Discrete wavelet steel rail damage acoustic emission signal detection method based on filtering parameter adaptive search

Abstract

The invention discloses a discrete wavelet steel rail damage acoustic emission signal detection method based on filtering parameter adaptive search, which comprises the following steps of: firstly, preprocessing acoustic emission data acquired by a sensor, and removing a pure noise segment in the signal to relieve the subsequent operation pressure; then, an evaluation function combining a reduction degree index and a sparseness index is designed, an evaluation index of a filtering parameter represented by each sparrow is calculated in combination with discrete wavelets, and a filtering parameter enabling the evaluation index to be minimum is searched by adopting a self-adaptive sparrow search method, namely an optimal filtering parameter group; and finally, filtering the original signal based on the optimal filtering parameter group, extracting a high-frequency component, obtaining the time-frequency characteristic of the high-frequency component by adopting short-time Fourier transform, and proposing a steel rail damage judgment condition based on the dual-band time-frequency characteristic, which can judge whether the damage exists or not through the time-frequency characteristic and calculate the moment when the damage is captured by a sensor. The problem that a traditional method generally depends on system parameters and prior information is solved.

Description

Technical Field

[0001] The present invention belongs to the field of non-destructive testing of rail damage, and relates to a method for detecting rail damage signals, specifically to a method for detecting acoustic emission signals of rail damage based on discrete wavelet with adaptive search of filtering parameters. Background Art

[0002] With the continuous progress of rail transit technology, the operating pressure of the track is gradually increasing, and the accompanying accident risks are also significantly increasing. To ensure the safety and stability of railway operation, relevant researchers have proposed various detection means for rail damage cracks. Among them, acoustic emission detection, as a non-destructive testing method, has shown great potential in the field of railway safety maintenance due to its strong real-time monitoring ability, high sensitivity, non-destructive testing method, wide application range and other advantages.

[0003] However, the application of acoustic emission technology in the current field of railway damage detection is still restricted by several technical defects. Acoustic emission is a transient elastic wave phenomenon generated by the rapid release of local stress in materials, and its signal intensity is usually weak. In contrast, the wheel-rail contact noise during train operation is often significant, which leads to the acoustic emission signal being easily masked by the wheel-rail contact noise. At the same time, the irregularity of the rail geometry, especially the features such as corners, protrusions and depressions in the I-shaped geometry, will significantly change the propagation path of sound waves. During the signal propagation process, phenomena such as reflection, refraction and diffraction frequently occur, and together with the combined action of environmental factors, they jointly lead to the complexity and high uncertainty of the acoustic emission signal. Therefore, the current detection of rail damage acoustic emission signals generally highly depends on system parameters and prior information, lacking wide generality and universality.

[0004] In summary, there is an urgent need to develop a rail damage acoustic emission signal processing technology that is both widely applicable to practical applications and does not depend on specific parameters, so as to effectively filter out the interference of wheel-rail rolling noise and improve the accuracy and reliability of acoustic emission signal detection. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for detecting acoustic emission signals of rail damage based on discrete wavelet with adaptive search of filtering parameters. This method designs an evaluation function that combines the reduction index and the sparsity index to evaluate the search results of the adaptive sparrow search algorithm, and proposes a damage determination condition for dual-band time-frequency characteristics, which can dynamically adjust parameters according to signal characteristics and obtain the optimal filtering parameter group, accurately determine damage and calculate the damage time, and solves the problem of the general dependence of traditional methods on system parameters and prior information.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A discrete wavelet rail damage acoustic emission signal detection method based on adaptive search of filtering parameters, comprising the following steps:

[0008] Step 1: Load the acoustic emission signal containing damage cracks and noise, frame the data, judge whether it contains valid signals by calculating the energy of each frame, and retain the non-overlapping part in the frames with energy higher than the threshold as the reconstructed signal;

