signal, and the initial signal set obtained is marked as N1 represents that the initial signal set has N1 groups of acoustic emission signals. Otherwise, it is an interference signal. Excluded.

[0036] Step 2: Calculate the comprehensive correlation index between each group of signals one by one to measure the degree of correlation between different signals and exclude noise signals with the same spectrum energy distribution as the electromagnetic acoustic emission signal. The specific steps are as follows:

[0037] Step 21: Obtain the initial signal set Then calculate one by one The correlation coefficient between each signal and other signals forms a correlation coefficient matrix

[0038]

[0039] in, represents the Pearson correlation coefficient between the i1-th group signal s1(i1) and the j-th group signal s1(j1), calculated as follows:

[0040]

[0041] Step 22: Based on calculate Comprehensive correlation index of each signal in

[0042]

[0043] Step 2 and 3: Calculate After that, it can be used to accurately measure the degree of correlation between different signals in the initial data set:

[0044]

[0045] in, Indicates based on the electromagnetic acoustic emission signal judgment result, i2∈{1,..,N1},θ is an adaptive correlation coefficient threshold, usually taking a value of 0.2; if the calculation result is 1, the correlation coefficient between the signal and other signals in the signal set is high, and it can be determined that the signal is an electromagnetic acoustic emission signal, otherwise, the signal is weakly correlated or not correlated with other signals in the signal set, and is determined as a noise signal, and is excluded from the initial signal set N2 represents a total of N2 groups of signals in the finally obtained electromagnetic acoustic emission signal set.

[0046] Step three: two features of barycenter frequency and Escort-Tsallis entropy are extracted from each group of electromagnetic acoustic emission signals, and a comprehensive health index is constructed accordingly to realize accurate monitoring of steel rail structure health, and the specific steps are as follows:

[0047] Step three one: according to the fast Fourier transform method, each signal in the electromagnetic acoustic emission signal set is converted from time domain to frequency domain to obtain frequency fk and frequency spectrum Y FFT :

[0048]

[0049]

[0050] Wherein, fs is the signal sampling rate, p∈{1,..,M},M represents that each group of electromagnetic acoustic emission signals contains M sampling points, FFT(·) represents fast Fourier transform operation, s2(i3) represents the i3th signal in the electromagnetic acoustic emission signal set

[0051] Step three two: according to the obtained frequency and frequency spectrum information, the barycenter frequency FC of each electromagnetic acoustic emission signal is calculated:

[0052]

[0053] Step three three: the Escort-Tsallis entropy ETE of the signal is calculated:

[0054]

[0055] Wherein, b is a normal number, usually taking 1, q represents a non-extended parameter, I represents the total number of possible configurations, and p(i5) is the relevant probability;

[0056] Step three four: the FC and ETE of each group of signals in the electromagnetic acoustic emission data set are calculated one by one, and the results are marked as​​​ and Then the two are normalized and quantified uniformly to avoid errors caused by amplitude differences, and a comprehensive health index is constructed based on the two normalized features.

[0057]

[0058] in, represents the comprehensive health index CHI of the i6th signal, i6∈{1, .., N2}, N(·) represents the normalization operation, and τ is a protection threshold that can ensure that CHI reflects the health status change information of the rail during the crack initiation stage;

[0059] Step 35: Based on It can accurately track the growth process of rail fatigue cracks and realize real-time monitoring of rail structural health. In addition, based on the change trend, it can also divide the growth of rail fatigue cracks into three health stages: initiation, growth and fracture.

[0060] Example:

[0061] Execute step 1: load a set of electromagnetic acoustic emission signals S, whose time-frequency information is as follows Figure 2 As shown, the S spectrum is divided into 16 sub-bands using the four-layer wavelet packet decomposition method. Calculate the energy proportion of each sub-band The energy distribution of each frequency band of electromagnetic acoustic emission signal is as follows: Figure 3 As shown, the sub-bands with energy proportion greater than 5% are screened out, and the sub-bands obtained by screening are marked as J S , at this time J S =2, 3, 4; Next, load all the acoustic emission signals obtained during the rail fatigue fracture process At this time, N0=30354, follow the same steps and filter out The sub-bands whose signal energy accounts for more than 5% in each group are marked as J. z According to the obtained electromagnetic acoustic emission signal spectrum energy distribution, we can judge whether each group of signals belongs to electromagnetic acoustic emission signals, and preliminarily judge that the signal is a suspected electromagnetic acoustic emission signal. The initial signal set obtained is marked as Otherwise, it is an interference signal. Finally, 4669 groups of acoustic emission signals are obtained, that is, N1 = 4669;

[0062] Execute step 2: calculate the initial signal set one by one The correlation coefficient between each signal and other signals forms a correlation coefficient matrix Then based on calculate Comprehensive correlation index of each signal in At this time, the correlation coefficient threshold θ is set to 0.2. The comprehensive correlation index of each signal in Figure 4 As shown; the correlation strength of each signal is determined according to the threshold θ. If the signal is If the CCM between other signals in the initial signal set is higher than the set threshold, the signal can be determined to be an electromagnetic acoustic emission signal, otherwise it is determined to be a noise signal. The final judgment result is as follows Figure 5 As shown in the figure, a total of 48 groups of interference noise signals are filtered out to obtain the electromagnetic acoustic emission signal set N2=4621;

