Acoustic emission sensor and acoustic emission energy signal separation method

Abstract

The invention belongs to the technical field of acoustic emission sensors, and particularly relates to an acoustic emission sensor and an acoustic emission energy signal separation method, a time domain signal data set is obtained by emitting a plurality of single-frequency excitation signal receiving echoes, then time-frequency analysis, feature extraction, principal component analysis and clustering analysis are performed, and dispersion model verification is input. And separating the signal into a reflection signal, a transmission signal and a noise signal, and finally reconstructing the transmission signal and evaluating the separation signal to obtain a high-quality separation transmission signal. By analyzing the diffusion characteristics of sound waves with different frequencies in the material, the reflection signal, the transmission signal and the environmental noise are effectively distinguished, so that the energy signal actually penetrating through the object is accurately identified. High-quality separation transmission signals are used for air coupling ultrasonic detection, interference of reflection signals and noise signals is effectively reduced, and the detection precision is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of acoustic emission sensors, and in particular relates to an acoustic emission sensor and an acoustic emission energy signal separation method. Background Art

[0002] Acoustic emission sensors are widely used in the field of non-destructive testing of materials. This technology mainly uses the characteristics of sound wave propagation to evaluate the structural integrity, defect distribution and mechanical properties of materials by analyzing the characteristics of sound wave signals in materials. Acoustic emission sensors evaluate internal damage by monitoring the transient elastic waves generated by materials under force, temperature or other external stimuli. Typical acoustic emission signals are closely related to the destructive behaviors such as microcracks, fiber debonding, and pore expansion inside the material.

[0003] Acoustic emission sensors based on air-coupled ultrasonic technology have the advantages of high sensitivity and high resolution, and realize effective non-contact and non-destructive detection. However, in practical applications, the echo signal of the acoustic emission sensor contains a mixture of reflected and transmitted signals, and the complex structure or multi-layer materials cause signal interference, as well as the influence of environmental noise, which reduces the accuracy and reliability of the test results and limits the application scope of non-destructive testing technology. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the defect in the prior art that the air-coupled ultrasonic technology is affected by the reflected signal and noise interference and affects the signal quality, thereby providing an acoustic emission sensor and an acoustic emission energy signal separation method.

[0005] A method for separating acoustic emission energy signals comprises the following steps:

[0006] Step S1: generating a plurality of single-frequency excitation signals, wherein the frequencies of the plurality of single-frequency excitation signals are logarithmically spaced;

[0007] Step S2: transmitting a plurality of the single-frequency excitation signals through a transducer, receiving echoes to obtain a time domain signal data set;

[0008] Step S3: performing time-frequency analysis on the time domain signal data set to obtain a time-frequency spectrum set;

[0009] Step S4: extracting features from the time-frequency spectrum set to obtain a feature matrix;

[0010] Step S5: performing principal component analysis on the feature matrix to obtain a dimension-reduced feature matrix;

[0011] Step S6: performing cluster analysis on the dimension reduction feature matrix to obtain a clustering result, wherein the clustering result includes a category label of each data point, and the category label includes reflection, transmission and noise;

[0012] Step S7: constructing a dispersion model according to the clustering results and the physical parameters of the material, incorporating the dispersion model into the dispersion model, and obtaining a verified clustering result;

[0013] Step S8: inputting the verification clustering result into the dispersion model, and based on the cluster center feature identification signal of the verification clustering result, marking the signal in the verification clustering result as a reflection signal, a transmission signal and a noise signal;

[0014] Step S9: superimposing and processing the transmission signals to obtain a reconstructed transmission signal;

[0015] Step S10: calculating a separation effect index based on the single-frequency excitation signal and the reconstructed transmission signal.

[0016] Furthermore, in step S1, the single-frequency excitation signal is a sinusoidal wave signal.

[0017] Furthermore, in step S3, the time-frequency analysis is continuous wavelet transform:

[0018] W(a,b)=(1 / sqrt(|a|))∫x(t)ψ*((tb) / a)dt;

[0019] Among them, ψ(t) represents the mother wavelet, a represents the scale parameter, and b represents the translation parameter.

[0020] Furthermore, in step S3, the time-frequency analysis is discrete wavelet transform:

[0021] W[m,n]=(1 / sqrt(a0 m ))∑x[k]ψ*((k-nb0 a0^m) / a0 m );

[0022] Among them, m and n are integers, a0>1 represents a fixed expansion step, and b0>0 represents a position parameter.

[0023] Furthermore, in step S4, each row of the feature matrix represents a signal of a frequency in the time-frequency spectrum set, and each column represents a feature; the features include peak amplitude, root mean square value, waveform factor, kurtosis, spectral entropy, energy density ratio, center frequency, frequency diffusion width, diffusion coefficient, and group velocity dispersion.

