Lamb wave modal blind separation method based on time-frequency space independent component

Abstract

A blind mode separation method for Lamb waves based on independent time-frequency spatial components includes: performing non-destructive testing of the test object using Lamb waves and acquiring signals using a single-channel piezoelectric sensor; performing independent component analysis on the time-domain and frequency-domain components based on the two-dimensional time-frequency spatial components to obtain time-domain atoms and frequency-domain atoms, and sorting them by a correlation index α; selecting the atoms other than the maximum and minimum values ​​from the sorted list as time- and frequency-domain reconstructed atoms; and performing time- and frequency-domain reconstructed atoms T... base ,F base The time-frequency spatial reconstruction component (TFR) of the signal is obtained, and the distance index D of the time-frequency spatial reconstruction component is established. Euk The component distance index D is reconstructed through time-frequency spatial reconstruction. Euk C-MEANS fuzzy clustering is performed on the time-frequency spatial reconstruction component (TFR) to separate and obtain time-frequency spatial sub-components; inverse time-frequency analysis is then performed on the time-frequency spatial sub-components to obtain the separated signals. This disclosure can effectively perform blind separation of overlapping modes with dispersion effects.

Description

Technical Field

[0001] This disclosure belongs to the technical field of signal processing technology, specifically relating to a blind separation method for Lamb wave modes based on time-frequency spatially independent components. Background Technology

[0002] Structural health monitoring is of significant scientific importance and urgent engineering need for ensuring the safe operation of critical equipment and timely detection of potential damage or defects. Unlike passive structural health monitoring technologies that obtain structural health information by analyzing operating parameters, active structural health monitoring technologies achieve online acquisition of structural health status information and direct detection of structural damage by transmitting excitation signals to the structure under test and processing the response signals. Among various excitation signals, ultrasonic guided waves have advantages such as long propagation distance, low energy attenuation, and high damage sensitivity, and are therefore widely used in active structural health monitoring. When ultrasonic guided waves propagate in a plate structure, they are repeatedly reflected and refracted on the upper and lower surfaces of the plate, forming plate waves, or Lamb waves, through waveform conversion and coupling superposition. However, the dispersion characteristics of Lamb waves cause their wave packet energy to disperse, and their multimodal characteristics cause wave packets of different modes to overlap. In recent years, various dispersion compensation techniques and various mode separation techniques have been proposed.

[0003] Ultrasonic guided waves are widely used in nondestructive testing (NDT) because they can detect potential damage or defects without damaging the structure. However, the dispersion and multimodal nature of ultrasonic guided waves can complicate the detection of received signals, leading to a degraded performance of detection algorithms. To address this issue, an efficient mode separation algorithm needs to be developed. Traditional ultrasonic mode separation methods mainly utilize prior time-frequency transforms (TFRs). Time signals are described as trajectories with different intrinsic wave numbers, and various types of TFR methods have been widely used to process and remove multimodal data in Lamb wave signal processing. These methods can effectively separate multiple modes of Lamb waves by analyzing the signal in the time-frequency domain, thereby improving the performance of detection algorithms. Furthermore, blind source separation (BSS) is also a commonly used signal processing method. One BSS method is Independent Component Analysis (ICA), which has been widely applied in various mixed signal processing applications, including voice verification, EEG source localization, communication systems, and fault analysis. ICA achieves signal separation by maximizing the statistical independence between signal sources, thus playing an important role in various applications.

[0004] Currently, existing ultrasonic guided wave mode separation methods mainly include time-frequency analysis and matched pursuit. Time-frequency analysis is constrained by the uncertainty principle, meaning its time-domain and frequency-domain accuracy cannot be improved simultaneously because different modes cannot be well separated on the time-frequency plane, thus limiting the accuracy and application range of this type of method. Matched pursuit requires a manually given sparsity parameter, but this parameter is practically impossible to obtain prior knowledge of, and the matched pursuit method has poor anti-interference capabilities. In practical applications, these algorithms can lead to incomplete separation or low separation accuracy, limiting the development and application of ultrasonic guided wave-based structural health monitoring technology.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of the present invention, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present disclosure aims to provide a blind single-channel mode separation technique that separates modes without prior knowledge of the dispersion curve of the board. Unlike traditional independent component analysis (ICA) methods that require multi-channel signals, this method uses only a single-channel signal as input. Due to the selection of the FastICA basis, the separation result depends on the choice of time and frequency atoms. This method not only separates different modes but also obtains non-dispersive signals with maximum energy and accurate arrival time. Therefore, this method can effectively process signals with limited computation time.