[0009] Step 2: Design a new adaptive sparrow search method to explore the optimal filtering parameter group suitable for discrete wavelet threshold filtering: Define the filtering parameter group to be searched as the sparrow position, construct an evaluation function based on the sparsity index and the restoration index, and combine the discrete wavelet filtering method to calculate the evaluation index of each sparrow in the sparrow group according to the evaluation function. This index is used to evaluate the quality of the sparrow position. The filtering parameters corresponding to the best sparrow position obtained through repeated iteration are the optimal filtering parameter group;

[0010] Step 3: Use the optimal filtering parameter group obtained in Step 2 for discrete wavelet threshold filtering to remove the noise in the original signal and obtain the high-frequency component of the filtered signal; perform short-time Fourier time-frequency analysis on the high-frequency component, and based on the time-frequency characteristics of the high-frequency component, judge whether there is damage according to the damage determination condition of the dual-band time-frequency characteristics and obtain the moment when the damage is captured by the acoustic emission sensor.

[0011] The present invention designs an evaluation function for evaluating the adaptive sparrow search, combines the adaptive sparrow search with the discrete wavelet and obtains the optimal filtering parameter group, and determines the rail damage in the acoustic emission signal by analyzing the high-frequency time-frequency characteristics of the filtered signal. Compared with the prior art, it has the following advantages:

[0012] 1. The present invention constructs an evaluation function combining sparsity and restoration, aiming to objectively evaluate the effect of the filtering parameter group represented by the sparrow position in the adaptive sparrow search algorithm for discrete wavelet threshold filtering. By dynamically adjusting the filtering parameters through the adaptive sparrow search method, the signal characteristics are effectively enhanced and the interference of wheel-rail rolling noise is significantly reduced.

[0013] 2. Based on the time-frequency analysis of the high-frequency component, the present invention proposes a damage signal determination condition based on the dual-band time-frequency characteristics, realizes the detection of the damage signal, and accurately obtains the moment when the damage signal appears.

[0014] 3. Compared with the traditional method, the present invention does not rely on system parameters and prior information, detects the acoustic emission signals of various types of damage under different environments and different test conditions, can adaptively adjust the filtering threshold according to the signal characteristics and successfully extract the damage characteristics, has wide applicability, and shows considerable application potential. Description of the Drawings

[0015] Figure 1 This is the flowchart of the discrete wavelet rail damage acoustic emission signal detection method based on adaptive search of filtering parameters in the present invention;

[0016] Figure 2 are the original acoustic emission signal and its reconstructed signal;

[0017] Figure 3 are the time-frequency diagram of the original signal and the time-frequency diagram of its high-frequency components;

[0018] Figure 4 is the time-frequency feature diagram of the high-frequency component of the filtered signal with fixed filtering parameters (decomposition level 5, wavelet basis db8, filtering threshold 0.2);

[0019] Figure 5 are the time-frequency feature diagram of the high-frequency of the adaptive discrete wavelet filtered signal and the filtered signal diagram. Specific implementation manners

[0020] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings, but it is not limited thereto. Any modification or equivalent replacement of the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention shall be covered by the protection scope of the present invention.

[0021] The present invention provides a discrete wavelet rail damage acoustic emission signal detection method based on adaptive search of filtering parameters. First, the acoustic emission data collected by the acoustic emission sensor is preprocessed to remove the pure noise segment in the signal to reduce the subsequent operation pressure; then, an evaluation function combining the reduction index and the sparsity index is designed to evaluate the performance of the filtering parameter group represented by the sparrow position obtained by the adaptive sparrow algorithm for discrete wavelet threshold filtering. The filtering parameter group represented by the sparrow that makes the evaluation index take the minimum value obtained by the adaptive sparrow search method is the optimal filtering parameter group of the current signal; finally, the optimal filtering parameter group obtained by the search is used for filtering the original signal and extracting the high-frequency component of the filtered signal. The short-time Fourier transform is used for time-frequency analysis of the high-frequency component and the time-frequency feature is obtained. A rail damage determination condition based on the double-band time-frequency feature is proposed. This condition can judge whether there is damage and calculate the moment when the damage is captured by the sensor. As Figure 1 shown, the specific steps are as follows:

