Wave packet self-identification method

Abstract

The invention discloses a wave packet self-identification method, which considers wave packet frequency dispersion and multi-mode compensation at the same time, and can realize analysis of various ultrasonic detection signals. The method comprises the following steps: firstly, constructing an over-complete dictionary for the waveform of a target wave packet, and considering the frequency dispersion characteristics of waves; then, on the basis of a Bayesian reasoning method, effective wave packets in the aliasing wave packets are recognized and reconstructed, and automatic calibration of the wave packets with the specified form in the detection signals is achieved; and finally, determining the physical meanings of different wave packets in combination with the arrival sequence of the wave packets. Due to the self-recognition characteristic of the method, rapid analysis of massive signals can be dealt with, so that a large number of repeated ultrasonic tests are realized, and the reliability and the precision of a final result are improved. The wave packet self-identification method can be popularized to various physical test signals of different detection objects.

Description

Technical Field

[0001] The present invention belongs to the field of damage monitoring, and particularly relates to a wave packet self-identification method. Background Art

[0002] Ultrasonic guided waves are a non-destructive detection method that propagates along structures and can cover a large range, and are widely used in the health monitoring of pipelines, plates, and large structures. Guided wave signals usually exhibit multiple wave packets with clear physical meanings, including excitation direct waves, defect reflection waves, structural boundary reflection waves, etc. Each wave packet carries information about the corresponding path or scattering source. Therefore, accurate identification and classification of wave packets are the key prerequisites for realizing structural damage location and quantification.

[0003] Currently, in engineering practice, the wave packet identification methods for guided wave signals still mainly rely on manual window division or searching for extreme values within a fixed time window. These methods not only rely on the experience judgment of the detector, and the identification results lack stability and repeatability, but also are prone to misjudgment in the case of strong noise or signal overlap. In addition, in actual structures, guided wave signals have multimodal propagation and dispersion characteristics, that is, the propagation speeds of different modes are related to frequency, resulting in strong time-domain aliasing and distortion of multiple wave packets at the receiving end, posing a serious challenge to traditional methods.

[0004] With the development of guided wave detection technology towards automation and online monitoring systems, there is an urgent need for an algorithm that can adaptively identify multiple effective wave packets in guided wave signals. This algorithm should be able to automatically identify the wave packets in the signal and back-calculate the defects based on the arrival time and amplitude of the abnormal wave packets.

[0005] In addition, in the long-term online monitoring scenario, changes in environmental temperature, humidity, and structural contact conditions will cause guided wave velocity drift and signal attenuation, resulting in further reduction of the stability of manual analysis. Therefore, it is necessary to rely on extracting stable wave packet features from a large amount of repeated detection data, and complete "multiple repeated positioning" through algorithms to improve the accuracy and robustness of defect identification, and eliminate false signals generated by non-structural factors.

[0006] For this reason, a wave packet self-identification algorithm is needed to automatically calibrate the information of defect reflection or transmission waves, and through a large number of repeated tests, achieve multiple repeated detections of the defect position and size, improve the accuracy and reliability of the detection, and avoid the interference of temperature, humidity, and scattering of other structures of the measured object. Summary of the Invention

[0007] To solve the above problems, the present invention discloses a wave packet self-identification method, which can automatically calibrate the wave packets with a specified form in the detection signal, and can cope with the rapid analysis of a large amount of signals, so as to achieve a large number of repeated ultrasonic tests and improve the reliability and accuracy of the final results.

[0008] To achieve the above object, the technical solution of the present invention is as follows:

[0009] A wave packet self-identification method, comprising the following steps:

[0010] S1: Measuring to obtain a time-domain signal;

[0011] S2: Constructing a target wave packet dictionary;

[0012] S3: Solving a sparse Bayesian model;

[0013] S4: Modal discrimination and waveform recognition;

[0014] S5: Inverting and precisely positioning the defect and bottom positions;

[0015] S6: Precisely positioning multiple arranged transducers.