[0007] To achieve the above objectives, this disclosure provides the following technical solutions:

[0008] A blind Lamb wave mode separation method based on time-frequency spatially independent components includes the following steps:

[0009] Step S100: Lamb wave non-destructive testing is performed on the object under test, and a single-channel piezoelectric sensor is used to acquire signals. The acquired signals include multi-channel multi-mode signals with raw Lamb wave signals. Time-frequency analysis is performed on the acquired raw Lamb wave signals to obtain two-dimensional time-frequency spatial components.

[0010] Step S200: Based on the two-dimensional time-frequency spatial components, perform independent component analysis on the time-domain components and frequency-domain components respectively to obtain time-domain atoms and frequency-domain atoms. Sort the time-domain atoms and frequency-domain atoms according to the correlation index α, and select the atoms other than the maximum and minimum values ​​from the sorted list as time-domain and frequency-domain reconstructed atoms: Time-domain reconstructed atom T base Frequency domain reconstruction of atom F base ;

[0011] Step S300: Reconstruct atom T based on time and frequency domains base,F base The time-frequency spatial reconstruction component (TFR) of the signal is obtained, and the distance index D of the time-frequency spatial reconstruction component is established. Euk The component distance index D is reconstructed through time-frequency spatial reconstruction. Euk C-MEANS fuzzy clustering was performed on the time-frequency spatial reconstruction components (TFR) to separate and obtain the time-frequency spatial sub-components;

[0012] Step S400: Perform inverse time-frequency analysis on the time-frequency spatial sub-components to obtain the separated signals.

[0013] In the method described, the time-frequency analysis employs the short-time Fourier transform:

[0014]

[0015] Where w(t) is the window function, τ is the short-time Fourier time, x(t) is the original time-domain signal of the Lamb wave, and e -j2πfτ It is a Fourier transform basis, where j is the imaginary part label, f is the frequency, τ is the short-time Fourier time, and π is the constant of pi. After the short-time Fourier transform, the original Lamb wave signal becomes a two-dimensional signal in the time and frequency domains, where t represents the time domain and f represents the frequency domain.

[0016] In the method described, the window function is a square window.

[0017] In the method described, the excitation signal of the piezoelectric sensor is a toneburst signal.

[0018] In the method described, step S200, the independent component analysis for a single dimension includes p independent components in n channels:

[0019] X = AS

[0020] Where X is the received multi-channel multimodal signal, S consists of p independent components, and A is the mixing matrix, which is obtained by estimating the mixing process:

[0021]

[0022] in To estimate the independent components of a multi-channel, multi-modal signal, where W is the separation matrix, the FastICA algorithm is used for independent component estimation. This is achieved by comparing the entropy between the signal and a pure Gaussian quantity. Since Gaussian random variables are the most random of all distributions, the difference between the entropy of the Gaussian random variable and the estimated variable yields a non-negative quantity J, defined as negative entropy.

[0023] J(s)=H(s gauss )-H(s),

[0024] The FastICA algorithm is simplified using kurtosis, and higher-order cumulants are used in conjunction with density polynomial expansion. Furthermore, the negative entropy is approximated using a non-quadratic expectation:

[0025]

[0026] Where v is a Gaussian random variable, w and x are elements in the separation matrix and the original signal matrix, and G(*) = (*). 4 / 4.