[0022] Step 1: Preprocess the signal collected by the acoustic emission sensor. Since the acoustic emission signal has a high sampling frequency and a large amount of data, in order to reduce the subsequent operation pressure, the noise segment without valid information in the signal needs to be removed. The acoustic emission data set is framed, and it is judged whether each frame contains a valid signal by calculating the energy of each frame. Only the non-overlapping part in the frame with energy higher than the threshold is retained as the reconstructed signal. The reconstructed signal is used for the iterative calculation in Step 2, which greatly reduces the operation time compared with the original signal. The specific steps are as follows:

[0023] Step 1.1: Collect the original signal:

[0024] The signal s(i) is a one-dimensional signal with a length of n obtained by sampling, where i = 1, 2, …, n, and can be expressed in the form of a one-dimensional vector:

[0025] S = [s(1) s(2) … s(n)] (1);

[0026] Step 1.2: Frame the original signal:

[0027] Frame the signal with a frame spacing of p, a frame length of q, and a total signal length of n. Divide the signal into m frames, where the number of frames m is the largest positive integer that satisfies .

[0028] Denote each frame as:

[0029] d im = [s((i - 1)×p + 1) s((i - 1)×p + 2) … s((i - 1)×p + q)], i = 1, 2, … m - 1 (2);

[0030] The length is q, and the non - overlapping part of each frame is:

[0031]

[0032] The length is p, and all the remaining signals form the last frame, that is:

[0033]

[0034] Then the original signal S can be further expressed as:

[0035]

[0036] Step 1.3: Reconstruct the signal according to the energy threshold:

[0037] Reconstruct the original signal. The reconstructed signal S d is the non - overlapping parts of all frames that meet the energy threshold recombined into a new signal, and can be expressed as:

[0038]

[0039] where {} represents the sequential recombination of all signals that meet the conditions, and γ is the energy threshold.

[0040] Step 2: Considering the significant impact of filter parameter selection on the final filtering result of the signal, the present invention proposes an improved adaptive sparrow search algorithm to explore the optimal filter parameter group. First, define the filter parameter group to be searched as the sparrow position; then construct an evaluation function based on the sparsity index and the restoration index to calculate the evaluation index of each sparrow position. The smaller the index, the better the sparrow position; update the sparrow position according to the iteration rule, and then recalculate the global evaluation index. When the best evaluation index no longer updates, output the current best filter parameter group. The specific steps are as follows:

[0041] Step 2-1: Define the sparrow position:

[0042] The main factors affecting the filtering effect of the discrete wavelet threshold filtering method include: the selection of the wavelet basis, the selection of the threshold, and the design of the threshold function. Different parameter selections have a significant impact on the final result. Based on the evaluation function constructed in the present invention, an adaptive sparrow search method is used to find the suitable wavelet basis and filtering threshold. Set the parameters to be optimized as 2-dimensional, namely the filtering threshold δ and the wavelet basis Ψ g of the group g. Use the simulated sparrow to represent the search parameters. The position X of the sparrow is expressed as:

[0043]

[0044] where, denotes rounding up the corresponding parameter, [x a,1 x a,2 is the position of the a-th sparrow, a = 1, 2, …, v, v is the total number of sparrows. The filter parameter group corresponding to the a-th sparrow is the filtering threshold δ a and the wavelet basis group Ψ ga .

[0045] Step 2-2: Obtain the wavelet decomposition coefficients and filtered signals corresponding to each group of filter parameters:

[0046] Since each sparrow represents a group of filter parameters, and for each group of filter parameters, wavelet decomposition is required to obtain the wavelet decomposition coefficients and the filtered signal residuals for calculating the evaluation index of this group of sparrows. The more sparrows there are, the greater the computational amount involved in this step. Therefore, in this step, the reconstructed signal is used to replace the original signal to participate in the iterative operation to reduce the operation time.