[0016] Further, the measuring to obtain a time-domain signal in step S1 is specifically as follows:

[0017] Setting a pair of ultrasonic transducer arrays in the area to be measured of the structure, exciting and collecting a guided wave signal y(t).

[0018] Further, the constructing a target wave packet dictionary in step S2 is specifically as follows:

[0019] Constructing an over-complete wave packet dictionary, which adopts a Gabor pulse model, and each atom in the dictionary is a modulated Gaussian pulse

[0020] S2.1 Definition of Gabor atom model

[0021] Each atom g(t) in the dictionary is defined as a Gabor function in the following form:

[0022]

[0023] where: s is a scale parameter, controlling the wave packet width; u is a time delay; ω is an angular frequency; A is an amplitude normalization constant S2.2 Parameter grid design

[0024] To construct an over-complete dictionary, it is necessary to perform combined sampling on the above parameters ω, s, u, specifically as follows:

[0025] Frequency parameter ω: Selecting m ω frequency points at equal intervals from the target frequency band;

[0026] Scale parameter s: Based on the center value s0 of the excitation signal bandwidth, sampling m s values at equal intervals within the range [0.5s0, 2s0];

[0027] Time delay parameter u: Dividing the entire time domain of the guided wave signal into mu equal-interval intervals, as the pulse center moments,

[0028] Thus, a total of m ω ×m s ×m u atoms are constructed to form a dictionary matrix

[0029] where:

[0030] N is the number of signal sampling points;

[0031] L is the number of columns of the dictionary matrix, L = m ω ×m s ×m u ;

[0032] S2.3 Dictionary Matrix Normalization

[0033] Perform amplitude normalization on each atom vector in the dictionary matrix. The specific normalization formula is as follows:

[0034] For the i-th atom vector in the dictionary matrix Perform the following operations

[0035]

[0036] where:

[0037] d i (n) represents the value of the i-th atom at the n-th sampling point, i = 1, 2... L;

[0038] ||d i ||2 is the second norm (i.e., energy) of this atom;

[0039] is the normalized atom vector,

[0040] the normalized dictionary matrix

[0041] Furthermore, the solution of the sparse Bayesian model described in step S3 is as follows:

[0042] The received signal y(t) is modeled as a linear combination of the dictionary matrix and the sparse coefficient vector , expressed as:

[0043] y = Φw + ∈

[0044] where:

[0045] Φ is the pre-constructed Gabor dispersion dictionary;

[0046] w = [w1, w2, …, w L T is the sparse coefficient vector;

[0047] is zero-mean Gaussian noise.

[0048] S3.1 Initialize the noise variance σ 2

[0049] σ 2 = var(y) × 0.1

[0050] where: var(y) is the variance of the observed signal,

[0051] S3.2 Initialize the model

[0052] All hyperparameters are initialized to α m is initialized to ∞

[0053] Select an atom Φ i Initialize and calculate its initial hyperparameter α i :

[0054]

[0055] S3.3 Calculate the covariance Σ and the mean μ for the first time

[0056] Construct the system matrix:

[0057] S = βΦ T Φ + A

[0058] where:

[0059] β is the noise precision, β = 1 / σ 2 , A is the hyperparameter diagonal matrix, Then calculate

[0060] Σ = S -1

[0061] μ = βΣc T y

[0062] S3.4 Select candidate atoms

[0063] Traverse all dictionary atoms Φ in the dictionary matrix in turn i , and prepare to evaluate whether to add / update / delete it;

[0064] S3.5 Calculate the evaluation factor

[0065] For the candidate atom Φ i , calculate:

[0066] ​

[0067] Among them:

[0068]

[0069] c i : Sparsity factor, representing atomic redundancy;

[0070] q i : Quality factor, representing the fitting degree of atoms to the signal;

[0071] Then calculate:

[0072]

[0073] S3.6 Model update rule

[0074] If θ i > 0 and α i < ∞: Update the existing hyperparameter α i ;

[0075] If θ i > 0 and α i = ∞: Add a new atom Φ i And calculate α i ;

[0076] If θ i ≤ 0 and α i < ∞: Delete Φ from the model i And set α i = ∞;

[0077] S3.7 Recalculate Σ and μ

[0078] Based on the new atom set and α, re - execute:

[0079] Σ = (βΦ T Φ + A) -1

[0080] μ = βΣΦ T y.

