An adaptive identification method for concrete damage mechanism

Through the adaptive method combined with Grid Search method and Gaussian hybrid model, feature selection and cluster analysis are optimized, which solves the problem of cumbersome feature selection in concrete acoustic emission damage recognition, improves recognition accuracy and adaptability, and is suitable for different types of Concrete specimens.

CN118734102BActive Publication Date: 2024-12-06HOHAI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411223601.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-12-06
Estimated Expiration
2044-09-03

AI Technical Summary

Technical Problem

The existing concrete acoustic emission damage recognition method is complicated to select features in the study of new material loss mechanisms, making it difficult to obtain ideal and realistic clustering results, which affects the recognition accuracy.

Method used

The Grid Search method combined with the Gaussian mixed model GMM is used to optimize multi-objectively through feature importance and similarity indicators, establish an adaptive concrete damage mechanism recognition model, and use normalized processing and cluster analysis to identify acoustic signal characteristics to improve cluster analysis computing efficiency and accuracy.

Benefits of technology

It improves the adaptability and accuracy of damage recognition of concrete specimens, reduces the complexity of feature selection, and provides stronger acoustic signal sensitivity characteristic parameters, which are suitable for different types of concrete specimens.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118734102B_ABST
    Figure CN118734102B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for adaptively identifying concrete damage mechanisms, including: performing a three-point bending loading test on a concrete beam, collecting the original acoustic emission signal of the sample; obtaining characteristic parameters characterizing the acoustic emission characteristics, and performing normalization processing; using Grid Search combined with Gaussian mixture model GMM clustering to establish an adaptive concrete damage mechanism identification model; and analyzing the damage mechanism of the sample based on the clustering results. The present invention not only realizes efficient and accurate damage mechanism identification by incorporating a multi-objective optimization mechanism, but can also be applied to the exploration of different concrete damage mechanisms, reducing the workload of manual analysis of acoustic emission characteristics.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of concrete material damage monitoring, and in particular relates to a concrete damage mechanism adaptive identification method. Background Art

[0002] Concrete materials are affected by external factors during use, which can cause various types of damage, which will significantly affect their mechanical properties and service life. Therefore, it is of great significance to identify the damage mechanism of concrete materials in a timely and accurate manner.

[0003] Acoustic emission detection is a real-time nondestructive testing technology that can capture the acoustic signals caused by tiny strains and microcracks when the material is subjected to stress, directly reflecting the generation and development of internal defects in the material. The existing concrete acoustic emission damage identification method is mainly based on unsupervised learning. By simply judging various acoustic emission features, the features with greater influence are found, and the trial algorithm is used to determine the appropriate number of features and clusters. Then, the clustering algorithm is used to divide the data into multiple categories, such as the paper Han Q., Yang G., Xu J., et al. Acoustic emission data analysesbased on crumb rubber concrete beam bending tests. Engineering FractureMechanics, 2019, 210, 189–202.

[0004] Although this type of method has been applied in damage identification of some composite materials, due to the diversity and complexity of acoustic signals of different concrete materials during the destruction process, the feature selection before clustering has a great impact on the accuracy of clustering identification. When applied to the study of the loss mechanism of a new material, the above method has the disadvantages of cumbersome feature selection process and difficulty in obtaining ideal and practical clustering results. Summary of the invention

[0005] The present invention aims to provide a method for identifying a concrete material self-adaptive damage mechanism, so as to solve the above-mentioned deficiencies in the prior art.

[0006] In order to solve the above problems, the present invention adopts the following solutions:

[0007] A method for adaptively identifying concrete damage mechanisms comprises the following steps:

[0008] Step S1: Prepare a concrete beam sample and conduct a three-point bending loading test, and collect acoustic emission signals from the entire process of the sample until failure.

[0009] Step S2: Acquire characteristic parameters representing concrete damage information based on the collected acoustic emission signals, and normalize the first data set corresponding to each characteristic parameter, and use the normalization result as the characteristic sample set for cluster analysis. The first data set corresponding to each characteristic parameter includes the value of the characteristic parameter at different times.

