Method for evaluating reliability of damage mechanism of single-lap bonding joint of composite material based on acoustic emission

Abstract

The invention provides a composite material single-lap bonding joint damage clustering reliability evaluation method based on acoustic emission. The method comprises the following steps: 1, carrying out a tensile experiment to obtain an acoustic emission signal; 2, providing a noise separation method combining variational mode decomposition (VMD) and fast independent component analysis FastICA to remove environmental noise and ineffective components; 3, proposing a credibility evaluation method based on statistics based on the characteristic parameters of the time-frequency domain of the acoustic emission signal, namely, quantifying the neighborhood change before and after the dimension reduction of the signal by adopting a Pearson's correlation coefficient, and evaluating the credibility of an event according to a credibility threshold value; and 4, based on a clustering algorithm, identifying three kinds of damages including matrix cracking, adhesive layer debonding and fiber fracture, and analyzing the development mechanism. According to the method, the VMD and the FastICA are innovatively combined for noise separation, the clustering credibility evaluation index is provided, the accuracy and reliability of acoustic emission data analysis are effectively improved, and technical support is provided for damage evaluation and health monitoring of composite material single-lap bonding joints and other complex structures.

Description

Technical Field

[0001] The invention relates to the technical field of composite material health monitoring, and in particular to a reliability assessment method for damage mechanism of composite material single-lap adhesive joints based on acoustic emission. Background Art

[0002] Lightweight design is of great significance to improving energy efficiency and achieving energy conservation and emission reduction. High-performance composite materials (such as carbon fiber reinforced polymers) are ideal for achieving this goal because of their advantages such as high strength, high stiffness, corrosion resistance, long fatigue life and light weight. They are currently widely used in aerospace, shipbuilding and automotive fields. Bonded structures are used in many fields because of their light weight and the absence of destructive processes such as drilling. However, bonded joints are prone to defects, which may lead to catastrophic failures. Therefore, it is crucial to use structural health monitoring technology to understand the damage mechanism and ensure structural reliability.

[0003] Acoustic emission monitoring technology is a highly sensitive real-time nondestructive testing method that has significant advantages in characterizing the damage behavior of composite materials and the progression of damage in bonded structures. However, despite the many advantages of acoustic emission technology, its practical application still faces challenges in signal analysis reliability, specifically:

[0004] (1) Noise interference increases the uncertainty of acoustic emission signal analysis. Acoustic emission signals are usually interfered by environmental sources, such as testing machines, ground vibrations, and other external noises. Techniques such as fast Fourier transform, wavelet decomposition, wavelet packet decomposition, modal decomposition, and local mean decomposition have achieved some success in mitigating random or non-periodic noise, but their effectiveness decreases significantly when the noise frequency overlaps or is close to the target signal frequency;

[0005] (2) In the unsupervised classification process, there is still a lack of methods to evaluate the credibility of acoustic emission signals. Existing algorithms usually evaluate the overall effect of clustering by "high similarity within clusters and low similarity between clusters". Dimensionality reduction methods such as t-SNE can retain the similarity between data points, which not only helps to explain clustering analysis, but also can intuitively present the clustering results of other algorithms. However, many studies rely on exhaustive search and empirical selection of hyperparameters (such as perplexity) when using t-SNE, and lack quantitative indicators to evaluate the neighborhood relationship of the data, which in turn affects the reliability of evaluating the clustering results.

[0006] Based on the above research gaps, it is necessary to propose a technique to separate effective signals from environmental noise and introduce a confidence index to quantify the credibility of damage mechanism classification. Summary of the invention

[0007] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a reliability assessment method for the damage mechanism of single-lap adhesive joints of composite materials based on acoustic emission, which uses variational mode decomposition (VMD) and fast independent component analysis (FastICA) to extract useful signals from mixed signals, and uses a t-SNE-based evaluation method to perform credibility assessment on acoustic emission signals. Finally, key features are extracted and Gaussian mixture model (GMM) is used to cluster acoustic emission signals. This method can reliably and accurately study the damage mechanism of single-lap joints of composite materials.

[0008] To achieve the above objectives, the present invention adopts the following technical solutions:

[0009] A reliability assessment method for damage mechanism of composite single-lap adhesive joints based on acoustic emission comprises the following steps:

[0010] Step 1: Conduct tensile test and collect acoustic emission signals

[0011] A mechanical loading experiment was conducted on the single-lap adhesive joint of the composite material to collect the acoustic emission signals generated during the mechanical loading process.