[0047] Perform discrete wavelet decomposition on the reconstructed signal S d :

[0048] S d = c1·ψ1 + c2·ψ2 + … + c M ·ψ M (8);

[0049] Among them, ψ1, ψ2, …, ψ M is a wavelet basis set with M wavelet bases, and c1, c2, …, c M are the discrete wavelet transform coefficients corresponding to this wavelet basis set. Denote Ψ = [ψ1 ψ2 … ψ M T , C Ψ = [c1 c2 … c M , then the above formula can be denoted as:

[0050] S d = C Ψ × Ψ (9);

[0051] Filter the wavelet transform coefficients. For wavelet coefficients less than the threshold δ, directly set them to 0. For wavelet coefficients greater than δ, set them to sgn(c jm )(c jm - δ), that is:

[0052]

[0053] where sgn() is the sign function. Denote as the wavelet transform coefficient matrix after filtering by the threshold δ corresponding to the wavelet basis set Ψ. Through the filtered signal

[0054]

[0055] Steps Two and Three. Construct an evaluation function and calculate the global evaluation index:

[0056] Construct an evaluation function for evaluating the quality of a group of sparrow positions:

[0057]

[0058] In the formula, || ||2 represents taking the 2-norm, is the reduction index, representing the energy ratio of the signal residual to the original signal, measuring the reduction degree of the filtered signal. The smaller this index value, the smaller the gap between the filtered signal and the original signal, and the more original features are retained; || ||0 represents taking the 0-norm, L is the length of the wavelet transform coefficient matrix , represents the number of non-zero elements in ​The ratio of length represents the sparsity of the wavelet coefficients. The larger the index is, the sparser the signal is and the better the filtering effect is. Generally, the better the filtering effect is, the less effective information of the original signal is retained, and the more the original signal is retained, the worse the filtering effect is. Therefore, a sparsity coefficient λ is added to adjust the proportion of the sparsity index in the overall index. The sparsity index can be adjusted as needed. The larger the index is, the better the filtering effect is, but the less effective information of the original signal is retained. Conversely, more effective information of the original signal is retained, but the filtering effect becomes worse.

[0059] Then the global evaluation index F X , that is, the evaluation index corresponding to all sparrows, can be calculated according to the following formula:

[0060]

[0061] Step 24: Update the sparrow position:

[0062] Update the sparrow position X according to the iteration rule. During each iteration, the parameter position is updated as follows:

[0063]

[0064] The superscripts represent the number of iterations of the corresponding parameters, for example Represents x a,b The t-th iteration result of is the evaluation index of the current (corresponding to the tth iteration) group a of sparrows, and are the current global best and worst evaluation indicators, respectively. is the current global optimal position, that is The corresponding set of x; is the current global worst position, that is A corresponding set of x; α is a random number with a mean of 0 and a variance of 1 that follows a normal distribution; K∈[-1,1] is a random number; ε is a very small constant used to avoid the error of division by 0.

[0065] Step 3: In order to extract the time-frequency features of the high-frequency part that is submerged in the noise signal, the optimal filter parameter group obtained in step 2 is first used for discrete wavelet threshold filtering to remove the noise in the original signal and obtain the high-frequency component of the filtered signal; then the high-frequency component is subjected to time-frequency analysis, and the damage signal is determined by the time-frequency features of the high-frequency part. Finally, a damage judgment standard based on dual-band time-frequency features and a method for calculating the time when damage occurs are proposed. The specific steps are as follows:

[0066] Step 31: Filter and decompose the signal into high-frequency components by frequency band:

[0067] Filter the original signal S according to the optimal filter parameter set obtained in Step 2 according to Step 22 (here, the original signal needs to be used because the reconstructed signal destroys the timing correspondence and it is inconvenient to calculate the damage time), and obtain the corresponding filtered signal. After that, for the filtered signal Perform high-pass filtering to obtain the high-frequency component of the filtered signal It can be expressed as:

[0068]

[0069] In the formula, is a high-pass filter, k is the frequency index, k c is the index corresponding to the filter cut-off frequency, and j represents the imaginary unit.