[0010] Step S3: Clustering is performed using the Grid Search method combined with the Gaussian mixture model GMM. The dimension of the feature samples and the value range of the number of clusters are used as constraints. The importance index of the feature samples, the similarity index and the silhouette coefficient of the clustering results are used as optimization targets to establish an adaptive concrete damage mechanism identification model.

[0011] Step S4: The acoustic signal sensitive characteristics of the concrete sample are clarified according to the characteristic parameter selection results, and the damage mechanism of the concrete sample is calibrated according to the comparative analysis of the clustering results.

[0012] Further optimized, the characteristic parameters characterizing the concrete damage information extracted in step S2 include several or all of ring count, amplitude, rise time, duration, energy, signal strength, peak frequency, centroid frequency, RA value and average frequency.

[0013] For further optimization, since different feature parameters in the data have different dimensions, in order to make the influence of different feature parameters on clustering equal, the normalization method is used to convert all data into the range of 0-1. The normalization processing formula is:

[0014] ;

[0015] In the formula, Y i,norm is any characteristic parameter Y The corresponding first data set i The value after normalization of samples; Y i is the first data set corresponding to the feature parameter i The value of samples before normalization; Y max and Y min are respectively the maximum and minimum values ​​in the first data set corresponding to the characteristic parameter.

[0016] Further optimization, the specific steps of step S3 are:

[0017] Step S31: Establish a Grid Search global search network based on the initial feature dimension of the acoustic emission signal, the dimension after dimension reduction based on feature importance, the constraint relationship of the dimension after dimension reduction based on feature similarity, and the number of clusters. Grid Search is suitable for global optimization of small data sets to avoid falling into the local optimum.

[0018] Step S32: Calculate the Laplace score of each feature parameter and sort them from low to high, retaining the top N 1 feature parameter; then calculate the feature similarity index of the retained feature parameters and sort them from low to high, retaining the top N 2 feature parameters. The present invention combines the Laplace score and the cosine similarity index as the basis for feature parameter selection, while taking into account feature importance and feature similarity, thereby improving the computational efficiency and accuracy of cluster analysis. Among them, the smaller the Laplace score of the feature, the greater the feature importance; the larger the cosine similarity index, the greater the overall similarity of the feature.

[0019] Step S33: Based on the feature parameters retained after two feature selections in step S32, a cluster number of k The Gaussian mixture model (GMM) is used for clustering, and the silhouette coefficient is used to evaluate the clustering results.

[0020] Step S34: Change N 1. N 2. k The value of is repeated, steps S32-S33 are repeated to generate the Pareto optimal solution that meets the objective function of clustering the damage mechanism of concrete specimens.

[0021] Further optimization, in step S31, the constraints constituting the Grid Search global search network model are as follows:

[0022] ;

[0023] In the above formula, N is the initial number of characteristic parameters obtained from the acoustic emission signal.

[0024] Further optimization, in step S32, the Laplace score of the feature parameter, that is, the importance of the feature parameter, is calculated as follows:

[0025] (1) The distance matrix between samples in the sample set of each feature parameter Y is obtained by the K nearest neighbor method S ,in, n is the total number of samples in the sample set of a characteristic parameter;

[0026] (2) Calculate the diagonal matrix D and the Laplacian matrix L

[0027] ;

[0028] L = DS;

[0029] In the above formula, b , c ∈[1, n ], S bc Represents the distance matrix S Middle b Line c Column data, D bc represents the diagonal matrix D b Line c Column data.

[0030] (3) Calculate the Laplace score

[0031] ;

[0032] In the above formula, f r Indicates r All samples in the sample set with feature parameters Y, r ∈[1,N], express f r The transpose of .

[0033] Further optimization, in step S32, the calculation process of the feature similarity index is as follows:

[0034] (1) Calculate the similarity between two feature parameters. The formula is:

[0035] ;

[0036] In the above formula, A and B represent any two of the N1 feature parameters retained after dimensionality reduction based on feature importance. A i The first i samples, B i The first i samples; The larger the value, the greater the similarity between the two feature parameters of A and B.

