Roller compacted concrete interface damage identification method based on acoustic emission technology

Abstract

The invention discloses a roller compacted concrete interface damage identification method based on an acoustic emission technology, which comprises the following steps: (1) preparing a cubic block of roller compacted concrete with a standard size, dividing the cubic block into two layers, and arranging at least two acoustic emission probes on each of two opposite surfaces of the cubic block; (2) applying an acting force to the vertical square block, and receiving a sound wave signal by the sound emission probe; (3) receiving a sound wave signal through the sound emission probe to determine a concrete damage position; (4) the concrete cracking damage type is obtained through the RA-AF value of the acoustic emission probe, wherein RA is the comparison value of the rising time and the amplitude; aF is the ratio of the signal ringing count to the duration. According to the roller compacted concrete interface damage identification method based on the acoustic emission technology, the problem of inaccurate damage positioning in the prior art is effectively solved. The sound emission technology is combined with the damage identification algorithm, the damage position and category can be accurately identified, and a reliable basis is provided for subsequent repair.

Description

TECHNICAL FIELD

[0001] The present application relates to a roller compacted concrete interface damage identification method based on acoustic emission technology, and belongs to the field of acoustic emission. BACKGROUND

[0002] The roller compacted concrete interface forms a weak interlayer due to layered paving and rolling during construction, resulting in weak interlayer bonding. Under load, stress concentration at the interface causes micro-cracks to initiate and expand, reducing overall strength and stiffness. Damage further reduces fracture toughness, making the material more prone to cracking and crack propagation, which affects the integrity and durability of the structure. The cracking failure is a dynamic evolution process from damage to fracture instability. At the initial stage, the crack opening displacement increases linearly with the load, and there is no macroscopic crack on the surface of the specimen. When the load approaches the cracking load, the crack opening displacement appears inflection point, and the macroscopic crack appears on the surface. With the continuous increase of load, the crack opening displacement increases, and the surface crack expands. When the load reaches the maximum load, the crack begins to expand unstably, the width increases, the opening displacement continues to grow, and the number increases. Finally, the specimen is broken. However, the weak interlayer of roller compacted concrete is located inside the concrete, and its damage cannot be directly measured. Therefore, it is of great significance to invent an effective monitoring means and a damage positioning and identification method to solve this problem.

[0003] The existing technology has limitations, mainly focusing on the quantitative characterization of damage, and it is difficult to accurately locate the damage position and identify the damage type. Especially for the damage of the roller compacted concrete interface, the existing acoustic emission technology lacks positioning accuracy and damage type identification ability, and cannot meet the actual needs of accurate identification and positioning of the roller compacted concrete interface damage. In addition, the existing technology is mostly based on a single unsupervised machine learning algorithm to perform cluster analysis on the roller compacted concrete interface damage, mainly based on the distribution characteristics of the data for clustering division, and lacks accuracy evaluation of the classification results. SUMMARY

[0004] The present application provides a roller compacted concrete interface damage identification method based on acoustic emission technology, which effectively solves the problem of inaccurate damage positioning in the prior art. Through acoustic emission technology combined with positioning algorithm, the damage position can be accurately calculated, providing a reliable basis for subsequent repair.

[0005] Technical scheme: To solve the above technical problems, a roller compacted concrete interface damage identification method based on acoustic emission technology provided by the present application comprises the following steps:

[0006] (1) Preparation of a standard size roller compacted concrete cube, divided into two layers, and arranging no less than 2 acoustic emission probes on each of the two opposite surfaces of the cube, and the 4 probes cannot be in the same plane, staggered arrangement;

[0007] (2) Applying force to the cube, and the acoustic emission probes receiving acoustic wave signals;

[0008] (3) Determining the concrete damage location through the acoustic emission probes receiving acoustic wave signals;

[0009] (4) Using the RA-AF value of the acoustic emission probes to obtain the concrete cracking damage type, RA is the rise time to amplitude ratio, and AF is the signal ring count to duration ratio.