[0024] Furthermore, the step S5 includes the following steps:

[0025] Step S5.1: Obtain the feature matrix X, the size of which is m×n, where m is the number of samples and n is the number of features;

[0026] Step S5.2: Calculate the central feature matrix of the feature matrix:

[0027] X centered =X–μ;

[0028] Wherein, μ is the mean vector of the feature matrix;

[0029] Step S5.3: Calculate the covariance matrix C of the centered feature matrix:

[0030] C=(1 / m)*X centered T *X centered ;

[0031] Step S5.4: Perform eigenvalue decomposition on the covariance matrix:

[0032] C = VΛV^T;

[0033] Where V represents the eigenvector matrix, and Λ represents the diagonal matrix of eigenvalues;

[0034] Step S5.5: Arrange the eigenvectors of the eigenvector matrix in descending order, and select the first k eigenvectors to form a projection matrix W; wherein k makes the cumulative explained variance ratio reach a preset threshold:

[0035] explained_variance_ratio = λ i / ∑λ i ;

[0036] where λ i represents the eigenvalue;

[0037] Step S5.6: Project the centered feature matrix to the new subspace to obtain the reduced dimension feature matrix:

[0038] Y=X centered *W.

[0039] Furthermore, in step S6, the cluster analysis includes the following steps:

[0040] Step S6.1: define the neighborhood radius and the minimum number of points in the dense area;

[0041] Step S6.2: For each point in the reduced dimension feature matrix, if the number of points in its neighborhood is greater than or equal to the minimum number of points in the dense area, it is defined as a core point;

[0042] Step S6.3: For a core point, add all points in its neighborhood to a cluster, and add all points in the neighborhood of the cluster to the cluster, and repeat until no new points are added to the cluster;

[0043] Step S6.4: define points that do not belong to any cluster as noise points, and output a clustering result, which includes all the clusters and the noise points.

[0044] Furthermore, the step S7 includes the following steps:

[0045] Step S7.1: According to the clustering result, the transmission signal dispersion analysis calculates the frequency-diffusion relationship, the reflection signal dispersion analysis calculates the frequency dispersion, and calculates the comprehensive dispersion function;

[0046] Step S7.1.1: Extract frequency diffusion features:

[0047] σf=sqrt(Σ(f-fc) 2 |X(f)| 2 / Σ|X(f)| 2 );

[0048] Determine the frequency dependence of α(ω) and v(ω)

[0049] Establishing the frequency-dispersion relationship:

[0050] D trans (ω)=f(σf,α(ω),v(ω));

[0051] Step S7.1.2: Extract frequency dispersion features and calculate v(ω) and R(ω)

[0052] Analyze dispersion patterns and study frequency dependence

[0053] Establish the frequency-dispersion relationship:

[0054] D ref (ω) = g(v(ω), R(ω));

[0055] Step S7.1.3: Calculate the integrated dispersion function:

[0056] D(ω)=Σ(wi*Di(ω)) / Σwi;

[0057] Where D(ω) represents the comprehensive dispersion function, Di(ω) represents the dispersion function of each group, and wi represents the weight based on signal quality and physical rationality;

[0058] Step S7.2: Calculate the diffusion function of each cluster group, check consistency and update the dispersion classification, and calculate the diffusion difference of each cluster group:

[0059] Step S7.2.1: extract frequency diffusion features for each cluster group and calculate σf and Dc;

[0060] Analyze the diffusion law and study the changes of α(ω) and v(ω);

[0061] Compute the diffusion function:

[0062] D(ω)=(f max -f min ) / f center ;

[0063] Step S7.2.2: Calculate the dispersion characteristic variance:

[0064] Var(D)=Σ(Di-D mean ) 2 / n;

[0065] Identify abnormal diffusion points:

[0066] |Di-D mean |>3σ;

[0067] Remove outliers, reclassify, and update dispersion classification;

[0068] Step S7.2.3: Analyze the dispersion differences of different categories: calculate the dispersion feature differences;

[0069] Verify the physical rationality of dispersion: Check whether it meets the physical laws of the material;

[0070] Calculate the diffusion difference:

[0071] ΔD=|D1(ω)-D2(ω)| / D ref ;

[0072] Step S7.3: According to the updated clustering model, based on the theoretical dispersion model verification, the dispersion classification is updated:

[0073] Step S7.3.1: Calculate the frequency spread width:

[0074] σf=sqrt(Σ(f-fc) 2 |X(f)| 2 / Σ|X(f)| 2 );

[0075] Calculate the diffusion characteristic deviation:

[0076] δD=|Dm-Dc| / Dc;

[0077] Where Dm represents the measured diffusion function, and Dc represents the theoretical diffusion function;

[0078] Calculate the frequency dispersion:

[0079] ΔV=|V(f2)-V(f1)| / V(f1);

[0080] Step S7.3.2: For each cluster group, if (σf<0.2)&&(δD<0.15)&&(ΔV<0.1) is satisfied, then the cluster group passes; if not, then adjust the model parameters;

[0081] Step S7.3.3: Remove points that do not conform to the dispersion law:

[0082] |DD mean |>2σ;

[0083] Re-divide the boundary based on ΔD, calculate the dispersion feature center and update D mea n and σ.