[0027] In the method described, the transformed original time-frequency domain signal x(t,f) is selected and subjected to FastICA analysis in both the time and frequency domains to obtain m and n independent components, respectively, where T i For time-domain components, F i As frequency domain components, from m and n independent components, a subset of components are extracted as reconstructed atoms, with the following indices:

[0028]

[0029] Correlation analysis was performed on the time-frequency domain atoms and the time-frequency signals in the original signal domain, and the atoms were arranged according to their correlation. The atoms in the middle part were selected as the corresponding time-domain reconstruction atoms and frequency-domain reconstruction atoms.

[0030] In the method described, six time-domain atoms and six frequency-domain atoms located in the middle part are selected as the corresponding time-domain reconstruction atoms and frequency-domain reconstruction atoms.

[0031] In the method described, based on the selected temporal reconstruction atom T base and frequency domain reconstruction of atom F base The time-frequency spatial reconstruction component TFR is obtained:

[0032]

[0033] Where, x rec (t,f) represents the reconstructed time-frequency signal domain; t represents time; i represents the index, and there are a total of x+y time-frequency domain representations, where x is T. base The number of y is F base The number.

[0034] In the method described, fuzzy clustering is applied to separate different TFRs.

[0035] x rec (t,f)=∑TFR modeA0 +∑TFR modeS0 Where modeA0 represents the TFR of mode A0; modeS0 represents the TFR of mode S0; the fuzzy clustering process is implemented by the fuzzy C-means algorithm, which is as follows:

[0036]

[0037] Where V = [v1, v2, ..., v c ],v i ∈R n , is a category vector. It is the distance norm of the square inner product, U = [μ] ik ] is the fuzzy classification result matrix, where μ ik This represents the membership degree of the i-th data point to the k-th cluster.

[0038] In the method described, inverse time-frequency analysis is performed on the separated time-frequency spatial reconstruction component (TFR) to obtain the separated signal:

[0039]

[0040] in,

[0041]

[0042] Where X(τ,ω) represents all time-frequency domain signals, τ is the short-time Fourier time, ω is the frequency, and e j2πfτ It is a Fourier basis, where j is the imaginary part label, f is the frequency, π is the constant of pi, x(t) is the original time-domain signal of the recovered Lamb wave, and t represents the time domain. After the inverse short-time Fourier transform, the Lamb wave reconstructs a two-dimensional signal into a one-dimensional recovered signal in the time domain.

[0043] Compared with existing technologies, the advantages of this disclosure are: it provides a blind single-channel mode separation technique that separates modes without prior knowledge of the dispersion curve of the plate. Unlike traditional independent component analysis (ICA) methods that require multi-channel signals, this method only requires a single-channel signal as input. Due to the selection of the FastICA basis, the separation result depends on the choice of time and frequency atoms. This method can not only separate different modes but also obtain non-dispersive signals with maximum energy and accurate arrival time. Attached Figure Description

[0044] Figure 1 This is a flowchart of a blind separation method for Lamb wave modes based on time-frequency spatially independent components provided in one embodiment of this disclosure;

[0045] Figure 2 This is a schematic diagram of the time-frequency analysis process for a dual-mode excitation signal with a propagation distance of 600mm and a center frequency of 500kHz, provided in one embodiment of this disclosure.

[0046] Figure 3This is a schematic diagram of the time-frequency analysis process for a dual-mode aliasing signal with a propagation distance of A0 (600mm), S0 (300mm), and a center frequency of 500kHz, provided in an embodiment of this disclosure.

[0047] Figure 4 This is a flowchart of time-frequency atom generation for FastICA provided in one embodiment of the present disclosure;

[0048] Figure 5 This is a schematic diagram of FastICA time-frequency atom correlation arrangement and screening for a dual-mode excitation signal with a propagation distance of 600mm and a center frequency of 500kHz, provided in one embodiment of this disclosure.

[0049] Figure 6 This is a schematic diagram of FastICA time-frequency reconstruction atoms for a dual-mode excitation signal with a propagation distance of 600mm and a center frequency of 500kHz, provided in one embodiment of this disclosure. The atoms are six frequency domain atoms and six time domain atoms.