[0012] Step 2: Noise separation of acoustic emission signals

[0013] The acoustic emission signal is separated using variational mode decomposition (VMD) to obtain multiple intrinsic mode functions (IMFs); in order to solve the invalid or pseudo modes that may be generated during the separation process of variational mode decomposition (VMD), the intrinsic mode function (IMFs) is screened using Pearson correlation analysis, and the threshold is set to select the intrinsic mode function (IMFs) with high correlation with the original signal; the original signal and the screened intrinsic mode function (IMFs) are combined into a multidimensional signal as the input of the fast independent component analysis (FastICA); since the arrangement order of the separation results is uncertain, the duration of the acoustic emission is used as the basis for distinguishing environmental noise from effective signals; all signals are subjected to noise separation processing;

[0014] Step 3: Credibility assessment of acoustic emission signals

[0015] Short-time Fourier transform (STFT) was used to extract the time-frequency domain characteristic indicators of the acoustic emission signal: peak frequency and peak amplitude. In addition, count, energy, duration, and amplitude were selected as input features of the t-SNE dimensionality reduction algorithm. The distance between each acoustic emission event and its neighboring acoustic emission events in the pre-embedded space before t-SNE dimensionality reduction and the embedded space after t-SNE dimensionality reduction was calculated, and these events were sorted from near to far. The credibility of a single acoustic emission event was defined as the Pearson correlation coefficient between its distance vectors in the two spaces. Among them, the neighborhood of the acoustic emission event was defined The total number of AE events is multiplied by the "similarity percentage"; the Pearson correlation coefficient between the AE event and the randomly sorted AE event neighborhood is calculated to obtain the zero distribution of the credibility score; based on the zero distribution, two thresholds of clustering credibility are defined: (1) credible threshold: the 95th percentile of the zero distribution; (2) suspicious threshold: the 5th percentile of the zero distribution; if the credibility score of the AE event is higher than the credible threshold, it is marked as a credible event; if the credibility score of the AE event is lower than the suspicious threshold, it is marked as a suspicious event; events with scores between the credible threshold and the suspicious threshold are not marked;

[0016] Step 4: Reliable clustering and mechanism analysis of acoustic emission signals

[0017] Based on the peak frequency and peak amplitude of the time-frequency domain characteristic indicators of the acoustic emission signal, the Gaussian mixture model (GMM) was used to cluster the acoustic emission signals, thereby obtaining the damage label of each acoustic emission event, and calculating the average clustering probability that each acoustic emission event was judged to be a specific damage; based on the above clustering results, the change curve of the normalized cumulative count of different bonding lengths of specific damage with the normalized displacement was calculated to explore the relationship between typical acoustic emission signals and damage evolution.

[0018] In the step 1, a tensile test based on displacement control is performed on the single-lap adhesive joint of the composite material, and the acoustic emission signal is recorded by an acoustic emission data acquisition system.

[0019] In step 2, the noise separation of the acoustic emission signal, the variational mode decomposition (VMD) decomposes the complex signal into multiple intrinsic mode functions (IMFs) by establishing a constrained variational model, and iteratively determines the optimal frequency center and bandwidth of each component; the specific steps are as follows:

[0020] 1) Perform Hilbert transform on each intrinsic mode function IMF to obtain the analytical signal, and down-convert it to baseband according to the estimated center frequency; then calculate the modulation signal bandwidth by derivative method; the problem of minimizing the sum of modal bandwidths and reconstructing the input signal is expressed as a constrained variational problem:

[0021]

[0022] Where:k is the center frequency of the kth mode; u k (t) is the kth mode function; δ(t) is the unit impulse function; * represents the convolution operator; Indicates u k (t) analysis signal, modulating the signal to baseband; Represents partial derivative operation, which is used to estimate the smoothness of bandwidth measurement; j represents the imaginary unit f(t) is the target signal;

[0023] 2) Introducing the Lagrangian function L(*) and the penalty factor α, the constrained variational problem in formula (1) is transformed into an unconstrained variational problem, thus obtaining the following expression:

[0024]

[0025] Where: t (t) is the Lagrange operator term with respect to time t;