[0084] Furthermore, the step S9 includes the following steps:

[0085] Step S9.1: constructing a signal model based on the transmission signal:

[0086] y=Φx+n;

[0087] Where y represents the measured signal, Φ represents the measurement matrix, x represents the original signal, and n represents the noise;

[0088] Step S9.2: Compressed sensing problem:

[0089] min||x||1subject to||y-Φx||2≤ε;

[0090] Among them, ||·||1 represents the L1 norm, ||·||2 represents the L2 norm, and ε represents the noise level;

[0091] Step S9.3: Implement the basis pursuit denoising algorithm:

[0092] min(1 / 2)||y-Φx||2 2 +λ||x||1;

[0093] Where λ represents the regularization parameter;

[0094] Step S9.4: Implement the orthogonal matching pursuit algorithm:

[0095] r0=y, x0=0;

[0096] For k = 1 to K:

[0097]

[0098] ii.Λ k = Λ_{k-1}∪{λ k};

[0099] iii.x k=arg min||y-Φ_Λ k x||2iv.r k =y-Φx k Return x k ;

[0100] Among them, φ j The column representing Φ, Φ_Λ k represents a submatrix of Φ, whose columns are composed of Λ k index.

[0101] Furthermore, the separation effect indicators in step S10 include: signal-to-noise ratio, mean square error, peak signal-to-noise ratio, structural similarity index, correlation coefficient, spectrum consistency, and energy recovery ratio.

[0102] An acoustic emission sensor realizes signal separation through the above-mentioned acoustic emission energy signal separation method.

[0103] Beneficial effects: The present invention discloses a method for separating acoustic emission energy signals, which obtains a time domain signal data set by transmitting multiple single-frequency excitation signals and receiving echoes, and then performs time-frequency analysis, feature extraction, principal component analysis, cluster analysis and inputs the dispersion model for verification, to separate the signal into a reflection signal, a transmission signal and a noise signal, and finally reconstructs the transmission signal and evaluates the separated signal to obtain a high-quality separated transmission signal. The present invention effectively distinguishes between reflection signals, transmission signals and environmental noise by analyzing the diffusion characteristics of sound waves of different frequencies in the material, thereby accurately identifying the energy signal that actually passes through the object. The high-quality separated transmission signal is used for air-coupled ultrasonic detection, which effectively reduces the interference of reflection signals and noise signals and improves the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0105] Figure 1 It is a schematic block diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0106] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present application, so the present application is not limited by the specific embodiments disclosed below.

[0107] In the description of this application, it should be understood that, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0108] Embodiment 1:

[0109] Reference Figure 1 As shown, this embodiment provides an acoustic emission energy signal separation method, comprising the following steps:

[0110] Step S1: generating a plurality of single-frequency excitation signals, wherein the frequencies of the plurality of single-frequency excitation signals are logarithmically spaced;

[0111] Step S2: transmitting a plurality of the single-frequency excitation signals through a transducer, receiving echoes to obtain a time domain signal data set;

[0112] Step S3: performing time-frequency analysis on the time domain signal data set to obtain a time-frequency spectrum set;

[0113] Step S4: extracting features from the time-frequency spectrum set to obtain a feature matrix;

[0114] Step S5: performing principal component analysis on the feature matrix to obtain a dimension-reduced feature matrix;

[0115] Step S6: performing cluster analysis on the dimension reduction feature matrix to obtain a clustering result, wherein the clustering result includes a category label of each data point, and the category label includes reflection, transmission and noise;

[0116] Step S7: constructing a dispersion model according to the clustering results and the physical parameters of the material, incorporating the dispersion model into the dispersion model, and obtaining a verified clustering result;

[0117] Step S8: inputting the verification clustering result into the dispersion model, and based on the cluster center feature identification signal of the verification clustering result, marking the signal in the verification clustering result as a reflection signal, a transmission signal and a noise signal;

[0118] Step S9: superimposing and processing the transmission signals to obtain a reconstructed transmission signal;

[0119] Step S10: calculating a separation effect index based on the single-frequency excitation signal and the reconstructed transmission signal.

[0120] Specifically, in step S1, the single-frequency excitation signal is a sine wave signal. In this embodiment, it is expressed as:

[0121] f i =f start *(f end / f start ) (i / (N-1))

[0122] Among them, f i represents the i-th frequency, f start and f end Indicates the starting and ending frequencies, indicates the total number of frequency points, i ranges from 0 to N-1.

