Damage identification and assessment method for composite panels based on Lamb wave and autoencoder

Abstract

The present invention discloses a method for damage identification and assessment of composite wall panels based on Lamb waves and autoencoders. The method comprises: placing a Lamb wave signal acquisition system on a healthy first wall panel and training a deep autoencoder model at each receiver; using the Lamb wave signal acquisition system to acquire Lamb wave response signals for a second wall panel to be inspected; determining the Lamb wave flight time at each receiver on the second wall panel using differential signals; constructing and solving a set of geometric model equations based on a Lamb wave isotropic propagation velocity distribution model to locate the damage position in the second wall panel; and reconstructing the Lamb wave response signals acquired by multiple receivers on the second wall panel using a deep autoencoder model and performing normalization processing to obtain a final comprehensive damage assessment score. The method has the advantages of high positioning accuracy, strong assessment sensitivity, and good ability to distinguish damage levels, and is suitable for nondestructive testing and assessment of composite structural wall panels.

Description

Technical Field

[0001] The present invention relates to the field of wall panel damage identification and assessment, and in particular to a composite material wall panel damage identification and assessment method based on Lamb waves and an autoencoder. Background Art

[0002] With the rapid development of high-end equipment manufacturing in aerospace, rail transportation, marine engineering, and new energy, lightweight and high-strength advanced composite materials are widely used in various key load-bearing structures. However, composite materials are extremely susceptible to various factors such as impact, fatigue, delamination, and foreign body penetration during actual service, resulting in invisible or subtle internal damage such as delamination, debonding, matrix cracks, and fiber breakage. These damages are often hidden and progressive. If they are not detected and identified in a timely and accurate manner, they are likely to cause structural degradation or even catastrophic failure, seriously threatening personnel safety and system reliability. Therefore, damage detection, identification, and assessment technology for composite wall panels is a research hotspot in the field of structural health monitoring. Among them, non-destructive testing technology based on guided waves (such as Lamb waves) has received widespread attention in recent years due to its high sensitivity, long-distance propagation capability, and adaptability to various damage types, and has gradually been applied to damage identification of composite structures.

[0003] Although a series of achievements have been made in Lamb wave detection technology, such as probabilistic imaging algorithms (CN114910562A, CN113376250A), data-driven modeling (CN114925716A), and signal time-frequency analysis (CN113390967A), all of which have demonstrated good performance in damage location and identification, the overall development trend shows that existing research has largely focused on detecting and determining damage location, with significantly insufficient attention paid to quantitative assessment of damage severity. In other words, current damage identification methods emphasize "whether" and "where" the damage is present, while the ability to determine "how severe" the damage is remains in its early stages of exploration. In particular, when dealing with minor damage, multi-stage damage, or continuous damage evolution that occurs during service, traditional methods based solely on characteristic quantities such as time difference, waveform changes, or wave velocity distribution struggle to obtain reliable and quantifiable assessment results. Therefore, it is necessary to develop Lamb wave detection technology that can both accurately locate damage in composite panels and comprehensively assess the damage severity. Summary of the Invention

[0004] The purpose of the present invention is to provide a composite material wall panel damage identification and assessment method based on Lamb waves and autoencoders, which is used for non-destructive detection and assessment of composite material wall panels, and has the characteristics of high positioning accuracy, strong assessment sensitivity, and good ability to distinguish damage levels.

[0005] In order to achieve the above tasks, this program adopts the following technical solutions:

[0006] The damage identification and assessment method for composite panels based on Lamb waves and autoencoders includes:

[0007] A Lamb wave signal acquisition system is arranged on the healthy first wall panel; Lamb wave response signals are acquired using receivers at different locations on the first wall panel under a preset excitation signal; and a deep autoencoder model corresponding to each receiver is trained based on the Lamb wave response signals acquired by each receiver;

[0008] For the second wall panel to be tested, the Lamb wave signal acquisition system is arranged in the same manner as for the first wall panel, and the Lamb wave response signal is acquired using a receiver; wherein the second wall panel is the same type as the first wall panel;

[0009] Calculating a differential signal using the Lamb response signals obtained from the first wall panel and the second wall panel, and determining the Lamb wave flight time at each receiver on the second wall panel using the differential signal;

[0010] Based on the Lamb wave flight time at some receivers and in combination with the Lamb wave isotropic propagation velocity distribution model, a geometric model equation group is constructed; an objective function is constructed to solve the geometric model equation group, thereby locating the damage position in the second wall panel;

[0011] Lamb wave response signals obtained by multiple receivers on the second wall panel are selected as original signals and input into the corresponding deep autoencoder models to obtain the reconstructed signals corresponding to each receiver. The original signals and the reconstructed signals are used to determine the average reconstruction error as the damage assessment value at each receiver and perform normalization. The comprehensive damage assessment score is determined based on the normalized damage assessment values ​​at the selected multiple receivers.

[0012] Furthermore, the Lamb wave signal acquisition system includes an exciter and a plurality of receivers arranged around the exciter; a damage location coordinate system is constructed with the location of the exciter as the origin, and the positive direction of the 𝑥-axis is defined to be along the main axis of the fiber of the first wall panel, and the positive direction of the y-axis and the positive direction of the x-axis are in the same plane;

[0013] The preset excitation signal adopts a single Lamb wave mode, and the center frequency is set in the non-dispersion section of the dispersion curve; the excitation signal adopts a sinusoidal pulse.

