Lamb wave frequency dispersion characteristic sparse reconstruction method based on non-convex optimization

By introducing the generalized maxima and minima concave penalty of the generalized Huber function and the forward-backward splitting algorithm, the problem of poor sparse reconstruction effect of Lamb wave dispersion curve in the prior art is solved, and high-precision, noise-resistant sparse reconstruction is achieved, which is suitable for thin-walled structure detection in heavy machinery and aerospace fields.

CN121167937APending Publication Date: 2025-12-19BEIJING UNIV OF TECH

Patent Information

Application Number
CN202511697257.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-11-19
Publication Date
2025-12-19

AI Technical Summary

Technical Problem

Existing sparse reconstruction methods for Lamb wave dispersion curves based on convex regularity do not show significant reconstruction effects at high frequencies and high-order modes, and suffer from limited sparsity capabilities and underestimation of signal amplitude.

Method used

A sparse reconstruction method based on non-convex optimization of Lamb wave dispersion characteristics is adopted. The generalized maxima-minus concave penalty function of the generalized Huber function is introduced as a regularization term and solved by forward-backward splitting algorithm to construct a non-convex sparse dispersion curve estimation model.

Benefits of technology

It achieves high-quality dispersion curve reconstruction with fewer measurement points, improves detection accuracy and noise immunity, shortens response time, and is suitable for real-time detection of large-size structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121167937A_ABST
    Figure CN121167937A_ABST
Patent Text Reader

Abstract

The invention relates to a Lamb wave frequency dispersion characteristic sparse reconstruction method based on non-convex optimization, and belongs to the technical field of high-end equipment perception and test.The method comprises the steps that excitation is conducted on a U-shaped suspension arm, and time domain sampling signals at the junction of a U-shaped suspension arm structure are obtained; the time domain sampling signals are preprocessed; constructing a non-convex sparse frequency dispersion curve estimation model based on the preprocessed signal; and converting the non-convex sparse frequency dispersion curve estimation model into an optimization problem, searching a convex preserving condition of a target function, selecting a forward and backward splitting algorithm for solving, obtaining frequency-wave number distribution, and further extracting a frequency dispersion curve. The method solves the problems that a traditional convex regularization method is limited in induction sparse capability and underestimates signal amplitude. Through a non-convex optimization method, the purpose of accurate estimation of the frequency dispersion curve based on fewer measurement points is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of high-end equipment sensing and testing technology, and in particular to a sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization. Background Technology

[0002] Thin-walled and plate / shell structures (such as crane U-shaped booms, aircraft skins, and wind turbine blades) are critical load-bearing components in heavy machinery, aerospace, and other fields, and their structural health directly affects the safety and reliability of the equipment. Lamb waves, as elastic waves propagating in thin-plate or thin-walled structures, possess multimodal, dispersion characteristics, and long-distance propagation capabilities, and are widely used for defect detection in metal / composite material structures (such as weld cracks and localized corrosion). The dispersion curve of Lamb waves is the theoretical basis for their use in defect detection, determining parameters such as the probe wave frequency, mode, phase velocity, and group velocity, thus affecting detection accuracy. Since the propagation characteristics of Lamb waves are significantly affected by material parameters and structural geometry, structures in service environments (such as U-shaped booms and composite plates) may experience dispersion curve shifts due to material aging, fatigue damage, and other factors. Therefore, accurately estimating the Lamb wave dispersion characteristics of the tested structure at different stages through experimental measurement is crucial for determining the detection parameters for structural health monitoring and quantitative damage assessment.

[0003] Two-dimensional Fourier transform is a classic estimation method for experimentally measuring dispersion curves. It converts the time-space domain signal into the frequency-wavenumber domain to obtain the wavenumber dispersion curve of the ultrasonic guided wave. However, this method requires the acquisition of a large amount of dense data and an increased spatial sampling rate to obtain an accurate dispersion curve. Sparse reconstruction methods derived from two-dimensional Fourier transform can reconstruct dispersion curves with a smaller amount of data. However, most existing convex regularized sparse reconstruction methods for dispersion curves suffer from limited ability to induce sparsity and underestimation of signal amplitude, resulting in poor reconstruction performance at high frequencies and high-order modes. Summary of the Invention

[0004] Based on the aforementioned problems in the technical field, the technical objective of this invention is to propose a sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization; this method introduces the generalized Huber function into the maxima and minima concave penalty to replace the traditional method. A norm regularization term is introduced to establish a non-convex sparse dispersion curve estimation model, which solves the problems of limited sparsity induction and underestimation of signal amplitude in traditional convex regularization methods. Through non-convex optimization, the goal of accurate dispersion curve estimation based on a smaller number of measurement points is achieved.