[0123] As a preferred embodiment of this embodiment, the starting frequency f start is 100kHz, the end frequency f end The logarithmic spacing allows fewer frequency points to cover a wider frequency range, ensuring that both low and high frequencies are fully represented.

[0124] In this embodiment, the time-frequency analysis of step S3 is wavelet transform, which can provide good time resolution for high-frequency components and good frequency resolution for low-frequency components. At the same time, wavelet transform can adapt to signal characteristics and capture transient characteristics more effectively.

[0125] In some implementations of this embodiment, in step S3, the time-frequency analysis is continuous wavelet transform:

[0126] W(a,b)=(1 / sqrt(|a|))∫x(t)ψ*((tb) / a)dt;

[0127] Among them, ψ(t) represents the mother wavelet, a represents the scale parameter, and b represents the translation parameter.

[0128] In some other implementations of this embodiment, in step S3, the time-frequency analysis is discrete wavelet transform:

[0129] W[m,n]=(1 / sqrt(a0m ))∑x[k]ψ*((k-nb0 a0^m) / a0 m );

[0130] Among them, m and n are integers, a0>1 represents a fixed expansion step, and b0>0 represents a position parameter.

[0131] In step S4, each row of the feature matrix represents a signal of a frequency in the time-frequency spectrum set, and each column represents a feature; the features include peak amplitude, root mean square value, waveform factor, kurtosis, spectral entropy, energy density ratio, center frequency, frequency diffusion width, diffusion coefficient, and group velocity dispersion.

[0132] Specifically, the following steps are included:

[0133] Step S4.1: Calculate the power spectral density (PSD) of the signal;

[0134] Step S4.2: normalizing the power spectrum density;

[0135] Step S4.3: Calculate spectral entropy based on the normalized power spectral density;

[0136] Step S4.4: dividing the signal into a high frequency band and a low frequency band based on the frequency range;

[0137] Step S4.5: Calculate the signal energy in the high frequency band and the low frequency band;

[0138] Step S4.6: Calculate the energy density ratio of the high frequency band and the low frequency band according to the signal energy.

[0139] Spectral entropy is used to reflect the complexity or uniformity of frequency distribution:

[0140] H=-∑(p i *log2(p i ));

[0141] Among them, p i represents the normalized power spectral density at frequency bin i.

[0142] The energy density ratio is used to compare the energy in different frequency bands:

[0143] EDR=E high / E low ;

[0144] Where E high Represents the energy in the high frequency band, E low Represents the energy in the low frequency band.

[0145] Energy is expressed as: E = ∑|X(f)|^2;

[0146] Where X(f) represents the Fourier transform of the signal.

[0147] The step S5 comprises the following steps:

[0148] Step S5.1: Obtain the feature matrix X, the size of which is m×n, where m is the number of samples and n is the number of features;

[0149] Step S5.2: Calculate the central feature matrix of the feature matrix:

[0150] X centered =X–μ;

[0151] Wherein, μ is the mean vector of the feature matrix;

[0152] Step S5.3: Calculate the covariance matrix of the centered feature matrix:

[0153] C=(1 / m)*X centered T *X centered ;

[0154] Step S5.4: Perform eigenvalue decomposition on the covariance matrix:

[0155] C = VΛV^T;

[0156] Where V represents the eigenvector matrix, and Λ represents the diagonal matrix of eigenvalues;

[0157] Step S5.5: Arrange the eigenvectors of the eigenvector matrix in descending order, and select the first k eigenvectors to form a projection matrix W; wherein k makes the cumulative explained variance ratio reach a preset threshold:

[0158] explained_variance_ratio = λ i / ∑λ i ;

[0159] where λ i represents the eigenvalue;

[0160] Step S5.6: Project the centered feature matrix to the new subspace to obtain the reduced dimension feature matrix:

[0161] Y=X centered *W.

[0162] In step S6, a K-means algorithm is used to perform cluster analysis. In this embodiment, the algorithm K=3, representing the reflection signal, the transmission signal and the noise signal respectively. The cluster analysis includes the following steps:

[0163] Step S6.1: define the neighborhood radius and the minimum number of points in the dense area;

[0164] Step S6.2: For each point in the reduced dimension feature matrix, if the number of points in its neighborhood is greater than or equal to the minimum number of points in the dense area, it is defined as a core point;

[0165] Step S6.3: For a core point, add all points in its neighborhood to a cluster, and add all points in the neighborhood of the cluster to the cluster, and repeat until no new points are added to the cluster;

[0166] Step S6.4: define points that do not belong to any cluster as noise points, and output a clustering result, which includes all the clusters and the noise points.