[0014] Furthermore, the deep autoencoder model includes an encoder and a decoder; the encoder is used to map the Lamb wave response signal to a low-dimensional feature space to form a low-dimensional feature vector at the hidden layer;

[0015] The decoder takes the low-dimensional feature vector at the hidden layer as input and outputs the reconstructed signal of the Lamb wave response signal; the training goal of the deep autoencoder model is to minimize the reconstruction error between the Lamb wave response signal and the reconstructed signal.

[0016] Furthermore, a differential signal is calculated using the Lamb response signals obtained from the first wall panel and the second wall panel, and the flight time of the Lamb wave at each receiver on the second wall panel is determined by the differential signal, including:

[0017] On the second wall The Lamb wave response signal collected by the receiver is , calculate the The average response signal at the receiver ; On the first wall panel The Lamb wave response signal collected by the receiver is , calculate the The average response signal at the receiver ;

[0018] Calculate the The differential signal at each receiver ;

[0019] For differential signals Apply complex Morlet wavelet transform, the wavelet transform result presents time-scale spectrum, extract the time corresponding to the peak of envelope spectrum as the first The arrival time of the Lamb wave response signal at each receiver ;

[0020] Determine the second wall Lamb wave flight time at the receiver ; Indicates the start time of the stimulus signal.

[0021] Furthermore, when locating the damage position in the second wall panel, firstly, the Lamb wave flight times of all receivers on the second wall panel are sorted from short to long, and the first Q Lamb wave flight times are selected;

[0022] Based on the selected Q Lamb wave flight times and the Lamb wave propagation velocity distribution model, a set of geometric model equations based on Lamb waves is constructed;

[0023] The least squares method is used to construct the objective function to solve the geometric model equations; by minimizing the calculated value of the Lamb wave flight time from the damage position to the i-th receiver and the actual Lamb wave flight time The residual square sum of is used to obtain the optimal estimated position coordinates of the damage location.

[0024] Furthermore, the geometric model equations are expressed as follows:

[0025] ;

[0026] Among them, the sixth equation of the equation group is the Lamb wave propagation velocity distribution model in all directions;

[0027] In the above formula, represents the Lamb wave flight time at the i-th receiver among the first Q Lamb wave flight times selected. There is a damage location D on the second wall, and its position coordinates are ; The position coordinates of the i-th receiver are ; 、 They represent the propagation paths of the Lamb wave from the exciter to the damage location D and from the damage location D to the i-th receiver. 、 length, 、 Represents the propagation paths 、 The angle between the positive direction of the x-axis of the damage location coordinate system, 、 Lamb wave at the angle 、 Propagation velocity in a direction; dimensionless parameter Characterization and the second wall panel fiber main axis direction Lamb wave velocity in the angle direction Lamb wave velocity in the direction of the fiber main axis The ratio of Obtained through actual measurement.

[0028] Furthermore, the least squares method is used to construct the objective function to solve the geometric model equations, thereby obtaining the optimal estimated position coordinates of the damage location; the objective function is expressed as follows:

[0029] ;

[0030] in, represents the optimal estimated position coordinates, 、 They represent the damage position D and the origin O of the damage location coordinate system, the damage position D and the i-th receiver respectively. The Euclidean distance between is the calculated value of the Lamb wave flight time.

[0031] Furthermore, the dimensionless parameter The expression is:

[0032] ;

[0033] in, is the equivalent Young's modulus of the damage location coordinate system on the second wall plate in the x-axis direction, is the equivalent Poisson's ratio along the y-axis on the plane whose normal is the x-axis; and They are respectively Equivalent Young's modulus and equivalent Poisson's ratio in the direction of the angle.

[0034] Furthermore, the original signal and the reconstructed signal are used to determine an average reconstruction error as a damage assessment value at each receiver and normalized. Based on the normalized damage assessment values ​​at the selected multiple receivers, a comprehensive damage assessment score is determined, including:

[0035] The damage assessment values ​​are normalized based on the minimum average reconstruction error among the damage assessment values ​​calculated based on the Lamb wave response signals collected by the selected multiple receivers;

[0036] The multiple standardized damage assessment values ​​are arithmetic averaged to obtain the average damage assessment value of the second wall panel in its current damage state. The standardized damage assessment value corresponding to the reconstruction error of the Lamb wave response signal obtained on the first wall panel in the healthy state is introduced as a benchmark, and a second normalization process is performed to obtain the comprehensive damage assessment score.

[0037] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, the composite material wall panel damage identification and assessment method based on Lamb waves and autoencoders is implemented.

[0038] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the composite material wall panel damage identification and assessment method based on Lamb waves and autoencoders is implemented.