[0005] To achieve the above objectives, the present invention provides the following solution: Sparse reconstruction methods for Lamb wave dispersion characteristics based on non-convex optimization include: Excitation is applied to the U-shaped boom to obtain the time-domain sampling signal at the junction of the U-shaped boom structure; The time-domain sampled signal is preprocessed; Based on the preprocessed signal, a non-convex sparse dispersion curve estimation model is constructed. The non-convex sparse dispersion curve estimation model is transformed into an optimization problem. The convexity preservation condition of the objective function is found, and the forward-backward splitting algorithm is selected to solve it, so as to obtain the frequency-wavenumber distribution and then extract the dispersion curve.

[0006] Optionally, preprocessing the time-domain sampled signal includes: The DC component is removed from the time-domain sampled signal, and then filtered, denoised, and smoothed. The smoothed signal is then standardized and scaled.

[0007] Optionally, based on the preprocessed signal, constructing a non-convex sparse dispersion curve estimation model includes: Based on the theory of ultrasonic guided wave propagation in a plate, a guided wave receiving signal model is constructed. Based on the guided wave receiving signal model, and combined with the sparsity of the preprocessed signal in the frequency wavenumber domain, a frequency domain sparse optimization model is constructed, and the sensing matrix and observation vector are designed according to the propagation characteristics of the Lamb wave in the plate. In the frequency domain sparse optimization model, a generalized maxima-minimum concave penalty function based on the generalized Huber function is introduced as a regularization term, which is defined as a non-convex sparse dispersion curve estimation model.

[0008] Optionally, the guided wave receiver signal model is: ; in, For the received frequency domain signal, For transfer functions, As an excitation signal, For noise, For the first l The distance between each receiving point and the excitation point For frequency, L This is the total measurement receiving point.

[0009] Optionally, the frequency domain sparse optimization model is: ; in, For the observation vector, For the perception matrix, For noise, To solve for the objective, namely the frequency-wavenumber distribution; The perception matrix is: ; in, i It is an imaginary number. e It is a natural constant. k It is the wave number.

[0010] Optionally, the generalized maxima-minus concave penalty function based on the generalized Huber function is: ; in, A A parameter matrix of the same dimension as the perception matrix. x For the target at a certain frequency, For vectors x of Norm, The norm is defined as the sum of the absolute values ​​of all elements in a vector. For generalized Huber functions, To The expression within the vector u Take the minimum value. u This is an auxiliary vector that has no actual physical meaning. For vectors u of Norm.

[0011] Optionally, the non-convex sparse dispersion curve estimation model is: ; in, For the expression on the right with respect to the vector x Take the minimum value. Y For the observation vector, as mentioned above Consistent, For the perception matrix With the target vector x The product of For regularization parameters, It is a generalized maxima and minima concave penalty function based on the generalized Huber function.

[0012] Optionally, the objective function is: ; The convexity preservation condition is: ; in, To preserve the convexity parameter, when 0 When the objective function is convex, This is the regularization parameter.

[0013] Optionally, choosing the forward-backward splitting algorithm to solve the problem includes: Each iteration requires two calculations: first, calculate the variables... x and auxiliary variables u The gradient is obtained, and the corresponding gradient descent is used as the forward gradient operator: calculate x The gradient is: ; in, For gradient operators, Let be the smooth, differentiable part of the objective function. This is the transpose of the perception matrix. For the perception matrix With the target vector x The product; calculate u The gradient is: ; The further forward operator is: ; ; in, v Let be the frequency domain signal vector to be determined. x The forward operator, w Auxiliary vector u The forward operator, The iteration step size, n Number of iterations; Then perform the calculation of the proximal operator and update the iteration value: ; in, For the first n The proximal operator in the next iteration, For the first n The target obtained by subgradient update x , The iteration step size, This represents the smooth, non-differentiable part of the objective function.

[0014] The beneficial effects of this invention are as follows: 1. Reduced number of measurement points: By using a non-convex sparse dispersion curve estimation model, this method can achieve high-quality dispersion curve reconstruction with a small number of measurement points during U-shaped boom dispersion curve testing. This not only significantly reduces sampling requirements but also reduces computational complexity and improves efficiency.