[0081] Furthermore, the mode discrimination and waveform recognition described in step S4 are specifically as follows:

[0082] After the sparse reconstruction of the guided - wave signal is completed using the sparse Bayesian learning algorithm, the signal is represented as a linear combination of a set of "atoms". A mode - matching and discrimination scheme is adopted to complete the automatic recognition of defect reflection waves.

[0083] Select the top 10 atoms with the largest amplitudes in the reconstruction coefficients. For each atom φ i , if its frequency f iIf it does not belong to the set excitation center frequency ± bandwidth range, then the corresponding w i is removed,

[0084] Adopt the "two-step matching strategy" to pair the remaining atoms to identify the mode pairs generated by defect reflection,

[0085] S4.1: Pairing according to time delay

[0086] Sort the atoms in ascending order of time delay

[0087] Starting from the earliest arriving wave packet, assume it is the L(0,2) mode;

[0088] Let the time delay of the wave packet be t0, the distance between the exciter and the receiver be l, and the group velocities of the L(0,2) / L(0,1) modes be Then the theoretical delay corresponding to its L(0,1) mode is:

[0089]

[0090] Search among the remaining atoms to see if there is a time delay falling near t', with the threshold set to 1 μs,

[0091] If the match is successful, then retain this pair of atoms, otherwise discard the current atom,

[0092] S4.2: Filtering according to amplitude ratio

[0093] For each pair of matched atoms, if the coefficient of the atom corresponding to the L(0,1) mode is greater than that of the atom corresponding to the L(0,2) mode, it is determined as a false match and this pair is removed;

[0094] Retain the wave packet pairs that satisfy |A L(0,2) | > |A L(0,1) |,

[0095] S4.3 End echo identification

[0096] For a single wave packet that fails to be paired, if its time delay and amplitude are much greater than those of the aforementioned defect wave packet pairs, it is determined as an end reflection wave.

[0097] Furthermore, the defect and bottom position inversion and precise positioning described in step S5 are as follows:

[0098] S5.1 Defect position calculation

[0099] The guided wave is emitted from the exciter, propagates to the defect point and then reflects back, then the defect position can be calculated as:

[0100]

[0101] Where

[0102] H D is the vertical position of the defect reflection point relative to the top;

[0103] H exc is the vertical distance from the actuator installation point to the top of the structure;

[0104] c L02 is the theoretical group velocity of the L(0,2) mode;

[0105] τ D is the propagation time of the defect reflection wave packet (L(0,2)) identified in step 4.

[0106] S5.2 Bottom position calculation

[0107] Calculate the bottom position as:

[0108]

[0109] Where:

[0110] H B is the position of the bottom reflection point from the top;

[0111] τ B is the propagation time of the bottom reflection wave packet L(0,2) identified in step 4.

[0112] Furthermore, the multi-array transducer precise positioning described in step S6 is specifically as follows:

[0113] Conduct multiple tests under different transducer arrays and repeat the above calculations:

[0114] Obtain multiple groups of defect / bottom position estimation values

[0115] Perform mean fusion to improve accuracy:

[0116]

[0117] The beneficial effects of the present invention are:

[0118] A wave packet self-identification method according to the present invention can automatically calibrate wave packets with a specified form in a detection signal, can cope with the rapid analysis of a large amount of signals, thereby realizing a large number of repeated ultrasonic tests and improving the reliability and accuracy of the final result. Brief description of the drawings

[0119] Figure 1 is a flowchart of the self-screening method for signal characteristics according to the present invention.