[0039] Compared with the prior art, the present invention has the following technical features:

[0040] This invention effectively addresses the common problem in existing technologies, which can only identify damage locations but struggle to precisely quantify the extent of damage. This approach uses differential enhancement and continuous wavelet transforms to effectively extract the Lamb wave flight time, improving signal processing accuracy. Furthermore, by introducing an autoencoder model trained on health status, reconstruction error is used to quantitatively assess different damage levels, resolving the difficulty of traditional methods in measuring "severity." This method can accurately locate and assess the extent of damage in composite wall panels, demonstrating strong application prospects and potential for widespread adoption. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the process of the present invention;

[0042] Figure 2 Schematic diagram of the damage location and assessment system of the method of the present invention;

[0043] Figure 3 Schematic diagram of the solution of Lamb wave velocity in each direction of the present invention;

[0044] Figure 4 Schematic diagram of the deep autoencoder model used in an embodiment of the present invention;

[0045] Figure 5 is the dispersion curve of the orthogonal laminated composite panel in the 0° direction in an embodiment of the present invention;

[0046] Figure 6 Schematic diagram of experimental test equipment and experimental arrangement in an embodiment of the present invention;

[0047] Figure 7 is the normalized Lamb wave response signal collected by the receiver 2 when the wall panel is in an intact state in an embodiment of the present invention;

[0048] Figure 8 The ratio of the Lamb wave velocity in each direction to the Lamb wave velocity in the main axis fiber direction derived in the embodiment of the present invention is distribution model of;

[0049] Figure 9 The figure is a result diagram of the damaged Lamb wave response signal collected by each receiver in the damage case of the present invention after differential processing and continuous wavelet transform;

[0050] Figure 10 A visual comparison of the estimated damage location and the actual damage location in an embodiment of the present invention;

[0051] Figure 11 The figure shows a comparison of the comprehensive assessment values ​​of the damage degree at each damage level in the embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention provides a method for identifying and assessing damage to composite material panels based on Lamb waves and autoencoders. This method achieves the identification of damage locations and the assessment of damage severity in composite material panels by combining the propagation characteristics of Lamb waves, wave velocity anisotropy modeling, wavelet transform to extract flight time, and a deep autoencoder model to calculate the reconstruction error between damaged and healthy signals.

[0053] See also Figure 1 This solution provides a composite panel damage identification and assessment method based on Lamb waves and autoencoders, including:

[0054] Step 1: Arrange a Lamb wave signal acquisition system on a healthy first wall panel; use receivers at different positions on the first wall panel to acquire Lamb wave response signals under a preset excitation signal; and train a deep autoencoder model corresponding to each receiver based on the Lamb wave response signal acquired by each receiver.

[0055] (1-1) Preset excitation signal.

[0056] In order to realize the damage location and degree assessment of composite wall panels based on Lamb waves, it is necessary to first determine the parameters of the excitation signal used to stimulate the propagation of Lamb waves, including the excited Lamb wave mode, center frequency, and signal form.

[0057] In order to reduce the interference caused by multi-mode superposition, this scheme selects a single Lamb wave mode with good propagation characteristics, weak dispersion effect, and relatively stable wave speed (such as symmetrical mode or antisymmetric mode ) as the main propagation mode of the excitation signal; the selection of Lamb wave mode can be determined based on factors such as the thickness of the composite material plate and the dispersion curve.

[0058] Center frequency It is advisable to set it in the non-dispersive section where the wave velocity is relatively stable in the dispersion curve to improve the stability of signal propagation and facilitate the subsequent extraction of flight time and time domain response characteristics.

[0059] The excitation signal is in the form of a 5-cycle Hanning amplitude modulated sinusoidal pulse, and its expression is as follows:

[0060] (1);

[0061] Among them, A=1 represents the amplitude of the excitation signal, is the angular frequency of the excitation signal, is the time parameter.

[0062] (1-2) Lamb wave signal acquisition system.

[0063] The panel used for damage identification in this solution is a composite panel. First, a healthy panel is selected as the first panel to collect relevant prior data. Since Lamb waves propagate at different speeds along different fiber directions in the panel, to facilitate the location of damage within the panel, this solution constructs a Lamb wave signal acquisition system on the panel and establishes a damage location coordinate system, as follows:

[0064] This scheme builds a Lamb wave signal acquisition system; see Figure 2 The system includes an exciter and multiple receivers arranged around the exciter, which together constitute a sensing network for Lamb wave signal excitation and reception; the exciter and receiver both use the same transducer element, such as piezoelectric ceramics; the damage localization coordinate system is a local Cartesian coordinate system established with the position of the exciter as the origin O(0,0); the positive direction of the 𝑥-axis is defined as along the main axis of the fiber of the first wall panel, and the positive direction of the y-axis is in the same plane as the positive direction of the x-axis and is perpendicular to the positive direction of the x-axis.

[0065] The area enclosed by all receivers arranged on the first wall panel is the Lamb wave damage detection area. The location, size, and shape of the Lamb wave damage detection area are determined by the user based on actual needs by adjusting the number and location of receivers. The subsequent work of this solution is to predict the coordinates of the damage location within the damage detection area of ​​the second wall panel to be inspected in the damage location coordinate system and assess the damage extent. Figure 2 Schematic diagram of the rectangular damage detection area formed by the arrangement of four receivers.

[0066] (1-3) Deep autoencoder model.

[0067] This solution uses an unsupervised learning method based on a deep autoencoder (DAE) model to quantitatively assess damage severity without pre-labeling. By building a DAE model that learns only the "healthy state characteristics" of the siding, the DAE model is trained to "understand" only the Lamb wave response characteristics in a healthy state. However, if the input is a Lamb wave response signal containing damage, the reconstruction result will produce a large reconstruction error. Based on this principle, this solution uses the reconstruction error of the Lamb wave response signal of the trained DAE model at different damage levels as a damage assessment indicator, achieving a quantitative assessment of the damage severity of the composite siding.