[0015] 2. High-precision, noise-resistant, sparse reconstruction: This invention integrates the piecewise penalty mechanism of the generalized Huber function with the concave penalty of the generalized maxima-minima concave penalty as a regularization term for estimating non-convex sparse dispersion curves. The piecewise penalty mechanism of the generalized Huber function effectively suppresses the interference of measurement noise on the reconstruction results, while the concave penalty of the generalized maxima-minima concave penalty avoids... The problem of amplitude deviation in regularization is solved, thereby achieving accurate sparse reconstruction of the dispersion characteristics of the U-shaped boom under strong noise interference.

[0016] 3. Enhance the quality of reconstructed modal images: In the signal preprocessing stage, the present invention adopts standardization to normalize the amplitude of the measured U-shaped boom ultrasonic guided wave signal, eliminating the difference in sensor dimensions; and after reconstruction, the frequency domain amplitude is subjected to power compression and normalization processing, which significantly enhances the higher-order modal components in the reconstructed dispersion curve.

[0017] 4. Reduced Response Time: By introducing a generalized maxima-minima concave penalty incorporating a generalized Huber function as a regularization term in the non-convex sparse dispersion curve estimation model, the response time for reconstructing the dispersion curve is reduced to that of traditional methods. The time required by the regularization method is about 4.5%, which enables real-time processing of U-shaped boom detection and is beneficial for the rapid measurement of dispersion curves of large-sized structures. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the experimental data acquisition system according to an embodiment of the present invention; Figure 3 The following is a comparison diagram of a measurement point signal before and after preprocessing in an embodiment of the present invention: where (a) is the original signal and (b) is the preprocessed signal; Figure 4 This is a dispersion curve diagram of the two-dimensional Fourier transform of an embodiment of the present invention; Figure 5 This is a reconstruction diagram of the non-convex sparse dispersion curves of 100 measurement points in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, this embodiment proposes a sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization, including: Excitation is applied to the U-shaped boom to obtain the time-domain sampling signal at the junction of the U-shaped boom structure; The time-domain sampled signal is preprocessed; Based on the preprocessed signal, a non-convex sparse dispersion curve estimation model is constructed. The non-convex sparse dispersion curve estimation model is transformed into an optimization problem. The convexity preservation condition of the objective function is found, and the forward-backward splitting algorithm is selected to solve it, so as to obtain the frequency-wavenumber distribution and then extract the dispersion curve.

[0023] This embodiment addresses the limitations of existing estimation methods in sparse reconstruction and the underestimation of signal amplitude. It improves sparse reconstruction performance by focusing on signal preprocessing, establishing and solving a non-convex sparse dispersion curve estimation model. The aim is to achieve accurate dispersion curve estimation using fewer measurement points. To further verify the feasibility of this model and algorithm in practical engineering environments, a piezoelectric excitation-laser reception Lamb wave detection experimental platform was built. Experimental signals were collected on a U-shaped boom structure, and the proposed method was used to estimate the dispersion curve of the noisy measured signal. Experimental results show that, compared to existing technologies, the proposed method can achieve accurate estimation of the dispersion curve of a U-shaped boom structure with fewer measurement points. It can be used to determine parameters such as probe wave frequency, mode, phase velocity, and group velocity for defect detection in U-shaped booms, providing a theoretical basis for further Lamb wave defect detection in U-shaped booms.

[0024] Specifically, in this embodiment, the sparse reconstruction method based on non-convex optimization of Lamb wave dispersion characteristics includes the following steps: S1. Signal Acquisition. Design a piezoelectric excitation-laser receiving experimental system with one excitation and multiple receivers. Excite the laser at one point on the U-shaped boom and receive it at multiple points along the junction. Obtain the time-domain signals from different receiving points at the junction of the U-shaped boom structure as the experimental data set.

[0025] Specifically, such as Figure 2 As shown, the entire experimental data acquisition system mainly consists of an arbitrary function generator, a power amplifier, a laser ultrasonic detection and receiving system, and the detection object (U-shaped boom). The piezoelectric transducer is positioned at the excitation point as the signal transmitter, and the laser receiver acquires experimental data point by point along the axial direction, obtaining a series of one-dimensional time signals, such as... Figure 3 The image shows the received time-domain signal at a certain measurement point.

[0026] S2. Signal preprocessing. The DC component and noise effects are removed from the received time-domain signals from different measurement points, and then standardized.