[0167] In some embodiments of the present invention, the dimension reduction feature matrix is ​​expressed as:

[0168] F = [F1, F2, F3, F4, F5];

[0169] Among them, F1 represents the frequency characteristic, F2 represents the energy characteristic, F3 represents the propagation characteristic, F4 represents the dispersion attenuation characteristic, and F5 represents the time characteristic, which are respectively expressed as:

[0170] F1=[σf,Bw,fc];

[0171] Where σf represents the frequency diffusion width: σf ​​= sqrt(∑(f-fc) 2 |X(f)| 2 / ∑|X(f)| 2 ); Bw represents the bandwidth, which is -3dB in this embodiment, thereby determining the frequency range in which the amplitude of the spectrum drops by 3dB: Bw = f upper -f lower , fc represents the center frequency: fc=Σ(f·|X(f)| 2 ) / Σ|X(f)| 2 ; Wherein, X(f) represents the signal spectrum, which is calculated by fast Fourier transform: X(f) = FFT(x(t));

[0172] F2=[E ratio ,SNR,A norm ];

[0173] Among them, [E ratio represents the energy ratio, [E ratio =E high / E low , E high =∫|X(f)| 2 df and f∈[f c ,fmax ], E low =∫|X(f)| 2 df and f∈[f min ,f c ] In this embodiment, since the actual distance does not need to be considered, a ratio is used to eliminate the influence of the distance, and the SNR is calculated by segmented integration after FFT; SNR = 10log 10 (P signal / P noise ) is called signal-to-noise ratio, where P signal =Σ|X s (f)| 2 Represents the signal power, P noise =Σ|X n (f)| 2 Represents the noise power X s and X n Respectively represent the signal / noise spectrum; A norm represents the normalized amplitude, A max = max{|x(t)|}, t∈[0,T], A norm =|x(t)| / A max .

[0174]

[0175] Among them, vg norm represents the normalized group velocity, vg = dω / dβ(ω), vg norm =vg / c0, c0 represents the longitudinal wave speed / transverse wave speed / other characteristic speeds; Where φ(f)=arg(X(f)), which is the phase angle of X(f), f1 and f2 represent the common characteristic frequency points, which are the main frequency ±Δf in this embodiment; t rel =t / t first Represents relative time characteristics, t is assumed to be the current time point, t first is the time when the signal exceeds the threshold for the first time.

[0176] F4=[D,R freq ,η];

[0177] Where D = -ω / v·dv / dω represents the dispersion coefficient, v represents the acoustic velocity, ω represents the angular frequency, and the phase velocity v(ω) = [v(ω) = ω / β(ω), R freq =|X(f1)| / |X(f2)| represents the frequency ratio. In this embodiment, a frequency ratio independent of the path is selected. f1 and f2 represent characteristic frequencies, i.e., the main frequency ±Δf; η = -ln(A(ω) / A0(ω)) / η refrepresents the normalized attenuation index, A(ω) and A0(ω) represent the current frequency amplitude and the reference amplitude respectively, and η_ref represents the normalized reference value;

[0178] F5=[t rel ,t ratio ,t spread ];

[0179] Among them, t rel =t / t first represents the normalized value relative to the first wave arrival time; t ratio =t peak / t first , represents the ratio of peak time to first wave time; t spread =(t end -t start ) / t first Indicates the time width.

[0180] Thus, in this embodiment, the cluster marked as a reflection signal has the following characteristics:

[0181] 0.8 <R freq <1.2;D <D th (D th ≈0.2);η(ω)<η th1 (η th1 ≈0.3); E ratio >0.7;t rel ≈1; t ratio <1.2; t spread <spread th1 ;

[0182] Clusters marked as transmission signals have the following characteristics:

[0183] 0.3 <R freq <0.8; D>D th ; η th1 <η(ω)<η th2 (η th2 ≈0.7); 0.3 <E ratio <0.7; t rel >1.2;t ratio >1.5; spread th1 <t spread <spread th2 ;

[0184] The clusters labeled as noise signals have the following characteristics:

[0185] R freq Random distribution; η(ω)>η th2 ;SNR <SNRth (SNR th ≈10dB); E ratio <0.3; t_rel randomly distributed; t ratio Unstable; spread >spread th2 .

[0186] In step S7, an accurate dispersion model is constructed by using the Kramers-Kronig relationship, which specifically includes the following steps:

[0187] Define the complex wave number: k(ω)=β(ω)+iα(ω);

[0188] Among them, β represents the phase constant and α represents the attenuation coefficient;

[0189] The Kramers-Kronig relation for wave number is expressed as:

[0190] β(ω)=β0+(2 / π)P∫[0,∞](xα(x) / (x 2 -ω 2 ))dx;

[0191] α(ω)=-(2ω / π)P∫[0,∞](β(x)-β0 / (x 2 -ω 2 ))dx;

[0192] Where P represents the Cauchy principal value integral and β0 represents the low-frequency limit of β.

[0193] In this embodiment, β(ω) and α(ω) are expressed by approximate realization:

[0194] β(ω)≈β0+(2 / π)∑[i](Δω i α i ω i / (ω i 2 -ω 2 ));

[0195] α(ω)≈-(2ω / π)∑[i](Δω i (β i -β0) / (ω i 2 -ω 2 ));

[0196] Among them, Δω i represents the frequency step, α i and β i Respectively represent the frequency ω i The measured value at .