[0068] The deep autoencoder model includes an encoder and a decoder, wherein:

[0069] The encoder is responsible for mapping the Lamb wave response signal X into a low-dimensional feature space to form a low-dimensional feature vector at the hidden layer. In this process, the encoder extracts the main features of the signal and compresses redundant information to obtain a concise but information-rich representation.

[0070] The decoder takes the low-dimensional feature vector at the hidden layer as input and tries to reconstruct a reconstructed signal X that is as close as possible to the original high-dimensional Lamb wave response signal. R ; The role of the decoder is to evaluate the expressiveness of the encoded features.

[0071] As attached Figure 4 As shown, in the training phase of this scheme, the input of the autoencoder model is the Lamb wave response signal X obtained by each receiver on the first wall plate = ; Output is the reconstructed signal X R = ;also,

[0072] In addition, the forward propagation process of the autoencoder model can be expressed as follows:

[0073] The encoding process from the input layer to the hidden layer:

[0074] (2);

[0075] The decoding process from the hidden layer to the output layer:

[0076] (3);

[0077] in, Represents the encoding process, Represents the decoding process, represents the activation function, 、 are the weight matrices for the encoding and decoding processes, 、 is the bias term, is the hidden layer feature representation.

[0078] The training objective of the deep autoencoder model is to minimize the reconstruction error Loss between the Lamb wave response signal and the reconstructed signal:

[0079] (4);

[0080] Where N is the length of the Lamb wave response signal X, , are the time domain amplitudes of the Lamb wave response signal X and the reconstructed signal at the nth sampling point, respectively. The deep autoencoder model was trained 1000 times, with a learning rate of 0.02 and an Adam optimizer.

[0081] The deep autoencoder model in this solution is trained using an unsupervised learning approach. The input Lamb wave response signal does not require pre-labeled input, and learning is based solely on healthy Lamb wave response signals. By training on a large number of Lamb wave response signal samples, the model can grasp the basic characteristic distribution of healthy Lamb wave signals and reconstruct them stably.

[0082] In order to improve the recognition accuracy and adaptability, this solution Build independent deep autoencoder models with the same model framework and with the receiver The collected Lamb wave response signal X is trained to learn the receiver The characteristic distribution of the Lamb wave response signal obtained for the first wall panel in a healthy state at the location ensures the reliability and regional sensitivity of the evaluation results.

[0083] Step 2: For the second wall panel to be tested, arrange the Lamb wave signal acquisition system in the same manner as the first wall panel, and use a receiver to obtain the Lamb wave response signal; wherein the second wall panel is the same type of wall panel as the first wall panel.

[0084] In actual application, the second wall panel to be tested is a wall panel of the same type as the first wall panel, that is, a composite wall panel with the same material, the same layup and the same size; first, a Lamb wave signal acquisition system is arranged on the second wall panel. The number, position and preset exciter and receiver in the system are the same as those on the first wall panel.

[0085] After the Lamb wave signal acquisition system is deployed, the exciter generates Lamb waves based on the preset excitation signal, and each receiver obtains the corresponding Lamb wave response signal.

[0086] Step 3: Calculate a differential signal using the Lamb response signals obtained from the first wall panel and the second wall panel, and determine the Lamb wave flight time at each receiver on the second wall panel through the differential signal.

[0087] This solution provides a method to extract the flight time of Lamb waves by combining differential operation and wavelet transform technology. signal processing method; by performing differential operation on the Lamb wave response signals collected on the first and second wall panels, the scattered signal characteristics caused by damage are enhanced, and then the time-frequency focusing capability of the complex Morlet wavelet is used to perform time-frequency transformation on the differential signal, thereby accurately extracting the arrival time of the scattered wave and calculating the Lamb wave flight time.

[0088] (3-1) Record the first The Lamb wave response signal collected by the receiver is , the sampling times are set to , No. The receiver is in The Lamb wave response signal obtained by sampling is recorded as ,in ; Calculate the average response signal at the receiver :

[0089] (5);

[0090] (3-2) On the first wall panel The Lamb wave response signal collected by the receiver is , then the average response signal at the receiver :

[0091] (6);

[0092] (3-3) To enhance the signal difference caused by damage, calculate the The differential signal at each receiver :

[0093] (7);

[0094] (3-4) For differential signals Applying complex Morlet wavelet transform, its wavelet function The expression is:

[0095] (8);

[0096] in, is the imaginary unit, is the time variable, is a natural constant, is the bandwidth, is the center frequency.

[0097] The wavelet transform result presents a time-scale spectrum, and the time corresponding to the peak of the envelope spectrum is extracted as the first The arrival time of the Lamb wave response signal at each receiver is recorded as .

[0098] (3-5) Determine the first Lamb wave flight time at the receiver :

[0099] (9);

[0100] in, Indicates the start time of the stimulus signal.

[0101] Step 4: construct a geometric model equation group based on the Lamb wave flight time at some receivers and in combination with the Lamb wave isotropic propagation velocity distribution model; construct an objective function to solve the geometric model equation group, thereby locating the damage position in the second wall panel.

[0102] (4-1) Damage localization model.