[0026] 3) Using alternating multiplication ADMM and λ n+1 Iterative updates are performed to find the saddle point of the Lagrangian function, which corresponds to the optimal solution. During the iteration process, the update of the modal components follows formula (3), and the update of the center frequency follows formula (4). Based on the updated modal components and the corresponding center frequencies, the update of the Lagrangian operator is performed according to formula (5):

[0027]

[0028] Where: Indicates u k (t) The nth iteration value in the Fourier domain; represents the Fourier transform of the target signal f(t), Represents ω k The nth iteration value; τ is the Lagrange operator term step size;

[0029] 4) The above iterative process terminates when the convergence condition is met, that is,

[0030]

[0031] Where: ε is the convergence parameter; when the solution reaches the discrimination accuracy ε, the iteration stops and K modal components are output; the decomposition quality of variational mode decomposition VMD depends especially on the choice of mode number;

[0032] Fast Independent Component Analysis (FastICA) is one of the improved ICA algorithms. This algorithm uses the Newton-Raphson method and takes the maximization of negative entropy as the objective function. Its main idea is to separate multiple mixed signals into independent signal sources. The observed signal is defined as x(t) = [x1(t), x2(t), ..., x K (t)], the source signal is defined as s(t) = [s1(t),s2(t),...,s K (t)]; the mixing relationship is defined as x(t) = A s(t), where A is the mixing coefficient matrix; the purpose of FastICA is to estimate the separation matrix W, which is the inverse matrix of the mixing matrix A, which is used to calculate the independent signals: y(t) = W x(t) = W A s(t); y(t) = [y1(t), y2(t), ..., y N (t)] is the estimate of the source signal s(t), and all signals in y are independent of each other;

[0033] By combining the above-mentioned variational mode decomposition VMD and fast independent component analysis FastICA, the effective signal can be efficiently separated from the mixed signal.

[0034] In step three, the optimal dimensionality reduction parameters of the acoustic emission data are determined by iteratively searching the perplexity values ​​in the range of 5 to 60, and the proportion of suspicious events in the signal space is calculated accordingly, thereby ensuring that the dimensionality reduction method is used to evaluate the credibility of events in specimens with different bonding lengths with consistency and fairness.

[0035] In step 4, the peak frequency and peak amplitude characteristic parameters are selected, and the Gaussian mixture model (GMM) is used to cluster the acoustic emission signals of different damages to obtain three types of damage: matrix cracking, adhesive debonding, and fiber breakage. The average clustering probability of the acoustic emission event being judged as a specific damage type is calculated.

[0036] In step 4, the grid search method is used to optimize the number of clusters and the covariance type in the Gaussian mixture model GMM based on the Bayesian information criterion BIC to find the best clustering parameters; at the same time, the clustering effect is evaluated by the silhouette coefficient and the Davies-Bouldin index. Among them, the silhouette coefficient Silhouette Coefficient is a measure of the internal consistency of the cluster, and the Davies-Bouldin index quantifies the degree of separation and closeness between clusters through the ratio of the intra-class distance to the inter-class distance. Combining these three indicators can ensure the accuracy of the clustering results more comprehensively and objectively.

[0037] In step 4, by analyzing the acoustic emission signals collected from specimens with different bonding lengths, the law and consistency of the change of the normalized cumulative counts of different damages with the normalized displacement are determined; based on the above reliability analysis, the development process of different damages in a single specimen is further studied to reveal the occurrence mechanism of adhesive layer damage.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. High accuracy: Through noise separation technology, useful signals can be effectively extracted from mixed signals, so that the analysis is based on pure signals without noise, thereby reducing the uncertainty of cluster analysis to a certain extent and improving the reliability of damage mechanism research;

[0040] 2. High reliability: Based on the credibility assessment method, it can provide a confidence reference for the clustering results;

[0041] 3. Real-time monitoring: Based on acoustic emission monitoring technology, the health status of the composite structure can be obtained in real time, providing a reliable reference for exploring the mechanism of the damage process. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of the method of the present invention;

[0043] Figure 2 It is a flow chart of the noise separation method in the embodiment;

[0044] Figure 3 The noise separation result in the embodiment;

[0045] Figure 4 It is a method for evaluating the credibility of acoustic emission events and data sets in the embodiment;

[0046] Figure 5 (a) is the t-SNE visualization result of the acoustic emission signal collected at a bonding length of 30 mm in the embodiment at the optimal perplexity;