[0037] (2) Calculate the similarity index of the features, the formula is:

[0038] ;

[0039] In the above formula, Represents characteristic parametersA and any characteristic parameter X The similarities between X Selected and retained based on feature importance N 1 characteristic parameter m feature parameters.

[0040] Further optimization, in step S33, the concrete damage mechanism clustering process based on GMM is as follows:

[0041] Step S331: The optimized dimensionality reduction based on feature similarity is N The second data set consisting of 2 feature parameters is used as the input of the Gaussian mixture model GMM, and the number of clusters is set to k , then keep the Gaussian distribution unchanged, and calculate the posterior probability of each data point belonging to each Gaussian distribution through the expectation step-maximization step algorithm. Each data point includes N Samples of two feature parameters at the same time; then update the Gaussian distribution parameters according to the posterior probability to maximize the loss function; the details are as follows:

[0042] (1) Expectation step: Under a given multidimensional Gaussian distribution, calculate any data point according to Bayes’ theorem x i The probability of belonging to the pth Gaussian distribution :

[0043] ;

[0044] (2) Maximization step: Update the Gaussian distribution parameters based on the probability obtained in the expectation step:

[0045] ;

[0046] ;

[0047] ;

[0048] In the above formula, x i For the i data points, For data points x i The expectation of the p-th Gaussian distribution, For data points x i The covariance belonging to the p-th Gaussian distribution, is the weight of the pth Gaussian distribution in the Gaussian mixture model, and satisfies ; is the probability density function of the acoustic emission characteristic sequence that obeys the pth Gaussian distribution, expressed as:

[0049] ;

[0050] In the above formula, .

[0051] Step S332: Use the logarithmic form of the maximum likelihood function as the loss function, and repeat step S331 until the change in the loss function is less than the threshold or the maximum number of iterations is reached. The loss function is expressed as:

[0052] .

[0053] Further optimization, in step S33, the calculation formula of the overall silhouette coefficient S is:

[0054] ;

[0055] in s ( i ) is the i The silhouette coefficient of the data points is expressed as: ;

[0056] In the above formula, a ( i ) is the i The average distance between a data point and other data points in its cluster; b ( i ) is the i The minimum value of the average distance between a data point and other data points in the cluster. The larger the overall silhouette coefficient, the closer the distance between the data points in the cluster, the farther the distance between the data points in the cluster, and the better the clustering effect.

[0057] Further optimization, in step S4, the identification process of the concrete sample damage mechanism includes aligning the clustering results with the event occurrence time, and drawing a time-event accumulation graph, and analyzing the growth characteristics of each damage signal through the graph to identify the damage mechanism of the sample.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. By constructing an adaptive damage identification method based on Grid Search, the type of concrete specimens selected is not restricted, which effectively improves the adaptability of existing methods in damage identification of different concrete specimens.

[0060] 2. The multi-objective optimization method is adopted to consider both feature importance and feature similarity at the same time, which can not only retain important indicators but also effectively remove redundancy, thereby improving the computational efficiency and accuracy of cluster analysis.

[0061] 3. The characteristic parameters obtained through two feature selections have stronger sensitivity and can be used as damage warning indicators for this type of concrete specimens. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a flow chart of the method for adaptively identifying concrete damage mechanisms according to the present invention;

[0063] Figure 2 This is a diagram showing the clustering results of the acoustic signal of sample 1 in Example 1;

[0064] Figure 3 This is a clustering result diagram of the acoustic signal of sample 2 in Example 1;

[0065] Figure 4 This is a schematic diagram of the Grid Search optimization result in Example 2;

[0066] Figure 5 This is the damage signal time-event cumulative distribution diagram of sample 3 in Example 2. DETAILED DESCRIPTION

[0067] The specific implementation method of the present invention is given below, and the technical solution of the present invention is clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0068] Embodiment 1: Figure 1 As shown in FIG. 1 , a method for adaptively identifying concrete damage mechanisms is described. The specific steps are as follows:

[0069] Step S1: prepare two ordinary concrete beam specimens, denoted as specimen 1 and specimen 2, and conduct three-point bending loading tests respectively, and collect acoustic emission signals of the whole process until the specimens are destroyed.