[0103] See also Figure 2 , take the second wall panel as an example; assume that there is a damage location D on the second wall panel, and its position coordinates in the damage location coordinate system are The black dotted line in the figure represents the propagation direction of the Lamb wave. The Lamb wave is emitted by the exciter and propagates to the receiver after being scattered by the damage position D. Receiver The position coordinates are .

[0104] Assume that the propagation time of Lamb wave from the exciter to the damage location D is , from the damage location D to the i-th receiver The propagation time is , then from the exciter excitation Lamb wave to the i-th receiver The flight time of the received Lamb wave satisfy:

[0105] (10);

[0106] in, 、 They represent the Lamb wave from the exciter to the damage location D and the damage location D to the i-th receiver respectively. The transmission path 、 length, 、 Represents the propagation paths 、 The angle between the positive direction of the x-axis of the damage location coordinate system, 、 Lamb wave at the angle 、 The speed of propagation in the direction.

[0107] At the same time, in the Lamb wave signal acquisition system, the propagation path 、 Length 、 The geometric relationship between them can be expressed as:

[0108] (11);

[0109] Angle 、 It can be expressed as:

[0110] (12);

[0111] Through the above damage location model, namely equations (10) to (12), using multiple receivers with known position coordinates and Lamb wave velocities in different directions , we can construct the inverse equations of the damage location and inversely calculate the position coordinates of the damage location.

[0112] (4-2) Lamb wave propagation velocity distribution model in all directions.

[0113] Considering the anisotropy of the Lamb wave propagation velocity in the composite wall, it is cumbersome to obtain the Lamb wave velocity in each direction by actual measurement. Therefore, in order to reflect the change of the Lamb wave propagation velocity in different directions in the second wall, see Figure 3 , introducing dimensionless parameters Characterization and the second wall panel fiber main axis direction Lamb wave velocity in the angle direction The ratio of the Lamb wave velocity in the direction of the fiber principal axis of the second wall panel (i.e., 0°, which is consistent with the positive direction of the x-axis of the damage location coordinate system established in this scheme) is shown in formula (13); , only the Lamb wave velocity along the main axis of the second wall panel fiber needs to be measured , that is, the other Lamb wave velocity in the angle direction .

[0114] (13);

[0115] in, is the Lamb wave propagation velocity along the main axis of the second wall fiber, With the main axis of the fiber The propagation speed of Lamb waves in the angle direction is is the equivalent Young's modulus of the damage location coordinate system on the second wall plate in the x-axis direction, is the equivalent Poisson's ratio along the y-axis on the plane whose normal is the x-axis; and They are respectively Equivalent Young's modulus and equivalent Poisson's ratio in the direction of the angle:

[0116] (14);

[0117] in, and are the equivalent Young's modulus of the wall panel in the x-axis and y-axis directions, is the equivalent shear modulus along the y-axis on the plane whose normal is the x-axis.

[0118] (4-3) Lamb wave velocity in the fiber main axis direction It is obtained through actual measurement. The specific steps are as follows:

[0119] A receiver is placed along the main axis of the second panel fiber (0°) and at a specified distance L from the exciter (coordinate origin O). The exciter then transmits a preset excitation signal, and the receiver receives the Lamb wave response signal. The flight time t of the Lamb wave is then extracted. Finally, the following formula is used to calculate: :

[0120] (15);

[0121] The Lamb wave propagation velocity distribution model of formula (13) to formula (15) can be used to determine the Lamb wave propagation velocity along the fiber main axis direction under the specified mode in the composite wall panel. The wave velocity in the angle direction is used for damage location calculation.

[0122] (4-4) Locate the damage in the second wall panel.

[0123] To accurately identify the damage location in the second wall panel, this scheme fully considers the flight time error caused by factors such as boundary reflection, material nonlinearity, and noise during the propagation of Lamb waves. By selecting data from receivers with shorter flight times and combining it with the Lamb wave propagation velocity distribution model in all directions, a set of geometric equations is constructed and solved using the least squares optimization method to ultimately obtain the optimal spatial position estimate of the damage location.

[0124] (4-4-1) In the actual detection process, as the Lamb wave propagation path lengthens, the Lamb response signal collected by the receiver farther away from the damage location is more susceptible to factors such as boundary reflection, material nonlinearity, and noise interference, thereby reducing the accuracy of Lamb wave flight time extraction.

[0125] Based on this, the Lamb wave flight times of all receivers on the second wall panel obtained in step 3 are sorted from short to long, and the receivers corresponding to the first Q Lamb wave flight times are selected for positioning modeling; the value of Q can be flexibly set by the user according to the detection accuracy requirements and actual structural conditions.

[0126] (4-4-2) Based on the selected Q Lamb wave flight times , combined with the Lamb wave isotropic propagation velocity distribution model, a geometric model equation group based on Lamb waves is constructed; this equation group describes the functional relationship between the propagation path from the damage location to each receiver and the Lamb wave flight time and Lamb wave velocity, as shown in Equation (16):

[0127] (16);

[0128] In the above formula, the flight time of Q Lamb waves selected for the second wall is and the corresponding receiver position coordinates , dimensionless parameters In each The values ​​in the angle direction and the Lamb wave velocity in the direction of the fiber main axis is a known quantity.