[0063] Execute step three: First, according to the fast Fourier transform method, the electromagnetic acoustic emission signal is collected Each signal is converted from the time domain to the frequency domain to obtain the frequency fk and spectrum Y FFT Then, based on the obtained frequency and spectrum information, the center of gravity frequency and Escort-Tsallis entropy are extracted from each set of electromagnetic acoustic emission signals to characterize the crack change information contained in the signal, which are recorded as and right and Normalization is performed to unify the quantization to avoid errors caused by amplitude differences. Next, a comprehensive health index is constructed based on the two normalized features. Then to Perform smoothing filtering; in addition, according to The change of rail structure health can also be divided into three different stages, namely crack initiation, expansion and fracture. The final monitoring structure is as follows: Figure 6 As shown, this shows that the method proposed in the present invention can accurately track the growth process of rail fatigue cracks and realize real-time monitoring of rail structural health.

Claims

1. A rail structure health monitoring method based on double-layer filtering and comprehensive health index, characterized by The method comprises the following steps: Step 1: Based on the unique spectrum energy distribution of electromagnetic acoustic emission signals, filter the original acoustic emission data set to eliminate interference signals. The specific steps are as follows: Step 1: Load a set of electromagnetic acoustic emission signals S and use the four-layer wavelet packet decomposition method to divide the S spectrum into 16 sub-bands: in, represents the j0th node in the i0th layer, i0∈{1,..,4}, j0∈{0,..,15}; Step 1 and 2: Calculate the energy proportion of each sub-band of the S spectrum in, represents the energy of the j0th node in the 4th layer, represents the j0th node in the 4th layer, represent The wavelet coefficients of n are the length of the electromagnetic acoustic emission signal, Represents the energy proportion of the j0th sub-band; Step 13: Filter the sub-bands whose energy accounts for more than 5% and mark the sub-bands that meet the conditions as J S : Step 14: Load all acoustic emission signals obtained during the rail fatigue fracture process N0 represents the total number of N0 groups of acoustic emission signals obtained during the fatigue fracture of the rail. Repeat steps 1 to 13 to obtain The sub-band number of each signal energy greater than 5% is J Z , according to the obtained electromagnetic acoustic emission signal spectrum energy distribution, determine whether each group of signals belongs to the electromagnetic acoustic emission signal: If the calculation result R is 1, the signal is preliminarily judged to be a suspected electromagnetic acoustic emission signal, and the initial signal set obtained is marked as N1 represents that the initial signal set has N1 groups of acoustic emission signals. Otherwise, it is an interference signal. Excluded; Step 2: Calculate the comprehensive correlation index between each group of signals one by one to measure the degree of correlation between different signals and exclude noise signals with the same spectrum energy distribution as the electromagnetic acoustic emission signal. The specific steps are as follows: Step 21: Obtain the initial signal set Then calculate one by one The correlation coefficient between each signal and other signals forms a correlation coefficient matrix in, represents the Pearson correlation coefficient between the i1-th group signal s1(i1) and the j-th group signal s1(j1); Step 22: Based on calculate Comprehensive correlation index of each signal in Step 2 and 3: Calculate After that, it can be used to accurately measure the degree of correlation between different signals in the initial data set: in, Indicates based on The electromagnetic acoustic emission signal judgment result, i2∈{1,..,N1}, θ is the adaptive correlation coefficient threshold; if the calculation result If the correlation coefficient between the signal and other signals in the signal set is 1, the signal is judged to be an electromagnetic acoustic emission signal. Otherwise, the signal is weakly correlated or uncorrelated with other signals in the signal set and is judged to be a noise signal. Finally, under the interference of complex noise, the electromagnetic acoustic emission signal set is obtained. N2 means that the final electromagnetic acoustic emission signal set has N2 groups of signals; Step 3: Extract the center of gravity frequency and Escort-Tsallis entropy from each set of electromagnetic acoustic emission signals, and construct a comprehensive health index based on them to achieve accurate monitoring of rail structure health. The specific steps are as follows: Step 31: According to the fast Fourier transform method, the electromagnetic acoustic emission signal is collected Each signal is converted from the time domain to the frequency domain to obtain the frequency fk and spectrum Y FFT : Where fs is the signal sampling rate, p∈{1,..,M}, M means that each set of electromagnetic acoustic emission signals contains M sampling points, FFT(·) represents the fast Fourier transform operation, s2(i3) represents the electromagnetic acoustic emission signal set The i3th signal in ; Step 32: Calculate the center of gravity frequency FC of each electromagnetic acoustic emission signal based on the obtained frequency and spectrum information: Step 3: Calculate the Escort-Tsallis entropy ETE of the signal: Where b is a positive constant, q represents a non-extensive parameter, I represents the total number of possible configurations, and p(i5) is the associated probability; Steps 3 and 4: Calculate the electromagnetic acoustic emission data sets one by one The FC and ETE of each signal group in the result are marked as and Then the two are normalized and quantified uniformly to avoid errors caused by amplitude differences, and a comprehensive health index is constructed based on the two normalized features. in, represents the comprehensive health index CHI of the i6th signal, i6∈{1,..,N2}, N(·) represents the normalization operation, and τ is a protection threshold; Step 35: Based on It can accurately track the growth process of rail fatigue cracks and realize real-time monitoring of rail structural health. In addition, based on the change trend, the growth of rail fatigue cracks can be divided into three health stages: initiation, growth and fracture.

Images