[0197] Calculate the phase velocity:

[0198] v p (ω) = ω / β(ω);

[0199] Calculate the group velocity:

[0200] v g (ω) = dω / dβ(ω);

[0201] Step S7.1: According to the clustering result, the transmission signal dispersion analysis calculates the frequency-diffusion relationship, the reflection signal dispersion analysis calculates the frequency dispersion, and calculates the comprehensive dispersion function;

[0202] Step S7.1.1: Extract frequency diffusion features, where f represents the signal frequency, fc represents the center frequency, X(f) represents the signal spectrum, |X(f)| 2 Represents the power spectral density, and the frequency diffusion width σf describes the degree of signal broadening in the frequency domain:

[0203] σf=sqrt(Σ(f-fc) 2 |X(f)| 2 / Σ|X(f)| 2 );

[0204] Determine the frequency dependence of α(ω) and v(ω)

[0205] Establishing the frequency-diffusion relationship, D trans (ω) describes the frequency broadening characteristics of the transmission signal, quantifies the discreteness of the signal frequency distribution, and reflects the frequency change law of sound waves when propagating in different media:

[0206] D trans (ω)=f(σf,α(ω),v(ω));

[0207] Step S7.1.2: Extract frequency dispersion features and calculate v(ω) and R(ω)

[0208] Analyze dispersion patterns and study frequency dependence

[0209] Establishing the frequency-dispersion relationship, D ref (ω) characterizes the dispersion characteristics of the reflected signal:

[0210] D ref (ω) = g(v(ω), R(ω));

[0211] Step S7.1.3: Calculate the integrated dispersion function:

[0212] D(ω)=Σ(wi*Di(ω)) / Σwi;

[0213] Where D(ω) represents the comprehensive dispersion function, Di(ω) represents the dispersion function of each group, and wi represents the weight based on signal quality and physical rationality;

[0214] Step S7.2: Calculate the diffusion function of each cluster group, check consistency and update the dispersion classification, and calculate the diffusion difference of each cluster group:

[0215] Step S7.2.1: extract frequency diffusion features for each cluster group and calculate σf and Dc;

[0216] Analyze the diffusion law and study the changes of α(ω) and v(ω);

[0217] Compute the diffusion function:

[0218] D(ω)=(f max -f min ) / f center ;

[0219] Step S7.2.2: Calculate the dispersion characteristic variance:

[0220] Var(D)=Σ(Di-D mean ) 2 / n;

[0221] Identify abnormal diffusion points:

[0222] |Di-D mean |>3σ;

[0223] Remove outliers, reclassify, and update dispersion classification;

[0224] Step S7.2.3: Analyze the dispersion differences of different categories: calculate the dispersion feature differences;

[0225] Verify the physical rationality of dispersion: Check whether it meets the physical laws of the material;

[0226] Calculate the diffusion difference:

[0227] ΔD=|D1(ω)-D2(ω)| / D ref ;

[0228] Step S7.3: According to the updated clustering model, based on the theoretical dispersion model verification, the dispersion classification is updated:

[0229] Step S7.3.1: Calculate the frequency spread width:

[0230] σf=sqrt(Σ(f-fc) 2 |X(f)| 2 / Σ|X(f)| 2 );

[0231] Calculate the diffusion characteristic deviation:

[0232] δD=|Dm-Dc| / Dc;

[0233] Where Dm represents the measured diffusion function, and Dc represents the theoretical diffusion function;

[0234] Calculate the frequency dispersion:

[0235] ΔV=|V(f2)-V(f1)| / V(f1);

[0236] Step S7.3.2: For each cluster group, if (σf<0.2)&&(δD<0.15)&&(ΔV<0.1) is satisfied, then the cluster group passes; if not, then adjust the model parameters;

[0237] Step S7.3.3: Remove points that do not conform to the dispersion law:

[0238] |DD mean |>2σ;

[0239] Re-divide the boundary based on ΔD, calculate the dispersion feature center and update D mean and σ.