[0129] (4-4-3) To obtain the optimal estimation result of the damage location, this scheme uses the least squares method to solve the above geometric model equations (16); by minimizing the calculated value of the Lamb wave flight time from the damage location to the i-th receiver and the actual Lamb wave flight time The residual square sum of the damage position is obtained to obtain the optimal estimated position coordinates of the damage position. ; Its objective function is expressed as follows:

[0130] (17);

[0131] in, Represents the Euclidean distance between point A and point B in the damage location coordinate system, that is, ; 、 Indicates that the main axis direction of the fiber of the second wall panel is 、 Lamb wave velocity in the angle direction.

[0132] By solving the above objective function, the optimal estimated position coordinates of the damage location in the second wall can be obtained: , to achieve damage localization of composite material panels.

[0133] Step 5: Select the Lamb wave response signals obtained by multiple receivers on the second wall panel as the original signals, and input them into the corresponding deep autoencoder models to obtain the reconstructed signals corresponding to each receiver; use the original signals and the reconstructed signals to determine the average reconstruction error as the damage assessment value at each receiver and perform normalization processing; determine the comprehensive damage assessment score based on the damage assessment values ​​after normalization at the selected multiple receivers.

[0134] Based on the positioning results of the damage position on the second wall panel, the damage extent is further assessed. Considering the complex factors such as energy attenuation, boundary reflection and material nonlinearity in the propagation of Lamb waves, which may cause the effective damage characteristics in the signal collected by the receiver far away from the damage location to be weakened, thereby affecting the accuracy of the damage extent assessment; Therefore, this scheme selects M receivers closest to the damage location (usually M=Q), and based on the trained deep autoencoder model corresponding to these receivers The collected Lamb wave response signals are reconstructed, and the average reconstruction error is calculated. Through normalization and comprehensive processing, the deviation degree between the health status of the second wall panel and the first wall panel is quantified to achieve an accurate assessment of the damage degree.

[0135] (5-1) The Lamb wave response signals collected by the selected M receivers on the second wall As the original signal, it is input into the corresponding trained deep autoencoder model , and obtain its reconstructed signal The average reconstruction error between the original signal and the reconstructed signal is used as the damage assessment value at the i-th receiver. , which is calculated as follows:

[0136] (18);

[0137] Where N represents the length of the Lamb wave response signal, Indicates the The number of samples per receiver; represents the average reconstruction error between the original signal and the reconstructed signal at the i-th receiver, i.e., the damage assessment value; is the Lamb wave response signal at the nth sampling point in the kth sampling of the i-th receiver The time domain amplitude of For the deep autoencoder model According to the input signal , the reconstructed time domain amplitude at the nth sampling point.

[0138] (5-2) In order to eliminate the difference in absolute error values ​​between different receivers and improve the comparability of the evaluation results, M damage evaluation values ​​are used. The minimum average reconstruction error As a normalized benchmark, the damage assessment value To perform standardization:

[0139] (19);

[0140] in, Indicates the The damage assessment value after the normalization of the average reconstruction error of the receiver.

[0141] (5-3) M standardized damage assessment values Perform arithmetic averaging to obtain the average damage assessment value of the second panel in its current damage state. To quantify the degree of deviation of this average damage assessment value from the healthy state, this scheme introduces the normalized damage assessment value corresponding to the reconstruction error of the Lamb wave response signal obtained on the first panel in the healthy state as a benchmark, performs a second normalization process, and determines the comprehensive damage assessment score. :

[0142] (20);

[0143] in, The Lamb wave response signal collected by the i-th receiver on the first wall is used as the original signal and input into the corresponding trained deep autoencoder model in the same way as step (5-1) The average reconstruction error between the reconstructed signal and the original signal is , and then calculate the normalized damage assessment value using the same method as in step (5-2). It should be noted that in this step, the M receivers selected on the first wall panel are configured in the same manner as those on the second wall panel.

[0144] The resulting comprehensive injury assessment score The numerical value reflects the degree of deviation between the health status of the current second wall panel and the first wall panel; the larger the value, the more serious the damage.

[0145] In a specific embodiment of the present invention, for a block using orthogonal ply ( ) carbon fiber composite wall panel, the damage location and assessment experiment based on Lamb wave and deep autoencoder model was carried out using the method of the present invention; the size of the wall panel is 500×500×1.5mm 3 , and its equivalent mechanical parameters are shown in Table 1.

[0146] Table 1: Equivalent mechanical parameters of cross-ply carbon fiber composite panels;

[0147]

[0148] See attached Figure 5 , the dispersion curve of the orthogonal laminated composite panel in the 0° (fiber axis) direction is plotted; it can be seen from the figure that when the frequency-thickness product is 0~0.7MHz·mm, the S0 modal phase velocity value is relatively stable; while when the frequency-thickness product is 0.2~0.7MHz·mm, the A0 modal phase velocity value is relatively stable; when the frequency-thickness product exceeds 0.7MHz·mm, multiple modes appear, which complicates signal processing. Considering the thickness of the panel is 1.5mm, the center frequency is selected in this embodiment. The S0 mode Lamb wave at kHz is used as the parameter of the excitation signal.