[0047] Figure 5 (b) is the t-SNE visualization result of the acoustic emission signal collected at a bonding length of 40 mm in the embodiment at the optimal perplexity;

[0048] Figure 5 (c) is the t-SNE visualization result of the acoustic emission signal collected at a bonding length of 50 mm in the embodiment at the optimal perplexity;

[0049] Figure 6 (a) is a credibility score heat map of the original acoustic emission signal collected at a bonding length of 50 mm in the embodiment;

[0050] Figure 6(b) is a credibility score heat map of the effective acoustic emission signal collected at a bonding length of 50 mm after noise separation in the embodiment;

[0051] Figure 7 (a) is the clustering result of the acoustic emission signal collected at a bonding length of 30 mm in the embodiment based on the noise reduction data.

[0052] Figure 7 (b) is the clustering result of the acoustic emission signal collected at a bonding length of 40 mm in the embodiment based on the noise reduction data.

[0053] Figure 7 (c) is the clustering result of the acoustic emission signal collected at a bonding length of 50 mm in the embodiment based on the noise reduction data.

[0054] Figure 8 (a) is the consistency curve of normalized displacement and normalized cumulative count of acoustic emission signal of substrate cracking at different bonding lengths in the embodiment.

[0055] Figure 8 (b) is the consistency curve of normalized displacement and normalized cumulative count of the acoustic emission signal of adhesive layer debonding at different bonding lengths in the embodiment.

[0056] Figure 8 (c) is the consistency curve of normalized displacement and normalized cumulative count of the acoustic emission signal of fiber breakage at different bonding lengths in the embodiment. DETAILED DESCRIPTION

[0057] In order to better understand the technical solution of the present invention, the specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings need not be drawn to scale unless otherwise specified.

[0058] The entire test system includes a tensile device, an acoustic emission sensor, a data acquisition system, a computer and a composite single-lap adhesive joint. The acoustic emission sensor is fixed to the surface of the composite single-lap adhesive joint by adhesion or clamping, and the acoustic emission signal is obtained through the data acquisition system during mechanical experiment loading.

[0059] like Figure 1 As shown, a reliability assessment method for damage mechanism of composite single-lap adhesive joint based on acoustic emission comprises the following steps:

[0060] Step 1: Conduct tensile test and collect acoustic emission signals

[0061] The composite adhesive joints were subjected to mechanical experimental loading to obtain the force-loading displacement curve of the composite adhesive joints and the acoustic emission signals during the loading process. The acquisition frequency of the acoustic emission signals was set to 2MHz, the pre-amplifier gain was set to 40dB, and the threshold was set to 50dB.

[0062] Step 2: Noise separation of acoustic emission signals

[0063] The acoustic emission signal is separated using variational mode decomposition (VMD) to obtain multiple intrinsic mode functions (IMFs). In order to solve the invalid or pseudo modes that may be generated during the separation process of variational mode decomposition (VMD), the intrinsic mode function (IMFs) is screened using Pearson correlation analysis, and a specific threshold is set to select IMFs with a high correlation with the original signal. The original signal and the screened IMFs are combined into a multidimensional signal as the input of the fast independent component analysis (FastICA). Since the arrangement order of the FastICA separation results is uncertain, acoustic emission indicators (such as duration) are used to distinguish between environmental noise and damage signals, and all signals are subjected to noise separation processing. The process is as follows: Figure 2 As shown;

[0064] The parameters of variational mode decomposition (VMD) are set as: alpha = 2000, init = 1, tol = 1e-7, K = 4; the parameters of fast independent component analysis (FastICA) are set as: n_component = 2, default = unit-variance, neg-entropy: logcosh. The separation results are shown in Figure 2. Figure 3 As shown, from Figure 3 It can be seen that after using the VMD-FastICA separation method, Figure 3 The original signal of (a) is decomposed into two components, Figure 3 The signal with a peak frequency of 316.41kHz in (b) has a short duration and a significant amplitude, and is identified as a valid damage signal. Figure 3 The signal in (c) has a longer duration, indicating that it belongs to the ambient noise. This shows that the proposed method can effectively separate the useful acoustic emission signal from the ambient noise.