[0027] Specifically, the experimental signals acquired in the experiment are preprocessed. For example... Figure 3 The image shows a comparison between the raw time-domain signal received at a certain measurement point and the preprocessed time-domain signal. This includes removing the DC component from the received time-domain signal, performing filtering, noise reduction, and smoothing, and then standardizing and scaling the processed signal. Figure 3 (a) is the original signal. Figure 3 (b) is the preprocessed signal.

[0028] Specifically, the data is standardized and scaled: ; in, s This represents the time-domain signal received at each measurement point. σ represents the time-domain signal after normalization and scaling, μ represents the average amplitude of the received time-domain signal at each measurement point, and σ represents the standard deviation of the received time-domain signal at each measurement point.

[0029] Furthermore, based on the preprocessed signal, the non-convex sparse dispersion curve estimation model is constructed as follows: Based on the theory of ultrasonic guided wave propagation in a plate, a guided wave receiving signal model is constructed. Based on the guided wave receiving signal model, and combined with the sparsity of the preprocessed signal in the frequency wavenumber domain, a frequency domain sparse optimization model is constructed, and the sensing matrix and observation vector are designed according to the propagation characteristics of the Lamb wave in the plate. In the frequency domain sparse optimization model, a generalized maxima-minimum concave penalty function based on the generalized Huber function is introduced as a regularization term, which is defined as a non-convex sparse dispersion curve estimation model.

[0030] Specifically, in this embodiment, S3, a non-convex sparse dispersion curve estimation model is constructed. A sensing matrix is ​​designed based on the propagation characteristics of Lamb waves, and 100 randomly selected time-domain signals are converted to the frequency domain and used as observation vectors. The sparsity of ultrasonic guided waves in the frequency-wavenumber domain is utilized to construct a non-convex sparse dispersion curve estimation model.

[0031] Specifically, based on the theory of ultrasonic guided wave propagation in a plate, a guided wave receiving signal model is constructed. Combining the sparsity of the signal in the frequency-wavenumber domain, a general form of a frequency-domain sparse optimization model is constructed, and the sensing matrix and observation vector are designed. To enhance the model's ability to induce sparsity, a generalized maxima-minimum concave penalty of the generalized Huber function is introduced as a regularization term in the general form of the sparse optimization model, defined as a non-convex sparse dispersion curve estimation model.

[0032] Specifically, the guided wave receiver signal model: ; in, It is the received frequency domain signal. It is a transfer function. It is an excitation signal. It's noise. It is the first l The distance between each receiving point and the excitation point It's frequency. L It is the main measurement receiving point.

[0033] Furthermore, the frequency domain sparse optimization model can be represented in matrix-vector form: ; in, For the observation vector, For the perception matrix, For noise, The objective is to find the frequency-wavenumber distribution.

[0034] Based on the propagation characteristics of Lamb waves in the plate, the sensing matrix is ​​designed as follows: ; in, i It is an imaginary number. e It is a natural constant. k It is the wave number.

[0035] Solving the objective can be transformed into... Unconstrained minimization problem under norm, where Norm is the regularization term in an optimization problem: ; in, It is an observation vector at a certain frequency. x It is the target at a certain frequency. l 1-norm, l 2-norm, This is the regularization parameter.

[0036] However Norm-induced sparsity has limited ability, and the sparsity of the solution is far lower than that of the actual sparse solution. Therefore, this invention adopts a non-convex regularization that is closer to the actual sparse solution, namely, a minimax concave penalty function based on the scaled Huber function. : ; in, It's a Huber function. a For scaling parameters, u An auxiliary variable with the same dimension as the signal.

[0037] The experimental data involves multiple measurement points, further extending to a multivariate problem. Therefore, a generalized maxima-minima concavity penalty is proposed, incorporating the generalized Huber function, defined as: ; in, A It is a parameter matrix of the same dimension as the perception matrix.

[0038] Replace the traditional method with a generalized maxima minima concave penalty. Using the norm as a regularization term, we obtain the estimation model for non-convex sparse dispersion curves: ; In this embodiment, S4 involves designing an optimized solution algorithm. The constructed non-convex sparse dispersion curve estimation model is transformed into an optimization problem. The convexity preservation condition of the objective function is sought, and the forward-backward splitting algorithm is selected for solving the problem to obtain the frequency-wavenumber distribution, thereby extracting the dispersion curve.