[0149] See attached Figure 6 In this embodiment, the Lamb wave signal acquisition system uses piezoelectric ceramics (receivers) as the Lamb wave exciter (receiver 0) and receivers (receiver 1-receiver 8); the distance between two adjacent piezoelectric ceramics is L = 150 mm. The excitation signal is generated by the exciter, amplified by the power amplifier, and then loaded onto the wall panel surface. After the Lamb wave propagates through the wall panel, the Lamb wave response signal is collected by each receiver and transmitted to a computer via a data acquisition device for storage and analysis. The sampling frequency is set to 50 MHz, and the sampling time is 200 ; In the experiment, magnets were selected as additional mass to simulate wall panel damage, and different levels of damage were simulated according to the weight of the magnets.

[0150] See attached Figure 7 , showing the normalized Lamb wave response signal received by receiver 2 in the undamaged state; since the distance between the excitation and the receiver is L = 150 mm, the Lamb wave propagation speed along the main axis of the fiber is calculated according to formula (15) to be approximately =6000m / s; The Lamb wave propagation velocity distribution model derived based on formula (13) is shown in the attached figure. Figure 8 shown.

[0151] The Lamb wave response signal on the wall panel (i.e., the first wall panel mentioned above) in the healthy state is collected and the deep autoencoder model is constructed and trained. In this embodiment, the sampling times of each receiver are set to Second-rate.

[0152] In order to verify the positioning and evaluation capabilities of the proposed method, Figure 6 A set of damage cases is set in the damage detection area shown in the figure; the specific setting parameters of the damage cases are shown in Table 2. Among them, the sampling times of each receiver are set to Second-rate.

[0153] Table 2: Specific setting parameters of damage cases;

[0154]

[0155] According to step 3, calculate the Lamb wave flight time at each receiver in the above damage case (equivalent to the second wall panel mentioned above); see Figure 9 , which gives the Lamb wave response signals collected by each receiver under the above damage case The result after differential processing and continuous wavelet transform; according to Figure 5 The dispersion curve information in the S0 mode is the frequency-thickness product ( ) has the fastest propagation speed, so it is determined that the first wave packet in the wavelet transform result corresponds to the S0 mode. Figure 9 In the figure, the arrival time of the Lamb wave response signal received by each receiver is clearly marked; considering that the waveform and timing of the excitation signal are known, combined with the experimental settings, it can be known that the excitation time of the Lamb wave is approximately ; Based on this, the flight time of each receiver is calculated and the results are shown in Table 3.

[0156] Table 3: Lamb wave flight time calculations at various receivers;

[0157]

[0158] In the damage location process, based on the Lamb wave flight time obtained in Table 3, this embodiment selects receivers 1 and 2 as the main basis for damage location. The least squares method is combined with the geometric model equations to solve the objective function, and the position coordinates of the damage position are estimated to be mm; the result is consistent with the actual damage position D mm is close, with small error. Figure 10 ,The estimated damage location and the actual damage location were visually compared to verify the ,accuracy and effectiveness of the proposed method.

[0159] During the damage assessment process, for each level of damage, the Lamb wave response signals on the damaged panel (i.e., the second panel mentioned above) collected by receivers 1 and 2 were extracted. and , and input it into the trained deep autoencoder models DAE1 and DAE2, and the final comprehensive damage assessment score is As attached Figure 11 shown.

[0160] The experimental results show that for various levels of damage with small differences in damage degree simulated by magnets, the proposed method can effectively distinguish different damage degrees and exhibit good monotonicity, that is, the comprehensive damage assessment score gradually increases with the increasing damage level. This verifies the advantages of this method in quantitative damage assessment, such as high positioning accuracy, strong assessment sensitivity, and good ability to distinguish damage levels.

[0161] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A composite material panel damage identification and assessment method based on Lamb waves and autoencoders is characterized by: include: Arrange a Lamb wave signal acquisition system on the first wall panel in a healthy state; Under a preset excitation signal, receivers at different positions on the first wall panel are used to obtain Lamb wave response signals; based on the Lamb wave response signals obtained by each receiver, a deep autoencoder model corresponding to the receiver is trained; For the second wall panel to be tested, the Lamb wave signal acquisition system is arranged in the same manner as for the first wall panel, and the Lamb wave response signal is acquired using a receiver; wherein the second wall panel is the same type as the first wall panel; Calculating a differential signal using the Lamb response signals obtained from the first wall panel and the second wall panel, and determining the Lamb wave flight time at each receiver on the second wall panel using the differential signal; Based on the Lamb wave flight time at some receivers and in combination with the Lamb wave isotropic propagation velocity distribution model, a geometric model equation group is constructed; an objective function is constructed to solve the geometric model equation group, thereby locating the damage position in the second wall panel; Lamb wave response signals obtained by multiple receivers on the second wall panel are selected as original signals and input into the corresponding deep autoencoder models to obtain the reconstructed signals corresponding to each receiver. The original signals and the reconstructed signals are used to determine the average reconstruction error as the damage assessment value at each receiver and perform normalization. The comprehensive damage assessment score is determined based on the normalized damage assessment values ​​at the selected multiple receivers.

2. The composite material wall panel damage identification and assessment method based on Lamb wave and autoencoder according to claim 1 is characterized in that: The Lamb wave signal acquisition system includes an exciter and a plurality of receivers arranged around the exciter; a damage location coordinate system is constructed with the location of the exciter as the origin, and the positive direction of the 𝑥-axis is defined to be along the main axis of the fiber of the first wall panel, and the positive direction of the y-axis and the positive direction of the x-axis are in the same plane; The preset excitation signal adopts a single Lamb wave mode, and the center frequency is set in the non-dispersion section of the dispersion curve; the excitation signal is in the form of a sine pulse.