[0240] The step S9 includes the following steps:

[0241] Step S9.1: constructing a signal model based on the transmission signal:

[0242] y=Φx+n;

[0243] Where y represents the measured signal, Φ represents the measurement matrix, x represents the original signal, and n represents the noise;

[0244] Step S9.2: Compressed sensing problem:

[0245] min||x||1subject to||y-Φx||2≤ε;

[0246] Among them, ||·||1 represents the L1 norm, ||·||2 represents the L2 norm, and ε represents the noise level;

[0247] Step S9.3: Implement the basis pursuit denoising algorithm:

[0248] min(1 / 2)||y-Φx||2 2 +λ||x||1;

[0249] Where λ represents the regularization parameter;

[0250] Step S9.4: Implement the orthogonal matching pursuit algorithm:

[0251] r0=y, x0=0;

[0252] For k = 1 to K:

[0253]

[0254] ii.Λ k = Λ_{k-1}∪{λ k};

[0255] iii.x k =arg min||y-Φ_Λ k x||2iv.r k =y-Φx k Return x k ;

[0256] Among them, φ j represents the columns of Φ, represents a submatrix of Φ, whose columns are composed of Λ k Index. By applying compressed sensing techniques, the complete transmission signal can be reconstructed from a reduced number of measurements, potentially speeding up the acquisition process and reducing data storage requirements.

[0257] The separation effect indicators in step S10 include: signal-to-noise ratio, mean square error, peak signal-to-noise ratio, structural similarity index, correlation coefficient, spectrum consistency, and energy recovery ratio. The specific calculation method is as follows.

[0258] Signal-to-Noise Ratio (SNR):

[0259] SNR = 10*log10(P 信号 / P 噪声 )

[0260] Wherein, P_signal represents the power of the signal, and P_noise represents the power of the noise.

[0261] P 信号 =∑|s[n]| 2 ;

[0262] P 噪声 =∑|x[n]-s[n]| 2 ;

[0263] Among them, s[n] represents the original signal and x[n] represents the reconstructed signal.

[0264] Mean square error (MSE): MSE = (1 / N) * ∑ (x[n] - s[n]) 2 Where N is the signal length.

[0265] Peak Signal-to-Noise Ratio (PSNR):

[0266] PSNR = 10 * log10 (MAX 2 / MSE);

[0267] Here, MAX represents the maximum possible value of the signal.

[0268] Structural Similarity Index (SSIM):

[0269] SSIM=(2μxμs+c1)(2σxs+c2) / ((μx 2 +μs 2 +c1)(σx 2 +σs 2 +c2);

[0270] Among them, μx and μs represent the means of x and s respectively, σx and σs represent the standard deviations, σxs represents the covariance, and c1 and c2 are constants.

[0271] Correlation coefficient:

[0272] R=∑((x[n]-μx)(s[n]-μs)) / (σx*σs);

[0273] Spectrum consistency:

[0274] SC=∑min(|X(f)|,|S(f)|) / max(∑|X(f)|,∑|S(f)|);

[0275] where X(f) and S(f) represent the spectra of x[n] and s[n] respectively.

[0276] Energy recovery ratio:

[0277] RR=∑|x[n]| 2 / ∑|s[n]| 2 .

[0278] Embodiment 2:

[0279] This embodiment provides an acoustic emission sensor, which realizes signal separation through the acoustic emission energy signal separation method described in the first embodiment.

[0280] This embodiment provides a method for separating acoustic emission energy signals, which obtains a time domain signal data set by transmitting multiple single-frequency excitation signals and receiving echoes, and then performs time-frequency analysis, feature extraction, principal component analysis, cluster analysis and inputs the dispersion model for verification, to separate the signal into a reflection signal, a transmission signal and a noise signal, and finally reconstructs the transmission signal and evaluates the separated signal to obtain a high-quality separated transmission signal. The present invention effectively distinguishes between reflection signals, transmission signals and environmental noise by analyzing the diffusion characteristics of sound waves of different frequencies in the material, thereby accurately identifying the energy signal that actually passes through the object. The high-quality separated transmission signal is used for air-coupled ultrasonic detection, which effectively reduces the interference of reflection signals and noise signals and improves the detection accuracy.

[0281] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0282] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the patent application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent application shall be subject to the attached claims.

Claims

1. A method for separating acoustic emission energy signals, characterized in that: The following steps are involved: Step S1: generating a plurality of single-frequency excitation signals, wherein the frequencies of the plurality of single-frequency excitation signals are logarithmically spaced; Step S2: transmitting a plurality of the single-frequency excitation signals through a transducer, receiving echoes to obtain a time domain signal data set; Step S3: performing time-frequency analysis on the time domain signal data set to obtain a time-frequency spectrum set; Step S4: extracting features from the time-frequency spectrum set to obtain a feature matrix; Step S5: performing principal component analysis on the feature matrix to obtain a dimension-reduced feature matrix; Step S6: performing cluster analysis on the dimension reduction feature matrix to obtain a clustering result, wherein the clustering result includes a category label of each data point, and the category label includes reflection, transmission and noise; Step S7: constructing a dispersion model according to the clustering results and the physical parameters of the material, incorporating the dispersion model into the dispersion model, and obtaining a verified clustering result; Step S8: inputting the verification clustering result into the dispersion model, and based on the cluster center feature identification signal of the verification clustering result, marking the signal in the verification clustering result as a reflection signal, a transmission signal and a noise signal; Step S9: superimposing and processing the transmission signals to obtain a reconstructed transmission signal; Step S10: calculating a separation effect index based on the single-frequency excitation signal and the reconstructed transmission signal.