[0039] Specifically, we obtain a non-convex sparse optimization problem with the objective function as follows: ; It is important to note that non-convex optimization problems are prone to getting trapped in local optima. Therefore, it is necessary to find conditions that maintain the convexity of the objective function. To obtain the globally optimal solution: ; in, It is the transpose of the matrix. yes x The affine function only needs the first term. If the objective function is positive semi-definite, then it can be guaranteed to be convex. Therefore, the convexity-preserving condition for the objective function is set as follows: ; in, To preserve the convexity parameter, when 0 When the objective function is convex, This is the regularization parameter.

[0040] Based on the objective function of the established non-convex sparse dispersion curve estimation model, the forward and proximal operators of the forward-backward splitting algorithm are designed. First, the objective function is decomposed into a smooth, differentiable part. and smooth non-differentiable parts : ; ; The optimization problem for estimating nonconvex sparse dispersion curves can then be rewritten as: ; Specifically, the iterative steps for solving the objective function using the forward-backward split algorithm are shown below. This algorithm requires two calculations in each iteration: first, calculating the variables... x and auxiliary variables u The gradient is obtained, and the corresponding gradient descent is used as the "forward" gradient operator: calculate x The gradient is: ; in, This is the gradient operator.

[0041] calculate u The gradient is: ; The further forward operator is: ; ; in, v Let be the frequency domain signal vector to be determined. x The forward operator, w Auxiliary vector u The forward operator, The iteration step size, n Number of iterations.

[0042] Then perform the calculation of the proximal operator and update the iteration value: ; S5. Application to U-shaped booms. Select a portion of the measurement points of the experimentally measured time-domain sampling signal of the U-shaped boom structure as input, and use the method mentioned in this embodiment to perform sparse reconstruction of the Lamb wave dispersion curve. Add parameter adjustments and continuously update and optimize the model parameters to obtain a better-reconstructed dispersion curve with fewer observations while more closely approximating the actual sparsity capability and without underestimating the signal amplitude.

[0043] Specifically, in this embodiment, 801 receiving points are set up, and the time-domain signals of 100 measurement points are randomly selected to perform sparse reconstruction of the Lamb wave dispersion curve.

[0044] S6. Verify the correctness of the model and algorithm. Perform a two-dimensional Fourier transform on the time-domain signals (time-space) of all measurement points, plot the dispersion curves, and compare them with the dispersion curves obtained by the non-convex sparse dispersion curve estimation model and the forward-backward splitting algorithm proposed in this embodiment to verify the correctness of the model and algorithm.

[0045] Specifically, a two-dimensional Fourier transform is performed on the preprocessed time-domain signal to convert the time-space domain distributed signal into a frequency-wavenumber domain distributed signal, obtaining the dispersion curve under the two-dimensional Fourier transform. This dispersion curve is then compared with the dispersion curve reconstructed by the non-convex sparse dispersion curve estimation model. The formula for the two-dimensional Fourier transform is: ; in, This is a time-domain signal, where r represents distance and t represents time. denoted by frequency, and k represents wave number.

[0046] like Figure 4 and Figure 5 As shown, a comparison between the two-dimensional Fourier transform dispersion curve of all measurement points and the dispersion curve reconstructed by selecting 100 measurement points as the observation matrix from the non-convex sparse dispersion curve estimation model shows that the non-convex sparse dispersion curve estimation model can completely, clearly and correctly reconstruct the dispersion curve using only 12.5% ​​of the data from all measurement points.

[0047] The proposed non-convex sparse dispersion curve estimation model: This model proposes using a generalized maxima-minima concave penalty incorporating a generalized Huber function as the regularization term, replacing the traditional method. The norm, as a regularization term, enhances the sparsity of the sparse reconstruction model.

[0048] Solving the Non-Convex Sparse Dispersion Curve Estimation Model: To prevent the solution of the non-convex sparse dispersion curve estimation model from getting trapped in local optima, we seek the convexity preservation condition of the objective function of the non-convex sparse dispersion curve estimation model. We use the forward-backward split algorithm to solve the objective function, which improves the stability of numerical computation, reduces the iteration time, and at the same time broadens the frequency range of sparse reconstruction and improves the accuracy of sparse reconstruction.