3. The composite material wall panel damage identification and assessment method based on Lamb wave and autoencoder according to claim 1 is characterized in that: The deep autoencoder model includes an encoder and a decoder; the encoder is used to map the Lamb wave response signal to a low-dimensional feature space to form a low-dimensional feature vector at the hidden layer; The decoder takes the low-dimensional feature vector at the hidden layer as input and outputs the reconstructed signal of the Lamb wave response signal; the training goal of the deep autoencoder model is to minimize the reconstruction error between the Lamb wave response signal and the reconstructed signal.

4. The composite material wall panel damage identification and assessment method based on Lamb wave and autoencoder according to claim 1 is characterized in that: Calculating a differential signal using Lamb response signals obtained from the first wall panel and the second wall panel, and determining the Lamb wave flight time at each receiver on the second wall panel using the differential signal, including: On the second wall The Lamb wave response signal collected by the receiver is , calculate the The average response signal at the receiver ; On the first wall panel The Lamb wave response signal collected by the receiver is , calculate the The average response signal at the receiver ; Calculate the The differential signal at each receiver ; For differential signals Apply complex Morlet wavelet transform, the wavelet transform result presents time-scale spectrum, extract the time corresponding to the peak of envelope spectrum as the first The arrival time of the Lamb wave response signal at each receiver ; Determine the second wall Lamb wave flight time at the receiver ; Indicates the start time of the stimulus signal.

5. The composite material wall panel damage identification and assessment method based on Lamb wave and autoencoder according to claim 1 is characterized in that: When locating the damage position in the second wall panel, first sort the Lamb wave flight times of all receivers on the second wall panel from short to long, and select the first Q Lamb wave flight times; Based on the selected Q Lamb wave flight times and the Lamb wave propagation velocity distribution model, a set of geometric model equations based on Lamb waves is constructed; The least squares method is used to construct the objective function to solve the geometric model equations; by minimizing the calculated value of the Lamb wave flight time from the damage position to the i-th receiver and the actual Lamb wave flight time The residual square sum of is used to obtain the optimal estimated position coordinates of the damage location.

6. The composite material wall panel damage identification and assessment method based on Lamb wave and autoencoder according to claim 5 is characterized in that: The geometric model equations are expressed as follows: ； Among them, the sixth equation of the equation group is the Lamb wave propagation velocity distribution model in all directions; In the above formula, represents the Lamb wave flight time at the i-th receiver among the first Q Lamb wave flight times selected. There is a damage location D on the second wall, and its position coordinates are ; The position coordinates of the i-th receiver are ; 、 They represent the propagation paths of the Lamb wave from the exciter to the damage location D and from the damage location D to the i-th receiver. 、 length, 、 Represents the propagation paths 、 The angle between the positive direction of the x-axis of the damage location coordinate system, 、 Lamb wave at the angle 、 Propagation velocity in the direction; dimensionless parameter Characterization and the second wall panel fiber main axis direction Lamb wave velocity in the angle direction Lamb wave velocity in the direction of the fiber main axis The ratio of Obtained through actual measurement.

7. The composite material wall panel damage identification and assessment method based on Lamb wave and autoencoder according to claim 6 is characterized in that: The least squares method is used to construct the objective function to solve the geometric model equations, thereby obtaining the optimal estimated position coordinates of the damage location; the objective function is expressed as follows: ； in, represents the optimal estimated position coordinates, 、 They represent the damage position D and the origin O of the damage location coordinate system, the damage position D and the i-th receiver respectively. The Euclidean distance between is the calculated value of the Lamb wave flight time.

8. The composite material wall panel damage identification and assessment method based on Lamb wave and autoencoder according to claim 6 is characterized in that: dimensionless parameters The expression is: ； in, is the equivalent Young's modulus of the damage location coordinate system on the second wall plate in the x-axis direction, is the equivalent Poisson's ratio along the y-axis on the plane whose normal is the x-axis; and They are respectively Equivalent Young's modulus and equivalent Poisson's ratio in the direction of the angle.

9. The composite material wall panel damage identification and assessment method based on Lamb wave and autoencoder according to claim 1 is characterized in that: The original signal and the reconstructed signal are used to determine the average reconstruction error as the damage assessment value at each receiver and perform normalization processing; Based on the normalized impairment assessment values ​​at the selected multiple receivers, a comprehensive impairment assessment score is determined, including: The damage assessment values ​​are normalized based on the minimum average reconstruction error among the damage assessment values ​​calculated based on the Lamb wave response signals collected by the selected multiple receivers; The multiple standardized damage assessment values ​​are arithmetic averaged to obtain the average damage assessment value of the second wall panel in its current damage state. The standardized damage assessment value corresponding to the reconstruction error of the Lamb wave response signal obtained on the first wall panel in the healthy state is introduced as a benchmark, and a second normalization process is performed to obtain the comprehensive damage assessment score.

10. A terminal device comprising a processor, a memory, and a computer program stored in the memory; characterized in that: When the processor executes the computer program, it implements the composite material wall panel damage identification and assessment method based on Lamb waves and autoencoders according to any one of claims 1 to 9.

Images