2. The method for separating acoustic emission energy signals according to claim 1, characterized in that: In the step S1, the single-frequency excitation signal is a sinusoidal wave signal.

3. The method for separating acoustic emission energy signals according to claim 1, characterized in that: In step S3, the time-frequency analysis is continuous wavelet transform: W(a,b)=(1 / sqrt(|a|))∫x(t)ψ*((tb) / a)dt; Among them, ψ(t) represents the mother wavelet, a represents the scale parameter, and b represents the translation parameter.

4. The method for separating acoustic emission energy signals according to claim 1, characterized in that: In step S3, the time-frequency analysis is discrete wavelet transform: W[m,n]=(1 / sqrt(a0 m ))∑x[k]ψ*((k-nb0a0^m) / a0 m ); Among them, m and n are integers, a0>1 represents a fixed expansion step, and b0>0 represents a position parameter.

5. The method for separating acoustic emission energy signals according to claim 1, characterized in that: In step S4, each row of the feature matrix represents a signal of a frequency in the time-frequency spectrum set, and each column represents a feature; the features include peak amplitude, root mean square value, waveform factor, kurtosis, spectral entropy, energy density ratio, center frequency, frequency diffusion width, diffusion coefficient, and group velocity dispersion.

6. The method for separating acoustic emission energy signals according to claim 1, characterized in that: The step S5 comprises the following steps: Step S5.1: Obtain the feature matrix X, the size of which is m×n, where m is the number of samples and n is the number of features; Step S5.2: Calculate the central feature matrix of the feature matrix: X centered =X–μ; Wherein, μ is the mean vector of the feature matrix; Step S5.3: Calculate the covariance matrix C of the centered feature matrix: C=(1 / m)*X centered T *X centered ; Step S5.4: Perform eigenvalue decomposition on the covariance matrix: C = VΛV^T; Where V represents the eigenvector matrix, and Λ represents the diagonal matrix of eigenvalues; Step S5.5: Arrange the eigenvectors of the eigenvector matrix in descending order, and select the first k eigenvectors to form a projection matrix W; wherein k makes the cumulative explained variance ratio reach a preset threshold: explained_variance_ratio=λ i / ∑λ i ; where λ i represents the eigenvalue; Step S5.6: Project the centered feature matrix to the new subspace to obtain the reduced dimension feature matrix: Y=X centered *W。 7. The method for separating acoustic emission energy signals according to claim 1, characterized in that: In step S6, the cluster analysis includes the following steps: Step S6.1: define the neighborhood radius and the minimum number of points in the dense area; Step S6.2: For each point in the reduced dimension feature matrix, if the number of points in its neighborhood is greater than or equal to the minimum number of points in the dense area, it is defined as a core point; Step S6.3: For a core point, add all points in its neighborhood to a cluster, and add all points in the neighborhood of the cluster to the cluster, and repeat until no new points are added to the cluster; Step S6.4: define points that do not belong to any cluster as noise points, and output a clustering result, which includes all the clusters and the noise points.

8. The method for separating acoustic emission energy signals according to claim 1, characterized in that: The step S7 includes the following steps: Step S7.1: According to the clustering result, the transmission signal dispersion analysis calculates the frequency-diffusion relationship, the reflection signal dispersion analysis calculates the frequency dispersion, and calculates the comprehensive dispersion function; Step S7.2: Calculate the diffusion function of each cluster group, check consistency and update the dispersion classification, and calculate the diffusion difference of each cluster group: Step S7.3: According to the updated clustering model, verify based on the theoretical dispersion model and update the dispersion classification.

9. The method for separating acoustic emission energy signals according to claim 1, characterized in that: The step S9 includes the following steps: Step S9.1: constructing a signal model based on the transmission signal: y = Φx + n; Where y represents the measured signal, Φ represents the measurement matrix, x represents the original signal, and n represents the noise; Step S9.2: Compressed sensing problem: min||x||1subject to||y-Φx||2≤ε; Among them, ||·||1 represents the L1 norm, ||·||2 represents the L2 norm, and ε represents the noise level; Step S9.3: Implement the basis pursuit denoising algorithm: min(1 / 2)||y-Φx||2 2 +λ||x||1; Where λ represents the regularization parameter; Step S9.4: Implement the orthogonal matching pursuit algorithm: r0=y, x0=0; For k = 1 to K: i. ii.L k =Λ_{k-1}∪{λ k }; iii. iv.r k =y-Φx k Return x k ; Among them, φ j represents the columns of Φ, represents a submatrix of Φ, whose columns are composed of Λ k index.

10. An acoustic emission sensor, characterized in that: The acoustic emission sensor realizes signal separation through the acoustic emission energy signal separation method according to any one of claims 1 to 9.

Images