[0070] Step 32: Perform time-frequency analysis on the high-frequency component:

[0071] Calculate the power spectrum of the high-frequency component using the short-time Fourier transform :

[0072]

[0073] In the formula, || represents taking the absolute value of the corresponding parameter, and A(τ,k) represents the power spectrum at the time index τ and frequency index k; w(i - τT) is the window function, τ is the time index of the window, and T is the moving step of the window, which is used to determine the position of the window in the signal.

[0074] Step 33: Construction of the damage determination condition based on the dual-band time-frequency characteristics and calculation of the damage time:

[0075] When the obtained power spectrum has independent peaks greater than the threshold at the same time in two frequency bands, it is considered that a damage signal appears at this time, which can be expressed as the following condition:

[0076]

[0077] In the formula, μ is the judgment threshold of the peak, k0 is the frequency index corresponding to the interval between the two frequency bands, max() represents finding the global maximum, arg max x {A|B} represents finding the value of x that maximizes A under the condition B, and σ is a minimum value. The above formula can be described in words as the following three conditions:

[0078] Condition 1: There is a power spectrum peak greater than the threshold μ between the frequency index k c and k0;

[0079] Condition 2: Between the frequency index k0 and 1 / 2×f s (f sThere is a power spectral peak greater than the threshold μ between the sampling frequencies (for the sampling frequency).

[0080] Condition 3: The power spectral peaks in the above Conditions 1 and 2 should occur at the same time.

[0081] When the above three conditions are satisfied simultaneously, it is considered that an injury signal is captured by the acoustic emission sensor at the time index τ max The time index τ max is expressed as:

[0082] τ max = argmax τ {A(τ,k)} (18).

[0083] Example:

[0084] The following describes the specific implementation manner of the present invention in combination with the experimental data obtained from the test platform:

[0085] Execute Step 1: Load the acoustic emission signals S1 and S2 containing injury signals. The signal sampling frequency is 5 MHz (f s = 5×10 6 Hz), as shown in Figure 2 (a) and (b). Whether there is an injury signal cannot be directly seen from the original signal. The original signal contains pure noise components, and there is no injury signal in the pure noise part. To reduce the computational burden, the original signal is reconstructed, the pure noise part is removed, the length of one frame of the signal is q = 100, the frame spacing is p = 50, and the energy threshold is γ = 2. The reconstructed signal obtained according to Step 1 is shown in Figure 2 (c) and (d). The length of signal S1 is reduced from 4×10 6 to 0.6319×10 6 , and the length of signal S2 is reduced from 4.5×10 6 to 0.7398×10 6 . Reconstructing the signal greatly reduces the amount of data required for iteration in Step 2.

[0086] Execute Step 2: For each group of test signals, optimize and determine the optimal filter parameter group to overcome the limitation that fixed filter parameters cannot be generally adapted to all signals. Define the upper and lower limits of the parameters to be solved as x up = [1, 18], x down = [0.01, 0.01], that is, the value range of the threshold is δ ∈ [0.01, 1], and the value range of the wavelet basis group is g ∈ {1, 2, 3,..., 18}, where the wavelet bases Ψ1 to Ψ 10 correspond to db1 to db10 respectively, and Ψ 11 to Ψ 18They respectively correspond to sym1 to sym8. The population size is taken as v = 50, the maximum number of iterations is 100, the sparsity coefficient is taken as λ = 5, and another ε = 0.001 is taken. The algorithm steps are as follows:

[0087] (1) Initialize parameters: Set the population size to 50, the maximum number of iterations to 100, the lower bound of the parameters to [0.01, 0.01], and the upper bound of the parameters to [1, 18];

[0088] (2) Initialize evaluation indicators: Randomly generate a set of initial positions of sparrows in the parameter space, and calculate the global evaluation indicators according to steps two and three;

[0089] (3) Iteratively update parameters: Iteratively update the sparrow positions according to steps two and four;

[0090] (4) Recalculate evaluation indicators: Recalculate the global evaluation indicators according to steps two and three based on the updated sparrow positions;

[0091] (5) If the evaluation indicators converge or reach the maximum number of iterations, end the iteration and output the optimal solution, otherwise jump to step (3).