[0050] Figure 7 This is a schematic diagram of the fuzzy C-clustering result of the reconstructed time-frequency representation (TFR) of a dual-mode excitation signal with a propagation distance of 600 mm and a center frequency of 500 kHz, provided in an embodiment of this disclosure;

[0051] Figures 8(a) to 8(c) This is a schematic diagram of the inverse time-frequency transformation of the reconstructed time-frequency representation (TFR) of a dual-mode excitation signal with a propagation distance of 600 mm and a center frequency of 500 kHz, provided in one embodiment of the present disclosure. Figure 8(a) shows the separated S0 signal, Figure 8(b) shows the separated A0 signal, and Figure 8(c) shows the original received signal.

[0052] Figure 9 This is a schematic diagram of the fuzzy C-clustering result of the reconstructed time-frequency representation of a dual-mode aliasing signal with a propagation distance of A0 (600mm), S0 (300mm), and a center frequency of 500kHz, provided in an embodiment of this disclosure;

[0053] Figures 10(a) to 10(c) This is a mode separation result of a dual-mode aliasing signal with a propagation distance of A0 (600mm), S0 (300mm), and a center frequency of 500kHz provided in an embodiment of this disclosure. Figure 10(a) shows the separated S0 signal, Figure 10(b) shows the separated A0 signal, and Figure 10(c) shows a schematic diagram of the original received signals A0 (600mm) and S0 (300mm).

[0054] Figure 11 It is 1000×1000×2mm 3An experimental plate with a propagation distance of 600mm was used on a T6061 aluminum alloy sheet. The diagrams show the three shortest paths from the emitted PZT to the received PZT, namely D1, D2, and D3.

[0055] Figure 12(a) shows Figure 11 The schematic diagram of the received signal after boundary reflection is shown in Figure 12(b). The three wave packets are separated in this embodiment. The separated S0 mode can be seen in Figure 12(c), and the separated A0 mode can be seen in Figure 12(d). Detailed Implementation

[0056] The following will refer to the appendix. Figure 1 Specific embodiments of the present disclosure are described in detail with reference to Figure 12(d). While specific embodiments of the present disclosure are shown in the figures, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0057] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this disclosure is determined by the appended claims.

[0058] To facilitate understanding of the embodiments of this disclosure, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. The accompanying drawings do not constitute a limitation on the embodiments of this disclosure.

[0059] like Figures 1 to 12(d) As shown, a Lamb wave mode blind separation method based on time-frequency spatially independent components includes the following steps:

[0060] Step S100: Acquire the PZT signal transmitted and received via a single channel (SISO), with the excitation signal being a toneburst signal. The acquired signal contains multi-channel, multi-mode signals. Perform time-frequency analysis (STFT) on the acquired raw Lamb wave signal to obtain its two-dimensional time-frequency spatial components.

[0061] Step S200: Perform independent component analysis on the time-domain and frequency-domain components respectively to obtain time-domain atoms and frequency-domain atoms. Sort the time-domain and frequency-domain atoms using the correlation index α, and select atoms far from the extreme values ​​as the time-frequency domain reconstruction atoms T. base ,F base ;

[0062] Step S300: Reconstruct the time-frequency domain atoms to obtain the time-frequency spatial reconstruction component (TFR) of the signal. Establish the distance index D of the time-frequency spatial reconstruction component. Euk By performing C-MEANS fuzzy clustering on the reconstructed time-frequency spatial components, the separation of time-frequency spatial sub-components is achieved.

[0063] Step S400: Perform inverse time-frequency analysis on the separated reconstructed components to obtain the separated signal.

[0064] Preferably, the time-frequency transformation of the signal in step S100 can be represented as follows:

[0065]

[0066] Here, w(t) is the window function, and a square window is chosen. x(t) is the original time-domain signal. After short-time Fourier transform, the original time-domain signal becomes a two-dimensional signal in the time-frequency domain.

[0067] Preferably, step S200 includes the following steps.

[0068] S201: Perform independent component analysis on the signals of both dimensions separately. The process of independent component analysis for a single dimension is as follows: Assume there are p independent components in n channels:

[0069] X = AS

[0070] Where X is the received multi-channel multimodal signal, S is composed of p independent components, and A is the mixing matrix.