[0070] Step S2: According to the collected acoustic emission signal, five characteristic parameters characterizing the signal characteristics are obtained, including rise time, count, duration, energy and amplitude. Then, the first data set corresponding to each characteristic parameter is normalized, and the normalized result is used as the sample set for cluster analysis. The first data set includes the values ​​of the characteristic parameter collected at different times, a total of n values. Partial data of the five characteristic parameters obtained from the acoustic emission signals of sample 1 and sample 2 are shown in Table 1 and Table 2.

[0071] Table 1 Partial data of acoustic emission characteristic parameters of sample 1

[0072]

[0073] Table 2 Partial data of acoustic emission characteristic parameters of sample 2

[0074]

[0075] Step S3: Clustering is performed using the Grid Search method combined with the Gaussian mixture model GMM, with the dimension of the feature sample and the value range of the number of clusters as constraints, and the importance index, similarity index of the feature sample and the silhouette coefficient of the clustering result as the optimization target, to establish an adaptive concrete damage mechanism identification model. The specific steps are:

[0076] Step S31: Initial number of characteristic parameters obtained based on acoustic emission signals N , the number of feature parameters after dimensionality reduction based on feature importance N 1 、 The number of feature parameters after dimensionality reduction based on feature similarity N 2, and the number of clusters, as constraints, to establish the Grid Search global search network model.

[0077] In this embodiment, the constraints constituting the Grid Search model are as follows:

[0078] ;

[0079] In this embodiment, N is 5.

[0080] Step S32: Calculate the Laplace score of each feature parameter and sort them from low to high, retaining the top N 1 feature parameter; then calculate the feature similarity index of the retained feature parameters and sort them from low to high, retaining the top N 2 characteristic parameters. Specifically:

[0081] The Laplace fraction calculation process of the characteristic parameter is as follows:

[0082] (1) The distance matrix between samples in the sample set of each feature parameter is obtained by the K nearest neighbor method S ,in, n is the total number of samples in the sample set of a characteristic parameter;

[0083] (2) Calculate the diagonal matrix D and the Laplace matrix L

[0084] ;

[0085] L = DS;

[0086] In the above formula, b,c∈[1,n], S bc Represents the distance matrixS Middle b Line c Column data, D bc represents the diagonal matrix D b Line c Column data.

[0087] (3) Calculate the Laplace score

[0088] ;

[0089] In the above formula, f r Indicates r All samples in feature parameters, r∈[1,N], represent f r The transpose of .

[0090] The calculation process of feature similarity index is as follows:

[0091] (1) Calculate the similarity between two feature parameters. The formula is:

[0092] ;

[0093] In the above formula, A and B represent any two of the N1 feature parameters retained after dimensionality reduction based on feature importance. A i The first i samples, B i The first i samples; The larger the value, the greater the similarity between the two feature parameters of A and B.

[0094] (2) Calculate the similarity index of the features, the formula is:

[0095] ;

[0096] In the above formula, Represents characteristic parameters A and any characteristic parameter X The similarities between X Selected and retained based on feature importance N 1 characteristic parameter m feature parameters.

[0097] Step S33: Establish clustering number of feature parameters retained after two feature selections in step S32 k The Gaussian mixture model GMM is used for clustering, and the silhouette coefficient is used to evaluate the clustering results. The specific process is as follows:

[0098] Step S331: The optimized dimensionality reduction based on feature similarity is N The second data set consisting of 2 feature parameters is used as the input of the Gaussian mixture model GMM, and the number of clusters is set to k , then keep the Gaussian distribution unchanged, and calculate the posterior probability of each data point belonging to each Gaussian distribution through the expectation step-maximization step algorithm. Each data point includes N The samples with two characteristic parameters at the same time, for example, the values ​​of five characteristic parameters in each row in Table 1, are normalized to form a data point; then the Gaussian distribution parameters are updated according to the posterior probability to maximize the loss function. The details are as follows:

[0099] (1) Expectation step: Under a given multidimensional Gaussian distribution, the probability of each data point belonging to the pth Gaussian distribution is calculated according to Bayes’ theorem. :

[0100] ;

[0101] (2) Maximization step: Update the Gaussian distribution parameters based on the probability obtained in the expectation step:

[0102] ;

[0103] ;

[0104] ;

[0105] In the above formula, x i is the data point, For data points x i The expectation of the p-th Gaussian distribution, For data points x i The covariance belonging to the p-th Gaussian distribution, is the weight of the pth Gaussian distribution in the Gaussian mixture model, and satisfies .

[0106] is the probability density function of the acoustic emission characteristic sequence that obeys the pth Gaussian distribution, expressed as:

[0107] ;

[0108] In the above formula, .

[0109] Step S332: Use the logarithmic form of the maximum likelihood function as the loss function, and repeat step S331 until the change in the loss function is less than the threshold or the maximum number of iterations is reached. The loss function is expressed as:

[0110] .

[0111] The calculation formula of the overall silhouette coefficient S is:

[0112] ;

[0113] in s ( i ) is the silhouette coefficient of the ith data point, expressed as:

[0114] ;

[0115] In the above formula, a ( i ) is the i The average distance between a data point and other data points in its cluster; b ( i ) is the i The minimum value of the average distance between a data point and other data points in the cluster. The larger the overall silhouette coefficient, the closer the distance between the data points in the cluster, the farther the distance between the data points in the cluster, and the better the clustering effect.

[0116] Step S34: Change N 1. N 2. k The above steps are repeated to generate the Pareto optimal solution that meets the objective function of clustering the damage mechanism of concrete specimens.

[0117] In this embodiment, the results of acoustic emission feature optimization for damage mechanism identification of two common concrete beam specimens of the same type are shown in Table 3. As can be seen from Table 3, the characteristic parameters selected by the concrete damage mechanism identification method of the present invention for the two specimens are both rise time and energy, which are clustered into 2 clusters, and the silhouette coefficients are both greater than 0.95. The clustering results of specimen 1 are shown in Table 3. Figure 2 The clustering results of sample 2 are shown in Figure 3 shown.

[0118] Table 3 Optimization results of acoustic emission characteristics of damage mechanism of sample 1 and sample 2

[0119]

[0120] Step S4: The sensitive characteristics of the acoustic emission signal of the concrete sample are clarified through the characteristic parameter selection results, and the damage mechanism of the concrete sample is calibrated based on the comparative analysis of the clustering results.

[0121] In this embodiment, the damage type clustering effect of the two samples is good, and the damage of the two samples is determined to be matrix cracking and aggregate cracking, which is consistent with the damage mechanism of ordinary concrete.

[0122] Example 2: In this example, the prepared sample is a steel fiber concrete sample, which is recorded as sample 3. The acoustic emission signal in this example is collected from a three-point bending loading test of a steel fiber concrete beam sample.

[0123] Compared with ordinary concrete, the acoustic emission characteristics of the damage mechanism of steel fiber concrete are not clear, and different selections of acoustic emission characteristic parameters will affect the clustering of acoustic emission signals. In this embodiment, there are 10 characteristic parameters of the sample acoustic emission signal collected, including ring count, amplitude, rise time, duration, energy, signal strength, peak frequency, centroid frequency, RA value and average frequency. Therefore, the initial constraints in the Grid Search search network are N Take it as 10. Some data of 10 characteristic parameters obtained from the acoustic emission signal of the steel fiber concrete sample are shown in Table 4.

[0124] Table 4 Partial data of acoustic emission characteristic parameters of sample 3

[0125]

[0126] The other steps are the same as those in Example 1.