[0092] Calculate the optimal filter parameter groups corresponding to two groups of acoustic emission signals according to the above steps. The filter parameter for signal S1 is wavelet basis db9, and the filtering threshold δ = 0.186123; the filter parameter for signal S2 is wavelet basis db10, and the filtering threshold is δ = 0.143945.

[0093] Execute step three: Filter the original signal according to the optimal filter parameters calculated in step two according to step two two, then extract the high-frequency components, and perform time-frequency analysis on the high-frequency components to obtain the time-frequency characteristics of the high-frequency components of the filtered signal.

[0094] First, use discrete wavelet for filtering. The decomposition level is selected as level = 5. The filter parameter for signal S1 is wavelet basis db9, and the filtering threshold δ = 0.186123; the filter parameter for signal S2 is wavelet basis db10, and the filtering threshold is δ = 0.143945. Then perform high-pass filtering on the filtered signal to obtain the high-frequency detail components. Select the cut-off frequency as 0.8 MHz, that is, select k c as the index corresponding to 0.8 MHz; finally, perform short-time Fourier time-frequency analysis on the high-frequency components. Select the window length as 256 and the shift step as 128. Since the sampling frequency f s = 5 MHz, the sampling interval dt between data points = 1 / f s , then the moving step of the window function should be T = dt × 128. Select the threshold μ = 0.2 × 10 -3 in the judgment condition, calculate the power spectrum energy A(τ, k) corresponding to time and frequency, and plot it as a three-dimensional graph, such asFigure 5 as shown in (a) and (b).

[0095] Figure 5 In (a) and (b), p1 and p2 marked by red circles represent two independent wave peaks in the time-frequency diagram, that is, the time-frequency characteristics of the damage. The two wave peaks of signal S1 are p1: 0.4051 s, 937500 Hz, 0.0065 and p2: 0.4051 s, 1562500 Hz, 0.0055 respectively. It can be seen from the figure that the two wave peaks are independent wave peaks, both at the moment of 0.4051 s, and the peak values are greater than the threshold, meeting the determination conditions 1-3; the two wave peaks of signal S2 are p1: 0.2665 s, 937500 Hz, 0.0022 and p2: 0.2665 s, 1562500 Hz, 0.0017 respectively. It can be seen from the figure that the two wave peaks are independent wave peaks, both at the moment of 0.2665 s, and the peak values are greater than the threshold, meeting the determination conditions 1-3. Damage signals are detected in both segments of signals, and very obvious time-frequency characteristics are obtained.

[0096] Figure 5 (c) and (d) are the filtered signals. The damage time points calculated from the characteristic wave peaks in the time-frequency diagram indicated by the positions of the red dotted lines, that is Figure 5 for signal S1 in (c) is at 0.4051 s, Figure 5 for signal S2 in (d) is at 0.2665 s, indicating that the acoustic emission sensor captures the damage signal at this moment.

[0097] Compare Figure 5 the high-frequency time-frequency characteristic diagram of the adaptive discrete wavelet filtered signal with Figure 3 the time-frequency analysis diagram of the original signal. It can be seen that there are multiple peaks without obvious differences in the signal obtained by directly performing time-frequency analysis on the original signal, which cannot be used as an indicator for damage indication. Moreover, the high-frequency time-frequency characteristics of the original signal are completely submerged by the wheel-rail rolling noise and cannot be recognized, while the method in the present invention can effectively extract the high-frequency time-frequency characteristics submerged by the wheel-rail rolling noise.