[0071] Preferably, in step S200, by estimating the mixing process, the following can be obtained:

[0072]

[0073] in For the independent component estimation of a multi-channel, multi-modal signal, W is the separation matrix. The estimation process uses the FastICA algorithm, which compares the entropy between the signal and a pure Gaussian quantity. Since the Gaussian random variable is the most random of all distributions, the difference between the entropy of the Gaussian random variable and the estimated variable yields a non-negative quantity J, defined as the negative entropy:

[0074] J(s)=H(s gauss )-H(s)

[0075] Step S202: The ICA algorithm is simplified using kurtosis. Since kurtosis is sensitive to edge values ​​and has low robustness, a higher-order cumulant is used in conjunction with a density polynomial expansion, and the negative entropy is approximated using a general form of non-quadratic expectation.

[0076]

[0077] Where v is a Gaussian random variable, and w and x are elements in the separation matrix and the original signal matrix, respectively.

[0078] Step S203: Here, a reasonable choice of non-quadratic function G is needed. To ensure robustness, the following is chosen:

[0079] G(*)=(*) 4 / 4.

[0080] Step S204: From the m and n independent components, extract a portion of the components as reconstructed atoms. Where T base For time-domain atoms, F base These are frequency domain atoms. The extracted indices are as follows:

[0081]

[0082] Correlation analysis was performed on the time-frequency domain atoms and the original signal domain time-frequency signals, and they were arranged according to their correlation. Atoms located in the middle part were selected as reconstruction atoms. The number of atoms was determined by the number of modes to be separated.

[0083] Preferably, in step S300, the selected time-domain atom T base and frequency domain atom F base Reconstruction yields a series of time-frequency domain representations (TFR):

[0084]

[0085] Where, x rec (t,f) represents the reconstructed time-frequency signal domain; t represents time; i represents the index, and there are a total of x+y time-frequency domain representations, where x is T. base The number of y is F base The number of; TFR is the time-frequency domain representation.

[0086] Preferably, in S300, fuzzy clustering separates different TFRs.

[0087] x rec (t,f)=∑TFR modeA0 +∑TFR modeS0

[0088] Preferably, in S400, the time-frequency representation signals of different modes are reconstructed to obtain time-domain signals.

[0089]

[0090] The parameters used in the inverse short-time Fourier transform are the same as those used in step S100.

[0091] In one embodiment, such as Figure 1 As shown, a blind Lamb wave mode separation method based on time-frequency spatially independent components includes the following steps:

[0092] S100: Lamb wave nondestructive testing typically involves single-input single-output (SISO) testing of piezoelectric transducers (PZTs), meaning only a single channel of signal is observed. However, to analyze multi-channel signals using ICA, multiple experiments or channels are usually required to increase the number of observations. This adds extra workload for single-channel test signals. Furthermore, no corresponding blind separation algorithm has been proposed for single-channel test signals. Signal filtering and compensation are only performed based on mode dominance at specific frequencies.

[0093] This step first requires performing a short-time Fourier transform (STFT) on the original time signal, selecting a square window function to divide the time-domain signal into discrete time and frequency domains. During the transform, the time resolution and frequency resolution need to be matched, with a corresponding number of discrete atoms being roughly equal.

[0094] For example, 1000×1000×2mm 3 After acquiring a toneburst signal with a center frequency of 500kHz and a propagation distance of 600mm on a T6061 aluminum alloy sheet, two modes, A0 and S0, were obtained. At this point, the two modes did not overlap. Different modes are often used for detection depending on the damage mode. Simulation 1 uses two modes with the same propagation distance, while Simulation 2 uses two modes with different propagation distances, resulting in mode overlap. After time-frequency analysis, the matrix size is 500*420, as shown below. Figure 2 As shown, its time-frequency diagram is plotted. It can be seen that the S0 mode exhibits some dispersion.