[0049] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A sparse reconstruction method based on non-convex optimization of Lamb wave dispersion characteristics, characterized in that, include: Excitation is applied to the U-shaped boom to obtain the time-domain sampling signal at the junction of the U-shaped boom structure; The time-domain sampled signal is preprocessed; Based on the preprocessed signal, a non-convex sparse dispersion curve estimation model is constructed. The non-convex sparse dispersion curve estimation model is transformed into an optimization problem. The convexity preservation condition of the objective function is found, and the forward-backward splitting algorithm is selected to solve it, so as to obtain the frequency-wavenumber distribution and then extract the dispersion curve.

2. The sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to claim 1, characterized in that, Preprocessing the time-domain sampled signal includes: The DC component is removed from the time-domain sampled signal, and then filtered, denoised, and smoothed. The smoothed signal is then standardized and scaled.

3. The sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to claim 1, characterized in that, Based on the preprocessed signal, the model for estimating nonconvex sparse dispersion curves includes: Based on the theory of ultrasonic guided wave propagation in a plate, a guided wave receiving signal model is constructed. Based on the guided wave receiving signal model, and combined with the sparsity of the preprocessed signal in the frequency wavenumber domain, a frequency domain sparse optimization model is constructed, and the sensing matrix and observation vector are designed according to the propagation characteristics of the Lamb wave in the plate. In the frequency domain sparse optimization model, a generalized maxima-minimum concave penalty function based on the generalized Huber function is introduced as a regularization term, which is defined as a non-convex sparse dispersion curve estimation model.

4. The sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to claim 3, characterized in that, The guided wave receiver signal model is as follows: ; in, For the received frequency domain signal, For transfer functions, As an excitation signal, For noise, For the first l The distance between each receiving point and the excitation point For frequency, L This is the total measurement receiving point.

5. The sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to claim 3, characterized in that, The frequency domain sparse optimization model is as follows: ; in, For the observation vector, For the perception matrix, For noise, To solve for the objective, namely the frequency-wavenumber distribution; The perception matrix is: ; in, i It is an imaginary number. e It is a natural constant. k It is the wave number.

6. The sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to claim 3, characterized in that, The generalized minimax concave penalty function based on the generalized Huber function is: ; in, A A parameter matrix of the same dimension as the perception matrix. x For the target at a certain frequency, For vectors x of Norm, The norm is defined as the sum of the absolute values ​​of all elements in a vector. For generalized Huber functions, To The expression within the vector u Take the minimum value. u This is an auxiliary vector that has no actual physical meaning. For vectors u of Norm.

7. The sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to claim 6, characterized in that, The estimation model for the non-convex sparse dispersion curve is as follows: ; in, For the expression on the right with respect to the vector x Take the minimum value. Y For the observation vector, For the perception matrix With the target vector x The product of For regularization parameters, It is a generalized maxima and minima concave penalty function based on the generalized Huber function.

8. The sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to claim 7, characterized in that, The objective function is: ; The convexity preservation condition is: ; in, To preserve the convexity parameter, when 0 When the objective function is convex, This is the regularization parameter.

9. The sparse reconstruction method for Lamb wave dispersion characteristics based on non-convex optimization according to claim 1, characterized in that, Choosing the forward-backward splitting algorithm for solving the problem includes: Each iteration requires two calculations: first, calculate the variables... x and auxiliary variables u The gradient is obtained, and the corresponding gradient descent is used as the forward gradient operator: calculate x The gradient is: ; in, For gradient operators, Let be the smooth, differentiable part of the objective function. This is the transpose of the perception matrix. For the perception matrix With the target vector x The product; calculate u The gradient is: ; The further forward operator is: ; ; in, v Let be the frequency domain signal vector to be determined. x The forward operator, w Auxiliary vector u The forward operator, The iteration step size, n Number of iterations; Then perform the calculation of the proximal operator and update the iteration value: ; in, For the first n The proximal operator in the next iteration, For the first n The target obtained by subgradient update x , The iteration step size, This represents the smooth, non-differentiable part of the objective function.

Citation Information

Patent Citations

  • Millimeter wave three-dimensional sparse imaging method based on generalized MCP constraint

    CN115469309A

  • Bearing fault diagnosis method and system based on non-convex optimization low-rank sparse decomposition

    CN116842470A

  • Composite material plate Lamb wave frequency dispersion characteristic sparse reconstruction method based on characteristic optimization

    CN120009420A

  • Method to estimate the rise-time of a pulse for single and multi-channel data

    US11156647B1

Cited By

  • Guide wave field damage imaging method based on two-dimensional sparse multi-frequency wave number spectrum reconstruction

    CN121703263A