[0098] Compare Figure 5 the high-frequency time-frequency characteristic diagram of the adaptive discrete wavelet filtered signal with Figure 4 the high-frequency time-frequency characteristic diagram of the fixed threshold filtering. It can be found that after the fixed threshold filtering process, signal S1 has relatively obvious independent wave peaks in two frequency bands at the same moment, and the signal characteristics are relatively obvious; however, the characteristics of signal S2 are poor, and there are multiple wave peaks distributed in a fan shape in each frequency band, and the specific moment when the damage occurs cannot be determined based on the high-frequency time-frequency characteristics. While the two signals after the adaptive discrete wavelet filtering in the present invention both show obvious independent wave peaks, and the signal characteristics are greatly improved compared with the fixed threshold filtering.

[0099] The above results confirm that the method proposed by the present invention can dynamically adjust parameters according to signal characteristics and environmental conditions, thereby effectively detecting the acoustic emission signals of rail acoustic damage from the noise background and determining the time when they are captured by the acoustic emission detector, providing a solid foundation for subsequent extraction and recognition of rail damage characteristics based on acoustic emission technology and large-scale automatic and intelligent detection of rail damage.

Claims

1. A discrete wavelet rail damage acoustic emission signal detection method based on adaptive search of filtering parameters, characterized in that The method includes the following steps: Step 1: Load the acoustic emission signal containing damage cracks and noise, frame the data, and determine whether it contains valid signals by calculating the energy of each frame. Retain the non-overlapping part in the frames with energy higher than the threshold as the reconstructed signal; Step 2: Design a new adaptive sparrow search method to explore the optimal filter parameter group suitable for discrete wavelet threshold filtering: Define the filter parameter group to be searched as the sparrow position, construct an evaluation function based on the sparsity index and the restoration index, and calculate the evaluation index of each sparrow in the sparrow group according to the evaluation function in combination with the discrete wavelet filtering method. This index is used to evaluate the quality of the sparrow position. The filter parameters corresponding to the best sparrow position obtained through repeated iteration are the optimal filter parameter group; Step 3: Use the optimal filter parameter group obtained in Step 2 for discrete wavelet threshold filtering to remove the noise in the original signal and obtain the high-frequency components of the filtered signal; Perform short-time Fourier time-frequency analysis on the high-frequency components. Based on the time-frequency characteristics of the high-frequency components, judge whether there is damage according to the damage determination condition of the dual-band time-frequency characteristics and obtain the moment when the damage is captured by the acoustic emission sensor.

2. The method for detecting discrete wavelet rail damage acoustic emission signals based on adaptive search of filtering parameters according to claim 1, wherein The specific steps of Step 1 are as follows: Step 1.1: Collect the original signal: The signal s(i) is a one-dimensional signal with a length of n obtained by sampling. i = 1, 2, …, n, and it is expressed in the form of a one-dimensional vector: S = [s(1) s(2) … s(n)] (1); Step 1.2: Frame the original signal: Frame the signal with a frame spacing of p, a frame length of q for each frame, and a total signal length of n. Divide the signal into m frames, where the number of frames m is the largest positive integer that satisfies ; Denote each frame as: d im = [s((i - 1)×p + 1)s((i - 1)×p + 2)…s((i - 1)×p + q)], i = 1, 2, … m - 1 (2); The length is q, then the non-overlapping part of each frame is: The length is p, and all the remaining signals form the last frame, that is: Then the original signal S is further expressed as: Step 1.3: Reconstruct the signal according to the energy threshold: Reconstruct the original signal, and the reconstructed signal is S d For all frames that meet the energy threshold, take the non-overlapping parts and recombine them into a new signal, which is expressed as: Among them, {} means reorganize all the signals that meet the conditions in sequence, and γ is the energy threshold.