[0120] Figure 2Schematic diagram of the connection between the lightweight detection system and the specimen used in the embodiments of the present invention.

[0121] Figure 3 Schematic diagram of the self-screening result of signal characteristics in the embodiments of the present invention. Specific embodiments

[0122] The present invention will be further illustrated below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

[0123] As shown in the figure, a wave packet self-identification method according to the present invention includes the following steps:

[0124] Step 1: Measure and obtain the time-domain signal y(t)

[0125] 1.1 Set a pair of ultrasonic transducer arrays in the area to be measured of the structure, and excite and collect the guided wave signal y(t).

[0126] Step 2: Construction of the target wave packet dictionary

[0127] In order to realize the automatic identification of wave packets in the guided wave signal, an over-complete wave packet dictionary is constructed in this step. The dictionary adopts the Gabor pulse model, and each atom of it is a modulated Gaussian pulse, which is used to approximate the local wave packet structure in the actual signal.

[0128] S2.1 Definition of the Gabor atom model

[0129] Each atom g(t) in the dictionary is defined as a Gabor function in the following form:

[0130]

[0131] where: s is the scale parameter, which controls the wave packet width; u is the time delay; ω is the angular frequency; A is the amplitude normalization constant.

[0132] S2.2 Parameter grid design

[0133] To construct an over-complete dictionary, the above parameters ω, s, and u need to be combined and sampled, specifically as follows:

[0134] Frequency parameter ω: Select m ω frequency points at equal intervals from the target frequency band;

[0135] Scale parameter s: Based on the center value s0 of the excitation signal bandwidth, sample m s values at equal intervals in the range [0.5s0, 2s0];

[0136] Time delay parameter u: Divide the entire time domain of the guided wave signal into m uEqual interval intervals are used as the pulse center moments.

[0137] Thus, a total of m ω × m s × m u atoms are constructed to form the dictionary matrix

[0138] where:

[0139] N is the number of signal sampling points;

[0140] L is the number of columns of the dictionary matrix, L = m ω × m s × m u ;

[0141] S2.3 Dictionary matrix normalization

[0142] Amplitude normalization is performed on each atom vector in the dictionary matrix. The specific normalization formula is as follows:

[0143] For the i-th atom vector in the dictionary matrix Perform the following operations

[0144]

[0145] where:

[0146] d i (n) represents the value of the i-th atom at the n-th sampling point, i = 1, 2... L;

[0147] ||d i ||2 is the two-norm of this atom;

[0148] is the normalized atom vector.

[0149] The normalized dictionary matrix

[0150] Step 3: Solve the sparse Bayesian model

[0151] The received signal y(t) is modeled as a linear combination of the dictionary matrix and the sparse coefficient vector expressed as:

[0152] y = Φw + ∈

[0153] where: Φ is the pre-constructed Gabor dispersion dictionary; w = [w1, w2,..., w L T is the sparse coefficient vector, representing the contribution of each Gabor model in the signal; is zero-mean Gaussian noise.​

[0154] S3.1 Initialize the noise variance σ 2

[0155] σ 2 = var(y) × 0.1

[0156] where: var(y) is the variance of the observed signal,

[0157] S3.2 Initialize the model

[0158] All hyperparameters are initialized to α m is initialized to ∞

[0159] Select an atom Φ i Initialize it and calculate its initial hyperparameter α i :

[0160]

[0161] S3.3 Calculate the covariance Σ and the mean μ for the first time

[0162] Construct the system matrix:

[0163] S = βΦ T Φ + A

[0164] where:

[0165] β is the noise precision, β = 1 / σ 2 , A is the hyperparameter diagonal matrix,