[0095] For example, 1000×1000×2mm 3After acquiring a toneburst signal with a center frequency of 500kHz from a T6061 aluminum alloy plate, two modes, A0 and S0, were obtained. To obtain aliased signals, two PZTs were used for transmission and one PZT for reception, ensuring that both A0 (propagation distance 300mm) and S0 (propagation distance 600mm) signals arrived at the receiving PZT simultaneously. The result was as follows: Figure 3 Plot the time-frequency diagram of the aliased signal shown.

[0096] S200: Perform independent component analysis on the time domain and frequency domain components to obtain time domain atoms and frequency domain atoms.

[0097] In this step, for example, such as Figure 4 As shown, the analysis considers both time and frequency dimensions separately. From the time-frequency domain x(t,f), FastICA transformations are performed on both the discrete time and discrete frequency dimensions to obtain a series of FastICA basis vectors T. base and F base Independent component analysis (ICA) is performed on the signals in both dimensions separately. The process of ICA for a single dimension is as follows: Assume there are p independent components in n channels:

[0098] X = AS

[0099] Where X is the received multi-channel signal, and S consists of p independent components. A is the mixing matrix. By estimating the mixing process, we can obtain:

[0100]

[0101] The estimation process used the FastICA algorithm, which compares the entropy between the signal and a pure Gaussian quantity. Since the Gaussian random variable is the most random of all distributions, the difference between the entropy of the Gaussian random variable and the estimated variable yields a non-negative quantity J, defined as negative entropy.

[0102] J(s)=H(s gauss )-H(s)

[0103] FastICA simplifies the ICA algorithm by using kurtosis. Since kurtosis is sensitive to edge values ​​and has low robustness, it uses higher-order cumulants in conjunction with density polynomial expansion, and approximates the negative entropy using a general form of non-quadratic expectation:

[0104]

[0105] Where v is a Gaussian random variable, and w and x are elements in the separation matrix and the original signal matrix, respectively.

[0106] A reasonable choice of non-quadratic function G is needed. To ensure robustness, we choose the following:

[0107] G(*)=(*) 4 / 4.

[0108] After performing fastICA analysis, m and n independent components were obtained, respectively. Among them, T... i For time-domain components, F i These are frequency domain components. Among the m and n independent components, a subset is extracted as reconstructed atoms. Where T... base For time-domain atoms, F base These are frequency domain atoms. The extracted indices are as follows:

[0109]

[0110] Correlation analysis was performed on the time-frequency domain atoms and the original signal domain time-frequency signals, and the results were arranged according to correlation. Atoms located in the middle of the sequence were selected as reconstruction atoms. The number of atoms is determined by the number of modes to be separated; for the separation of two modes A0 and S0, six time-domain atoms and six frequency-domain atoms are sufficient. An example is shown below. Figure 5 The correlation arrangement of the selected 6 time-domain atoms and 6 frequency-domain atoms. Example: Figure 6 The six reconstructed time-domain atoms and six frequency-domain atoms were selected.

[0111] Careful selection of the time and frequency bases is necessary. We need to choose only the base containing the most important and useful information from the many independent components generated by FastICA. ICA algorithms typically generate hundreds of independent components, and computing all of them would require significant computational power and storage. Therefore, only the most valuable components are selected for analysis. However, this approach can lead to distortion of the recovered signal. Therefore, after reconstruction, clustering algorithms are needed to compensate for the mode mixing caused by this selection.

[0112] S300: Reconstruct the time-frequency domain atoms to obtain the time-frequency spatial reconstruction component (TFR) of the signal. Establish the distance index D of the time-frequency spatial reconstruction component. Euk By performing C-MEANS fuzzy clustering on the reconstructed time-frequency spatial components, the separation of time-frequency spatial sub-components is achieved.