[0065] Step 3: Credibility assessment of acoustic emission signals

[0066] Credibility assessment process Figure 4As shown. First, the short-time Fourier transform (STFT) is used to extract the time-frequency domain feature indicators of the acoustic emission signal, and the amplitude, count, energy and duration are used as the input features of the t-SNE dimensionality reduction. Among them, the parameters of the short-time Fourier transform (STFT) are set to: nperseg=256, noverlap=128, the length of the Kaiser window is 256, and the shape parameter is 5. In the pre-embedded space before t-SNE dimensionality reduction and the embedded space after dimensionality reduction, the distance between each acoustic emission event and other events in its neighborhood is calculated, and these events are sorted from near to far according to distance. The confidence of a single acoustic emission event is defined as the Pearson correlation coefficient between its two distance vectors. Among them, the neighborhood of an acoustic emission event is defined as the total number of acoustic emission events multiplied by the "similarity percentage". The zero distribution of the credibility score can be obtained by calculating the Pearson correlation coefficient between the acoustic emission event and the randomly sorted acoustic emission event neighborhood. Based on the null distribution, two confidence thresholds are defined: (1) credible threshold: the 95th percentile of the null distribution; (2) suspicious threshold: the 5th percentile of the null distribution. If the confidence score is higher than the credible threshold, it is marked as a credible event; if the confidence score is lower than the suspicious threshold, it is marked as a suspicious event; events with scores between the credible threshold and the suspicious threshold are recorded as intermediate events. The optimal perplexity for dimensionality reduction is determined by iteratively searching within the perplexity range of 5 to 60 and combining the proportion of suspicious events in the acoustic emission signal space. Table 1 lists the dimensionality reduction of acoustic emission data of three specimens with different bonding lengths at different perplexity values, where PDE represents the proportion of suspicious events and PTE represents the proportion of credible events. For specimens with bonding lengths of 30 mm and 40 mm, the dimensionality reduction effect is best when the perplexity is 60; while for specimens with a bonding length of 50 mm, the dimensionality reduction effect is best when the signal perplexity is 5.

[0067] Table 1 Basis for selecting hyperparameters of acoustic emission data of specimens with different bonding lengths

[0068]

[0069] The credibility evaluation results and thermal images of acoustic emission signals collected on specimens with three different bonding lengths are shown in Figure 2. Figure 5 (a), (b), (c) and Figure 6 As shown in (a) and (b). Figure 5 The clustering characteristics of the dataset under the optimal hyperparameter dimensionality reduction are shown. There are clear boundaries between different damage types, which are usually located at the boundary between isolated data clusters or two clusters (such as black points). In contrast, credible events show a high degree of consistency with the characteristic patterns of the data in the neighborhood, and are usually distributed in the center of the cluster area (such as medium gray points), indicating that the credibility of their clustering is high. Figure 6As shown in the figure, after using VMD-FastICA noise separation, the number of light gray points increases, indicating that the credibility of the data has generally improved. Specifically, after the VMD-FastICA noise separation algorithm, the confidence of events with lower peak amplitudes increases. At the same time, in areas with higher peak amplitudes, the number of low-confidence events also decreases due to the removal of noise signals.

[0070] Step 4: Reliable clustering and mechanism analysis of acoustic emission signals

[0071] Based on the characteristic index of acoustic emission signal characteristics, Gaussian mixture model GMM is used to cluster acoustic emission signals. The results based on Bayesian information criterion BIC, silhouette coefficient and Davies-Bouldin index show that GMM can achieve the best clustering when the number of clusters is 4. The distribution of different damages in peak frequency, peak amplitude and time is shown in Figure 2. Figure 7 As shown in (a), (b), and (c), it can be seen from the figure that the acoustic emission signals are divided into four clusters based on the peak frequency. The acoustic emission events with a frequency lower than 150kHz correspond to matrix cracking, and the acoustic emission events with a frequency of about 300kHz are related to fiber breakage, while the signals between these two frequencies are judged to be adhesive debonding. Signals with larger peak amplitudes are classified as other signals. These signals mostly occur when the specimen is about to break and may contain multiple types of damage. Since the proportion of this type of signal is small, it is not further analyzed in this invention.

[0072] The average probability of acoustic emission events of different damages being clustered into specific damages is shown in Table 2. The average probability of acoustic emission signals being predicted to be different damages after noise separation using VMD-FastICA is significantly improved. The change curve of normalized cumulative counts of different bonding lengths under specific damage with normalized displacement is shown in Figure 8 As shown in (a), (b), and (c), it can be seen from the figure that the development process of different damages of specimens with different bonding lengths has a certain consistency.