[0166] Then calculate

[0167] Σ = S -1

[0168] μ = βΣc T y

[0169] S3.4 Select candidate atoms

[0170] Traverse all dictionary atoms Φ in the dictionary matrix in turn i , and prepare to evaluate whether to add / update / delete it;

[0171] S3.5 Calculate the evaluation factor

[0172] For the candidate atom Φ i , calculate:

[0173]

[0174] where:

[0175]

[0176] c i : Sparsity factor, representing the atomic redundancy;

[0177] q i : Quality factor, representing the fitting degree of the atom to the signal;

[0178] Then calculate:

[0179]

[0180] S3.6 Model update rule

[0181] If θ i > 0 and α i < ∞: Update the existing hyperparameter α i ;

[0182] If θ i > 0 and α i = ∞: Add a new atom Φ i and calculate α i ;

[0183] If θ i ≤ 0 and α i < ∞: Delete Φ from the model i and set α i = ∞;

[0184] S3.7 Recalculate Σ and μ

[0185] Based on the new atom set and α, re - execute:

[0186] Σ=(βΦ T Φ + A) -1

[0187] μ = βΣΦ T y.

[0188] Step 4 Modal discrimination and waveform recognition

[0189] After the sparse reconstruction of the guided - wave signal is completed using the Sparse Bayesian Learning (SBL) algorithm, the signal is represented as a linear combination of a set of "atoms". Since the constructed over - complete dictionary covers a variety of frequency, bandwidth, and propagation - time parameters, in order to extract the physically meaningful reflected waves (i.e., defect - reflected waves or end - reflected waves) from it, the present invention proposes a modal matching and discrimination scheme for automatically identifying defect - reflected waves.

[0190] Select the top 10 atoms with the largest amplitudes in the reconstruction coefficients. For each atom φ i , if its frequency f i does not belong to the range of the set excitation center frequency ± bandwidth, then the corresponding w i is removed.

[0191] The retained atoms are paired using a "two-step matching strategy" to identify the mode pairs generated by defect reflections.

[0192] S4.1: Pairing according to time delay

[0193] Sort the atoms in ascending order of time delay.

[0194] Starting from the earliest arriving wave packet, assume it is the L(0,2) mode.

[0195] Let the time delay of the wave packet be t0, the distance between the actuator and the receiver be l, and the group velocities of the L(0,2) / L(0,1) modes be Then the theoretical delay corresponding to the L(0,1) mode is:

[0196]

[0197] Search among the remaining atoms to see if there is a time delay falling near t', with the threshold set to 1 μs.

[0198] If the match is successful, retain this pair of atoms; otherwise, discard the current atom.

[0199] S4.2: Filtering according to the amplitude ratio

[0200] For each pair of matched atoms, if the coefficient of the atom corresponding to the L(0,1) mode is greater than that of the atom corresponding to the L(0,2) mode, it is determined as a false match and this pair is excluded.

[0201] Retain the wave packet pairs that satisfy |A L(0,2) | > |A L(0,1) |.

[0202] S4.3 End echo identification

[0203] For a single wave packet that fails to be paired, if its time delay and amplitude are much greater than those of the aforementioned defect wave packet pairs, it is determined as an end reflection wave.

[0204] Step 5: Inversion and precise positioning of the defect and the bottom

[0205] Taking the defect detection of a steel pipe as an example.

[0206] S5.1 Defect position calculation

[0207] The guided wave is emitted from the actuator, propagates to the defect point and then reflects back. Then the defect position can be calculated as:

[0208]

[0209] Where:

[0210] HD The vertical position of the defect reflection point relative to the top

[0211] H exc The vertical distance from the actuator installation point to the top of the structure

[0212] c L02 Is the theoretical group velocity of the L(0,2) mode,

[0213] τ D Is the propagation time of the defect reflection wave packet (L(0,2)) identified in step 4,

[0214] S5.2 Bottom position calculation

[0215] The bottom position is calculated as:

[0216]

[0217] Where:

[0218] H B Is the position of the bottom reflection point from the top;

[0219] τ B Is the propagation time of the bottom reflection wave packet L(0,2) identified in step 4.