[0127] In this embodiment, the results of acoustic emission feature optimization for damage mechanism identification of steel fiber concrete beam specimens are shown in Table 5. Among them, the number of feature parameters selected based on feature importance is N 1 The number of selected feature parameters is 9, and the selected feature parameters are ring count, amplitude, rise time, signal strength, peak frequency, centroid frequency, RA value, AF value and duration. The number of selected feature parameters determined based on feature similarity N 2 The characteristic parameters selected are signal intensity, peak frequency, centroid frequency, RA value and AF value. The characteristic parameters selected based on similarity are the features involved in cluster analysis.

[0128] The parameter optimization results based on Grid Search are shown in Figure 4 , the optimization results in the figure are expressed in coordinate form. The darker the color of the coordinate point, the larger the clustering silhouette coefficient. It can be seen that the clustering silhouette coefficient of sample 3 obtained by the method of the present invention is greater than 0.92, and the clustering effect is good. The selected features obtained by the method are well representative and can provide a reference for the establishment of early warning indicators for steel fiber concrete damage.

[0129] Table 5 Optimization results of acoustic emission characteristics of damage mechanism of sample 3

[0130]

[0131] The cumulative distribution of damage signal time-event of specimen 3 is as follows: Figure 5 As shown in the figure, the damage signals are divided into three clusters, corresponding to matrix cracking, matrix and steel fiber separation, and aggregate cracking, which is consistent with the damage mechanism of steel fiber concrete. Figure 5 In the figure, curve cluster2 represents matrix cracking, curve cluster1 represents separation of matrix and steel fiber, and curve cluster3 represents aggregate cracking.

[0132] In summary, the method of the present invention can use the acoustic emission signal characteristics to achieve good concrete damage mechanism identification, and optimize the number of features by integrating the multi-objective optimization mechanism, thereby improving the clustering effect and computational efficiency. In addition, the present invention has wide applicability and can be applied to the exploration of new concrete damage mechanisms, reducing the workload of manual analysis of acoustic emission characteristics.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for adaptive identification of concrete damage mechanism, characterized in that: The steps include: Step S1: preparing a concrete beam sample and performing a three-point bending loading test, collecting acoustic emission signals from the entire process of the sample until failure; Step S2: acquiring characteristic parameters representing concrete damage information according to the collected acoustic emission signals, and normalizing the first data set corresponding to each characteristic parameter, and using the normalized processing result as a characteristic sample set for cluster analysis; The first data set corresponding to each characteristic parameter includes the value of the characteristic parameter at different times; The extracted characteristic parameters representing the concrete damage information include some or all of the ring count, amplitude, rise time, duration, energy, signal strength, peak frequency, centroid frequency, RA value and average frequency; Step S3: clustering is performed using the Grid Search method combined with the Gaussian mixture model GMM, with the dimension of the feature samples and the value range of the number of clusters as constraints, and the importance index, similarity index of the feature samples and the silhouette coefficient of the clustering results as optimization targets, to establish an adaptive concrete damage mechanism identification model; The specific steps are: Step S31: Establishing a Grid Search global search network model based on the initial number of characteristic parameters obtained from the acoustic emission signal, the number of characteristic parameters after dimensionality reduction based on characteristic importance, the number of characteristic parameters after dimensionality reduction based on characteristic similarity, and the number of clusters as constraints; The constraints are as follows: ; In the above formula, N is the initial number of characteristic parameters obtained from the acoustic emission signal; Step S32: Calculate the Laplace score of each feature parameter and sort them from low to high, retaining the top N 1 feature parameter; then calculate the feature similarity index of the retained feature parameters and sort them from low to high, retaining the top N 2 characteristic parameters; The Laplace score of the feature parameter, that is, the importance of the feature parameter, is calculated as follows: (1) The distance matrix between samples in the sample set of each feature parameter Y is obtained by the K nearest neighbor method S ,in, n is the total number of samples in the sample set of a characteristic parameter; (2) Calculate the diagonal matrix D and the Laplacian matrix L ; L = DS; In the above formula, b , c ∈[1, n ], S bc Represents the distance matrix S Middle b Line c Column data, D bc represents the diagonal matrix D b Line c Column data; (3) Calculate the Laplace score ; In the above formula, f r Indicates r All samples in the sample set with feature parameters Y, r ∈[1,N], express f r The transpose of Step S33: Based on the feature parameters retained after two feature selections in step S32, a cluster number of k The Gaussian mixture model GMM is used for clustering, and the silhouette coefficient is used to evaluate the clustering results; Step S34: Change N 1. N 2. k The value of is repeated, steps S32-S33 are repeated to generate a Pareto optimal solution that meets the objective function of clustering the damage mechanism of the concrete specimen; Step S4: The acoustic signal sensitive characteristics of the concrete sample are clarified according to the characteristic parameter selection results, and the damage mechanism of the concrete sample is calibrated according to the comparative analysis of the clustering results.