[0113] In this step, the selected time-domain atom T base and frequency domain atom F base Reconstruction yields a series of time-frequency domain representations (TFR):

[0114]

[0115] Where, x rec(t,f) represents the reconstructed time-frequency signal domain; t represents time; i represents the index, and there are a total of x+y time-frequency domain representations, where x is T. base The number of y is F base The number of modes; TFR is represented in the time-frequency domain. The reconstructed signal is restored to the time-frequency domain. At this point, by comparing it with the original signal, the distinguishing index D of different modes can be obtained. Euk :

[0116] D Euk (i)=||TFR i -TFR origin || 2

[0117] In this step, TFR origin The time-frequency domain representation of the original signal, i.e., x(t,f), is obtained by obtaining the index D. Euk The reconstructed time-frequency domain representation (TFR) is separated along both time and frequency dimensions. The separation utilizes a fuzzy C-clustering algorithm, the process of which is as follows:

[0118]

[0119] In this step, V = [v1, v2, ..., v c ],v i ∈R n , 是类别向量 . It is the distance norm of the square inner product. U = [μ] ik ] is the fuzzy classification result matrix, where μ ik This represents the membership degree of the i-th data point to the k-th cluster. The clustering results are shown below. Figure 7 The overlapping parts of two modalities can be removed using fuzzy clustering.

[0120] S400: Perform inverse time-frequency analysis on the separated reconstructed components to obtain the separated signal.

[0121] In this step, the time-frequency representation signals of different modes are reconstructed to obtain time-domain signals.

[0122]

[0123] The parameters used in the inverse short-time Fourier transform are the same as those used in step S100.

[0124] like Figures 8(a) to 8(c) To make 1000×1000×2mm 3 Mode separation results of a non-overlapping signal propagating over a distance of 600mm and a center frequency of 500kHz on a T6061 aluminum alloy sheet. Figure 9The reconstructed time-frequency representation clustering result of a dual-mode aliasing signal with a propagation distance of A0 (600 mm), S0 (300 mm), and a center frequency of 500 kHz is given. Figures 10(a) to 10(c) The mode separation results are for a dual-mode aliasing signal with a propagation distance of A0 (600mm), S0 (300mm), and a center frequency of 500kHz.

[0125] from Figures 8(a) to 8(c) , Figures 10(a) to 10(c) The results show that this method has a good separation effect on blind mode separation. Further separation of the experimental results was performed, and imaging was conducted using the A0 mode. Figure 11 It is 1000×1000×2mm 3 An experimental plate with a propagation distance of 600mm was used on a T6061 aluminum alloy sheet. The three shortest paths from the transmitting PZT to the receiving PZT were marked as D1, D2, and D3. After boundary reflection, the received signal had three wave packets, as shown in Figure 12(b). After processing and separation in this embodiment, the separated S0 mode can be seen in Figure 12(c), and the separated A0 mode can be seen in Figure 12(d). This disclosure has a good separation effect.

[0126] Existing modal separation and compensation techniques often use known dispersion prior curves to compensate for modes that dominate energy at specific frequencies. This is computationally intensive and requires prior acquisition of the dispersion curves of experimental samples, necessitating additional testing and experiments. Furthermore, the selection of frequency-dominant modes cannot completely eliminate interference from non-dominant modes. In contrast, this disclosure utilizes a reconstruction algorithm that extracts independent components from the video space. Because it extracts only a certain number of major independent components, it has the advantage of low computational cost. Moreover, it can perform blind separation estimation without prior knowledge and exhibits good adaptability to Lamb waves. Therefore, the technical solution of this disclosure can separate multimodal signals by mode, obtaining multiple single-mode ultrasonic guided wave component sub-signals.

[0127] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this application to the necessity of employing the aforementioned specific details.

Claims

1. A Lamb wave mode blind separation method based on time-frequency space independent components, characterized in that, It comprises the following steps: Step S100: the object to be tested is subjected to Lamb wave nondestructive testing, and a piezoelectric sensor using single channel transceiving is used to collect signals, wherein the collected signals include multi-channel multi-modal signals having Lamb wave original signals; time-frequency analysis is performed on the collected Lamb wave original signals to obtain two-dimensional time-frequency space components; Step S200: based on the two-dimensional time-frequency space components, respectively, time domain component and frequency domain component are analyzed by independent component analysis, time domain atom and frequency domain atom are obtained, time domain atom, frequency domain atom are sorted by correlation index , select other part atoms of the sorting removing maximum value and minimum value as time domain reconstruction atom and frequency domain reconstruction atom , time domain reconstruction atom , frequency domain reconstruction atom Step S300: reconstructing atoms based on time and frequency domain obtaining time-frequency space reconstruction components of the signal establishing a time-frequency space reconstruction component distance index separating time-frequency space sub-components by using the time-frequency space reconstruction component distance index performing C-MEANS fuzzy clustering on the time-frequency space reconstruction components to separate time-frequency space sub-components Step S400: inverse time-frequency analysis is performed on the time-frequency space sub-components to obtain separated signals.