[0073] Table 2 Average clustering probability of different damages

[0074]

[0075] The present invention innovatively proposes a noise separation method and a credibility assessment method, which utilizes mature signal processing algorithms and dimensionality reduction algorithms. The operation is simple and convenient, and the method can adapt to complex and harsh external field conditions, ensure the accuracy of the acoustic emission clustering process, and realize reliable research on the structural damage mechanism of single-lap adhesive joints of composite materials.

[0076] Final Notes

[0077] The above-mentioned embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them. Although the present invention has been described in detail with reference to the above-mentioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the above-mentioned embodiments can still be modified, or some or all of the technical features therein can be replaced by equivalents; these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A reliability assessment method for damage mechanism of composite single-lap adhesive joints based on acoustic emission, characterized by: The following steps are involved: Step 1: Conduct tensile test and collect acoustic emission signals A mechanical loading experiment was conducted on the single-lap adhesive joint of the composite material to collect the acoustic emission signals generated during the mechanical loading process. Step 2: Noise separation of acoustic emission signals The acoustic emission signal is separated using variational mode decomposition (VMD) to obtain multiple intrinsic mode functions (IMFs); in order to solve the invalid or pseudo modes that may be generated during the separation process of variational mode decomposition (VMD), the intrinsic mode function (IMFs) is screened using Pearson correlation analysis, and the threshold is set to select the intrinsic mode function (IMFs) with high correlation with the original signal; the original signal and the screened intrinsic mode function (IMFs) are combined into a multidimensional signal as the input of the fast independent component analysis (FastICA); since the arrangement order of the separation results is uncertain, the duration of the acoustic emission is used as the basis for distinguishing environmental noise from effective signals; all signals are subjected to noise separation processing; Step 3: Credibility assessment of acoustic emission signals Short-time Fourier transform (STFT) was used to extract the time-frequency domain characteristic indicators of the acoustic emission signal: peak frequency and peak amplitude. In addition, count, energy, duration, and amplitude were selected as input features of the t-SNE dimensionality reduction algorithm. The distance between each acoustic emission event and its neighboring acoustic emission events in the pre-embedded space before t-SNE dimensionality reduction and the embedded space after t-SNE dimensionality reduction was calculated, and these events were sorted from near to far. The credibility of a single acoustic emission event was defined as the Pearson correlation coefficient between its distance vectors in the two spaces. Among them, the neighborhood of the acoustic emission event was defined The total number of AE events is multiplied by the "similarity percentage"; the Pearson correlation coefficient between the AE event and the randomly sorted AE event neighborhood is calculated to obtain the zero distribution of the credibility score; based on the zero distribution, two thresholds of clustering credibility are defined: (1) credible threshold: the 95th percentile of the zero distribution; (2) suspicious threshold: the 5th percentile of the zero distribution; if the credibility score of the AE event is higher than the credible threshold, it is marked as a credible event; if the credibility score of the AE event is lower than the suspicious threshold, it is marked as a suspicious event; events with scores between the credible threshold and the suspicious threshold are not marked; Step 4: Reliable clustering and mechanism analysis of acoustic emission signals Based on the peak frequency and peak amplitude of the time-frequency domain characteristic indicators of the acoustic emission signal, the Gaussian mixture model (GMM) was used to cluster the acoustic emission signals, thereby obtaining the damage label of each acoustic emission event, and calculating the average clustering probability that each acoustic emission event was judged to be a specific damage; based on the above clustering results, the change curve of the normalized cumulative count of different bonding lengths of specific damage with the normalized displacement was calculated to explore the relationship between typical acoustic emission signals and damage evolution.

2. The reliability assessment method for damage mechanism of composite single-lap adhesive joints based on acoustic emission according to claim 1 is characterized in that: In the step 1, a tensile test based on displacement control is performed on the single-lap adhesive joint of the composite material, and the acoustic emission signal is recorded by an acoustic emission data acquisition system.