[0220] Step 6: Precise positioning with multiple transducers

[0221] Perform multiple tests under different transducer arrangements (adjusting the actuator and receiver up and down), and repeat the above calculations:

[0222] Obtain multiple sets of defect / bottom position estimates

[0223] Perform mean fusion to improve accuracy:

[0224]

[0225] Due to the self-identification characteristics of this method, it can cope with the rapid analysis of a large number of signals, thereby realizing a large number of repeated ultrasonic tests and improving the reliability and accuracy of the final results.

[0226] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can still be made, and these improvements and refinements all fall within the protection scope of the claims of the present invention.

Claims

1. A wave packet self-identification method, characterized in that: It includes the following steps: S1: Measure and obtain the time-domain signal; S2: Construct the target wave packet dictionary; S3: Solve the sparse Bayesian model; S4: Modal discrimination and waveform recognition; S5: Invert and accurately locate the defect and bottom positions; S6: Accurately locate multiple arranged transducers.

2. The wave packet self-identification method according to claim 1, wherein: The measurement to obtain the time-domain signal described in step S1 is specifically as follows: Set a pair of ultrasonic transducer arrays in the area to be measured of the structure, and excite and collect the guided wave signal y(t).

3. A wave packet self-identification method according to claim 1, characterized in that: The construction of the target wave packet dictionary described in step S2 is specifically as follows: Construct an over-complete wave packet dictionary, which adopts the Gabor pulse model, and each atom of which is a modulated Gaussian pulse, used to approximate the local wave packet structure in the actual signal. S2.1 Definition of Gabor atom model Each atom g(t) in the dictionary is defined as a Gabor function in the following form: Where: s is the scale parameter, controlling the wave packet width; u is the time delay; ω is the angular frequency; A is the amplitude normalization constant S2.2 Parameter grid design To construct an over-complete dictionary, it is necessary to perform combined sampling on the above parameters ω, s, u, specifically as follows: Frequency parameter ω: Select m ω frequency points at equal intervals from the target frequency band; Scale parameter s: Based on the center value s0 of the excitation signal bandwidth, m values are sampled at equal intervals within the range [0.5s0, 2s0]. s values; Time delay parameter u: The time domain of the entire guided wave signal is divided into m u equally spaced intervals, serving as the pulse center moments. Thus, a total of m ω × m s × m u atoms are constructed to form a dictionary matrix Where: N is the number of signal sampling points; L is the number of columns of the dictionary matrix, L = m ω × m s × m u ; S2.3 Dictionary matrix normalization Perform amplitude normalization processing on each column vector atom in the dictionary matrix, and the specific normalization formula is as follows: For the i-th atomic vector in the dictionary matrix Perform the following operations Where: d i (n) represents the value of the i-th atom at the n-th sampling point; ||d i ||2 is the two-norm of the atom; is the normalized atomic vector, Normalized dictionary matrix 4. A wave packet self-identification method according to claim 1, characterized in that: The solution of the sparse Bayesian model described in step S3 is specifically as follows: The received signal y(t) is modeled as a linear combination of the dictionary matrix and the sparse coefficient vector as expressed by: y = Φw + ∈ Where: Φ is the pre-constructed Gabor dispersion dictionary w = [w1, w2, …, w L T is a sparse coefficient vector;​ is zero-mean Gaussian noise S3.1 Initialize the noise variance σ 2 σ 2 = var(y) × 0.1 Where: var(y) is the variance of the observed signal S3.2 Initialize the model All hyperparameters are initialized to α m is initialized to ∞ Select an atom Φ i Initialize and calculate its initial hyperparameter α i : S3.3 Calculate the covariance Σ and mean μ for the first time Construct the system matrix: S = βΦ T Φ + A Where: β is the noise precision, β = 1 / σ 2 ; A is a hyperparameter diagonal matrix, Then calculate Σ = S -1 μ = βΣc T y S3.4 Select candidate atoms Traverse all dictionary atoms Φ in the dictionary matrix in sequence i , and prepare to evaluate whether to add / update / delete it; S3.5 Calculate the evaluation factor For candidate atom Φ i , calculate: Where: c i : Sparsity factor, indicating atomic redundancy; q i : Quality factor, indicating the degree of fitting of the atom to the signal; Then calculate: S3.6 Model update rule If θ i > 0 and α i < ∞: Update the existing hyperparameter α i ; If θ i > 0 and α i = ∞: Add new atom Φ i and calculate α i ; If θ i ≤ 0 and α i < ∞: Delete Φ from the model i and set α i = ∞; S3.7 Recalculate Σ and μ Based on the new atom set and α, re-execute: Σ = (βΦ T Φ + A) -1 μ = βΣΦ T y.