2. The method for adaptive identification of concrete damage mechanism according to claim 1, characterized in that: In step S2, the normalization processing formula is: ; In the formula, Y i,norm is any characteristic parameter Y The corresponding first data set i The value after normalization of samples; Y i is the first data set corresponding to the feature parameter i The value of samples before normalization; Y max and Y min are respectively the maximum and minimum values ​​in the first data set corresponding to the characteristic parameter.

3. The method for adaptive identification of concrete damage mechanism according to claim 2, characterized in that: In step S32, the calculation process of the feature similarity index is as follows: (1) Calculate the similarity between two feature parameters. The formula is: ; In the above formula, A and B represent any two of the N1 feature parameters retained after dimensionality reduction based on feature importance. A i The first i samples, B i The first i samples; The larger it is, the greater the similarity between the two feature parameters of A and B; (2) Calculate the similarity index of the features, the formula is: ; In the above formula, Represents characteristic parameters A and any characteristic parameter X The similarities between X Selected and retained based on feature importance N 1 characteristic parameter m feature parameters.

4. The method for adaptive identification of concrete damage mechanism according to claim 3, characterized in that: In step S33, the concrete damage mechanism clustering process based on GMM is as follows: Step S331: The optimized dimensionality reduction based on feature similarity is N The second data set consisting of 2 feature parameters is used as the input of the Gaussian mixture model GMM, and the number of clusters is set to k , then keep the Gaussian distribution unchanged, and calculate the posterior probability of each data point belonging to each Gaussian distribution through the expectation step-maximization step algorithm. Each data point includes N Samples of two feature parameters at the same time; then update the Gaussian distribution parameters according to the posterior probability to maximize the loss function; the details are as follows: (1) Expectation step: Under a given multidimensional Gaussian distribution, calculate any data point according to Bayes’ theorem x i The probability of belonging to the pth Gaussian distribution : ; (2) Maximization step: Update the Gaussian distribution parameters based on the probability obtained in the expectation step: ; ; ; In the above formula, x i For the i data points, For data points x i The expectation of the p-th Gaussian distribution, For data points x i The covariance belonging to the p-th Gaussian distribution, is the weight of the pth Gaussian distribution in the Gaussian mixture model, and satisfies ; is the probability density function of the acoustic emission characteristic sequence that obeys the pth Gaussian distribution, expressed as: ; In the above formula, ; Step S332: Use the logarithmic form of the maximum likelihood function as the loss function, and repeat step S331 until the change in the loss function is less than the threshold or the maximum number of iterations is reached. The loss function is expressed as: 。 5. The method for adaptive identification of concrete damage mechanism according to claim 4, characterized in that: In step S33, the calculation formula of the overall silhouette coefficient S is: ; in s ( i ) is the i The silhouette coefficient of the data points is expressed as: ; In the above formula, a ( i ) is the i The average distance between a data point and other data points in its cluster; b ( i ) is the i The minimum value of the average distance between a data point and other data points in the cluster. The larger the overall silhouette coefficient, the closer the distance between the data points in the cluster, the farther the distance between the data points in the cluster, and the better the clustering effect.

6. The method for adaptive identification of concrete damage mechanism according to claim 1, characterized in that: In step S4, the identification process of the concrete sample damage mechanism includes aligning the clustering results with the event occurrence time, and drawing a time-event accumulation graph, and analyzing the growth characteristics of each damage signal through the graph to identify the damage mechanism of the sample.