3. The method for detecting discrete wavelet rail damage acoustic emission signals based on adaptive search of filtering parameters according to claim 2, characterized in that The specific steps of Step 2 are as follows: Step 2.1: Define the sparrow position: The parameters to be optimized are set to two dimensions, namely the filtering threshold δ and the wavelet basis Ψ g For the group g, the simulated sparrows are used to represent the search parameters, and the position X of the sparrows is expressed as: Among them, represents rounding up the corresponding parameter, [x a,1 x a,2 is the position of the a-th sparrow, where a = 1, 2, …, v and v is the total number of sparrows. The filtering parameter group corresponding to the a-th sparrow is the filtering threshold δ a and the wavelet basis group Step 2.2: Obtain the wavelet decomposition coefficients and filtered signals corresponding to each group of filter parameters: Reconstruct the signal S d Perform discrete wavelet decomposition: S d = c1·ψ1 + c2·ψ2 + … + c M ·ψ M (8); Among them, ψ1, ψ2, …, ψ M is a wavelet basis set with M wavelet bases, c1, c2, …, c M are the discrete wavelet transform coefficients corresponding to the wavelet basis set. Denote Ψ = [ψ1 ψ2 … ψ M T , C Ψ = [c1 c2 … c M , then the above formula is denoted as:​ S d = C Ψ × Ψ(9); Filter the wavelet transform coefficients. For wavelet coefficients less than the threshold δ, directly set them to 0. For wavelet coefficients greater than δ, set them to sgn(c jm )(c jm - δ), that is: where sgn() is the sign function, denoted as the wavelet transform coefficient matrix corresponding to the wavelet basis set Ψ after filtering by the threshold δ, and the filtered signal is resynthesized through the resynthesized filtered signal Step 2.3: Construct an evaluation function and calculate the global evaluation index: Construct an evaluation function for evaluating the quality of a group of sparrow positions: where, || ||2 represents the calculation of the 2-norm, is the restoration degree index, representing the energy ratio of the signal residual to the original signal, || ||0 represents the calculation of the 0-norm, and L is the length of the wavelet transform coefficient matrix ; represents the ratio of the number of non-zero elements in to the length of Then the global evaluation index F X , that is, the evaluation index corresponding to all sparrows, is calculated according to the following formula: Step 2.4: Update the sparrow position: Update the sparrow position X according to the iteration rule. During each iteration, the parameter position is updated as follows: Among them, the superscript represents the iteration number of the corresponding parameter; is the evaluation index of the current a-th group of sparrows, and are the current global best and worst evaluation indexes respectively, is the current global optimal position, that is, the corresponding group of x; is the current global worst position, that is, the corresponding group of x; α is a random number obeying the normal distribution with a mean of 0 and a variance of 1; K ∈ [-1, 1] is a random number; ε is an extremely small constant used to avoid the error of division by zero.

4. The method for detecting discrete wavelet rail damage acoustic emission signals based on adaptive search of filtering parameters according to claim 3, characterized in that The specific steps of Step 3 are as follows: Step 3.1: Filter and decompose the signal into high-frequency detail components by frequency band: Filter the original signal S according to the optimal filter parameter set obtained in step two according to step two-two and obtain the corresponding filtered signal After that, for the filtered signal Perform high-pass filtering to obtain the high-frequency detail component of the filtered signal It is expressed as: wherein, is a high-pass filter, k is the frequency index, and k c is the index corresponding to the filter cut-off frequency, and j represents the imaginary unit; Step 3.2: Perform time-frequency analysis on the high-frequency components: Calculating the power spectrum of high-frequency components using the short-time Fourier transform : In the formula, | | means taking the absolute value of the corresponding parameter, A(τ, k) represents the power spectrum at the time index τ and the frequency index k; w(i - τT) is the window function, τ is the time index of the window, and T is the moving step of the window; Step 3.3: Construction of the damage determination condition based on the dual-band time-frequency characteristics and calculation of the damage moment: When the obtained power spectrum has independent peaks greater than the threshold at two frequency segments at the same moment, it is considered that an AE sensor captures a damage signal at the time index τ max and is expressed as the following condition: where μ is the judgment threshold of the wave crest, k0 is the frequency index corresponding to the interval between two frequency segments, max() represents finding the global maximum, and arg max x {A|B} represents finding the value of x that maximizes A under condition B, and σ is a minimum value.

5. The method for detecting discrete wavelet rail damage acoustic emission signals based on adaptive search of filtering parameters according to claim 4, wherein The time index τ max is expressed as: τ max = argmax τ {A(τ, k)} (18).

Images