3. The reliability assessment method for damage mechanism of composite single-lap adhesive joints based on acoustic emission according to claim 1 is characterized by: In step 2, the noise separation of the acoustic emission signal, the variational mode decomposition (VMD) decomposes the complex signal into multiple intrinsic mode functions (IMFs) by establishing a constrained variational model, and iteratively determines the optimal frequency center and bandwidth of each component; the specific steps are as follows: 1) Perform Hilbert transform on each intrinsic mode function IMF to obtain the analytical signal, and down-convert it to baseband according to the estimated center frequency; then calculate the modulation signal bandwidth by derivative method; the problem of minimizing the sum of modal bandwidths and reconstructing the input signal is expressed as a constrained variational problem: Where: k is the center frequency of the kth mode; u k (t) is the kth mode function; δ(t) is the unit impulse function; * represents the convolution operator; Indicates u k (t) analysis signal, modulating the signal to baseband; Represents partial derivative operation, which is used to estimate the smoothness of bandwidth measurement; j represents the imaginary unit f(t) is the target signal; 2) Introducing the Lagrangian function L(*) and the penalty factor α, the constrained variational problem in formula (1) is transformed into an unconstrained variational problem, thus obtaining the following expression: Where: t (t) is the Lagrange operator term with respect to time t; 3) Using alternating multiplication ADMM and λ n+1 Iterative updates are performed to find the saddle point of the Lagrangian function, which corresponds to the optimal solution. During the iteration process, the update of the modal components follows formula (3), and the update of the center frequency follows formula (4). Based on the updated modal components and the corresponding center frequencies, the update of the Lagrangian operator is performed according to formula (5): Where: Indicates u k (t) The nth iteration value in the Fourier domain; represents the Fourier transform of the target signal f(t), Represents ω k The nth iteration value; τ is the Lagrange operator term step size; 4) The above iterative process terminates when the convergence condition is met, that is, Where: ε is the convergence parameter; when the solution reaches the discrimination accuracy ε, the iteration stops and K modal components are output; the decomposition quality of variational mode decomposition VMD depends especially on the choice of mode number; Fast Independent Component Analysis (FastICA) is one of the improved ICA algorithms. This algorithm uses the Newton-Raphson method and takes the maximization of negative entropy as the objective function. Its main idea is to separate multiple mixed signals into independent signal sources. The observed signal is defined as x(t) = [x1(t), x2(t), ..., x K (t)], the source signal is defined as s(t) = [s1(t),s2(t),...,s K (t)]; the mixing relationship is defined as x(t) = A s(t), where A is the mixing coefficient matrix; the purpose of FastICA is to estimate the separation matrix W, which is the inverse matrix of the mixing matrix A, which is used to calculate the independent signals: y(t) = W x(t) = W A s(t); y(t) = [y1(t), y2(t), ..., y N (t)] is the estimate of the source signal s(t), and all signals in y are independent of each other; By combining the above-mentioned variational mode decomposition VMD and fast independent component analysis FastICA, the effective signal can be efficiently separated from the mixed signal.

4. The reliability assessment method for damage mechanism of composite single-lap adhesive joints based on acoustic emission according to claim 1 is characterized by: In step three, the proportion of suspicious events in the AE signal space is calculated by iteratively searching different perplexity values ​​ranging from 5 to 60 to find the optimal hyperparameters for AE data dimensionality reduction.

5. The reliability assessment method for damage mechanism of composite single-lap adhesive joints based on acoustic emission according to claim 1 is characterized in that: In step 4, the peak frequency and peak amplitude are selected as characteristic parameters, and the Gaussian mixture model (GMM) is used to cluster the acoustic emission signals to obtain three types of damage: matrix cracking, adhesive layer debonding, and fiber breakage. The average clustering probability of the acoustic emission event being judged as a specific damage type is calculated.

6. The reliability assessment method for damage mechanism of composite single-lap adhesive joints based on acoustic emission according to claim 1 is characterized by: In step 4, based on the Bayesian information criterion (BIC), the grid search method is used to optimize the number of clusters and the covariance type in the Gaussian mixture model (GMM) to find the optimal clustering parameters; at the same time, the clustering effect is evaluated by the Silhouette Coefficient and the Davies-Bouldin Index to determine the optimal number of clusters.

7. The reliability assessment method for damage mechanism of composite single-lap adhesive joints based on acoustic emission according to claim 1 is characterized by: In step 4, by analyzing the acoustic emission signals collected from specimens with different bonding lengths, the law and consistency of the change of the normalized cumulative counts of different damages with the normalized displacement are determined; based on the above reliability analysis, the development process of different damages in a single specimen is further studied to reveal the occurrence mechanism of adhesive layer damage.

Images