5. A wave packet self-identification method according to claim 1, characterized in that: The modal discrimination and waveform recognition described in step S4 are specifically as follows: After the sparse reconstruction of the guided wave signal is completed using the sparse Bayesian learning algorithm, the signal is represented as a linear combination of a group of "atoms", and a modal matching and discrimination scheme is adopted to complete the automatic recognition of the defect reflection wave. Select the top 10 atoms with the largest amplitudes in the reconstruction coefficients. For each atom φ i , if its frequency f i does not belong to the range of the set excitation center frequency ± bandwidth, then the corresponding w i is removed. Adopt a "two-step matching strategy" to pair the remaining atoms to identify the mode pairs generated by defect reflection. S4.1: Pair according to the time delay Sort the atoms in ascending order of time delay Starting from the earliest arriving wave packet, assume it is the L(0,2) mode; Let the time delay of the wave packet be \(t_0\), the distance between the exciter and the receiver be \(l\), and the group velocities of the \(L(0,2) / L(0,1)\) modes be Then the theoretical delay corresponding to the \(L(0,1)\) mode is: Search in the remaining atoms whether there is a time delay falling near t', and the threshold is set to 1 μs. If the matching is successful, retain this pair of atoms, otherwise discard the current atom. S4.2: Filter according to the amplitude ratio For each pair of matched atoms, if the coefficient of the atom corresponding to the L(0,1) mode is greater than the atom corresponding to the L(0,2) mode, it is determined as a false match and this pair is excluded; Retain the wave packet pairs that satisfy |A L(0,2) | > |A L(0,1) |, S4.3 End echo recognition For a single wave packet that fails to be paired, if its time delay and amplitude are much larger than the aforementioned defect wave packet pair, it is determined as the end reflection wave.

6. A wave packet self-identification method according to claim 5, characterized in that: The inversion and accurate positioning of the defect and bottom positions described in step S5 are specifically as follows: S5.1 Defect position calculation The guided wave is emitted from the exciter, propagates to the defect point and then reflects back, then the defect position can be calculated as: Where: H D is the vertical position of the defect reflection point relative to the top; H exc is the vertical distance from the actuator installation point to the top of the structure; c L02 is the theoretical group velocity of the L(0,2) mode; τ D is the propagation time of the defect reflection wave packet (L(0, 2)) identified in step 4 S5.2 Bottom Position Calculation The calculated bottom position is as follows: Where: H B is the position of the bottom reflection point from the top, τ B is the propagation time of the bottom reflection wave packet L(0, 2) identified in step 4.

7. A wave packet self-identification method according to claim 1, characterized in that: The precise positioning of the multiple arrayed transducers described in step S6 is as follows: Conduct multiple tests under different transducer arrangements and repeat the above calculations: Obtain multiple sets of defect / bottom position estimation values Execute mean fusion to improve the accuracy:

Images