2. The method of claim 1, wherein, The time-frequency analysis uses short-time Fourier transform: ， wherein is a window function, is a short-time Fourier time, is a Lamb wave original time domain signal, is a Fourier transform base, wherein is a virtual part mark, is a short-time Fourier time, is a constant of pi, and the Lamb wave original signal becomes a two-dimensional signal in a time-frequency domain after a short-time Fourier transform, denotes a time domain, denotes a frequency domain.

3. The method of claim 2, wherein, The window function is a square window.

4. The method of claim 1, wherein, The excitation signal of the piezoelectric sensor is a toneburst signal.

5. The method of claim 1, wherein, In step S200, the single dimension independent component analysis includes, independent components in the channel​ ， wherein, is a received multi-channel multi-modal signal, is a mixing matrix, obtained by estimating the process of mixing: is composed of individual components, ， wherein is an independent component estimate of the multichannel multimodal signal, is a separation matrix, the FastICA algorithm is used for independent component estimation, and by comparing the entropy between the signal and the pure Gaussian quantity, since the Gaussian random variable is the most random in all distributions, the difference between the entropy of the Gaussian random variable and the estimated variable is obtained , defined as negative entropy, ， The FastICA algorithm is simplified using kurtosis, high-order cumulants are used in combination with density polynomial expansion, and the negative entropy is approximated using non-quadratic expectation: ， wherein is a Gaussian random variable, and are elements in the separation matrix and the original signal matrix, .

6. The method of claim 5, wherein, selecting the transformed original time-frequency domain signal performing fastICA analysis on the signal in time domain and frequency domain respectively, and obtaining independent components respectively, wherein, is a time domain component, is a frequency domain component, and in the independent components, part of the components are extracted as reconstruction atoms, and the indexes of the extracted components are as follows: ， Correlation analysis is performed on the time-frequency signals in the time-frequency domain atom and the original signal domain respectively, and the atoms in the middle part are selected as the corresponding time-domain reconstruction atoms and frequency-domain reconstruction atoms according to the correlation.

7. The method of claim 6, wherein, The 6 time-domain atoms and 6 frequency-domain atoms in the middle part are selected as the corresponding time-domain reconstruction atoms and frequency-domain reconstruction atoms.

8. The method of claim 6, wherein, based on the selected time-domain reconstruction atoms and frequency-domain reconstruction atoms resulting in time-frequency space reconstruction components : ， wherein represents a reconstructed time-frequency signal domain; represents time; i represents a serial number, and there are x+y time-frequency domain representations in total, x is the number of, and y is the number of.

9. The method of claim 8, wherein, Fuzzy clustering is applied to separate different TFRs, where modeA0represents the TFR of the A0mode; modeS0represents the TFR of the S0mode; the process of fuzzy clustering is realized by a fuzzy C-means algorithm, and the fuzzy C-means algorithm is: ， wherein, is a category vector, is a squared inner product distance norm, is a fuzzy classification result matrix, wherein denotes the membership of the i-th data point to the k-th cluster.

10. The method of claim 9, wherein, Inverse time-frequency analysis is performed on the separated time-frequency space reconstruction components TFR to obtain separated signals: ， Wherein, ， wherein is all time-frequency domain signals, is a short-time Fourier time, is a frequency, is a Fourier base, wherein is a imaginary part marker, is a frequency, is a constant of pi, is a recovered Lamb wave original time domain signal, denotes a time domain, a Lamb wave reconstructed two-dimensional signal becomes a one-dimensional recovered signal in a time domain after an inverse short-time Fourier transform.

Images