[0010] As preferred, the step (3) comprises the following steps:

[0011] (31) Acoustic emission probe AE waveform TOA picking up

[0012] AIC(t w )=t w log(var(R w (t w ,1)))+(T w -t w -1)log(var(R w (1+t w ,T w )))

[0013] In the formula, R w is a time series window containing elastic wave arrival time, T w is the end time of the time window, var is the variance function, and t w represents the time variation value within the time window;

[0014] (32) AE signal TOA accurate picking up, In the formula, t is time, R(u) and R(t) are both time series windows containing elastic wave arrival time; when the singular point u=t, the value is the Cauchy principal value of integral, after Hilbert transformation of a time series, a time series with a phase shift of π / 2 is obtained, and then the envelope time equation is obtained: According to the Hilbert transformation result, the AIC calculation time window is determined, and the AIC value at the time corresponding to the minimum value is the TOA of the AE signal;

[0015] (33) Establishing a TOA function equation group, and proposing a TOA residual equation, when the TOA residual takes the minimum value, the estimated coordinates are the actual spatial position of the AE source.

[0016] As preferred, the step (33) is realized by the following method: assuming that the elastic wave emitted by the AE source propagates along a straight line in the concrete slab, the propagation velocity of the elastic wave in the concrete slab is v p For the AE signal with spatial coordinates (x s ,y s ,z s ), the TOAt s of the elastic wave emitted at time t i to the i-th AE probe (x i ,y i ,z i,c ) is:

[0017]

[0018] In the formula, there are 4 unknowns, respectively, the spatial coordinates (x s ,y s ,z s ) of the AE source and the time ts of the elastic wave emitted by the AE source, and therefore, to realize the spatial positioning of the AE source, an equation group containing at least 4 elastic wave propagation time equations needs to be established, that is, at least 4 AE probes are needed, and the spatial coordinates of the AE source can be solved based on the propagation time equation group established by any 4 AE probes. According to the actually collected AE signal TOAt i,o , the residual equation of the calculated TOAt i,c and the actual TOAt i,o is established:

[0019] t i = |t i,o -t i,c |

[0020] According to the residual equation, the AE source positioning problem is converted into the minimization problem of the residual equation:

[0021]

[0022] In the formula, N s is the number of AE probes.

[0023] As preferred, the step (4) comprises the following steps:

[0024] (41) Obtain the corresponding relationship between the input data and the output mode by the Kmeans clustering analysis method, divide the data into different category subsets, and form a pseudo-labeled data set Y;

[0025] (42) Introduce the SVM classification algorithm again, train and test the clustered data, and thus obtain a classification model suitable for the damage mode recognition of roller compacted concrete.

[0026] As a preference, said step (41) comprises the following steps:

[0027] The principle of Kmeans clustering is to minimize the sum of distances between data points X and the center points of the clusters to which they belong, i.e. the following objective function:

[0028]

[0029] An unlabeled acoustic emission data set X = {x(1), x(2), …, x(n)} is selected as the input layer, where n is the number of acoustic emission events, the acoustic emission signal received by the sensor, and the features of the acoustic signal are extracted, one of which is the number of acoustic emission samples. Each x(i) in the data set is a two-dimensional array [RA i , AF i ], RA i is the rise time to amplitude ratio of each probe, and AF i is the ratio of signal ringing count to duration of each probe. The acoustic emission probe collects acoustic signal features and obtains their characteristic parameters, and calculates their corresponding ratios. The original data is normalized and preprocessed using the following formula before clustering: where RA nori and AF nori represent the normalized RA and AF values, respectively. The processed unlabeled AE data set is divided into two classes using the Kmeans algorithm, namely the stretching cluster and the shearing cluster. After Kmeans clustering, the original data set X becomes a labeled data set Y.

[0030] As a preference, said step (42) comprises the following steps: The normalized linear function of the hyperplane is H·V+b=0, where H is a vector on the hyperplane, V is the normal vector of the hyperplane, and b is the threshold value of the support vector machine. The boundary of the hyperplane is y sv (x sv ·V+b) = 1, where x sv ∈X is a vector on the boundary of the hyperplane, specifically a data point on the boundary, called a support vector, y sv ∈Y is the label of vector x sv , is the perpendicular distance between the hyperplane and the origin, and ||V|| is the Euclidean norm of vector v. Since the distance between the hyperplane boundary and the optimal hyperplane is then the solution of the formula is equal to the following constraint condition:

[0031]

[0032] where xi is a vector of X, corresponding to the coordinate vector of the i-th data point, and yi is a vector of Y, corresponding to the classification label of the i-th data point,

[0033] In order to avoid the non-convergence of the hyperplane solution, a non-negative relaxation coefficient ζ is introduced i And the penalty coefficient C, then the formula is changed to:

[0034]

[0035] The Lagrange functional expression of the formula is:

[0036]

[0037] Where, δ i And μ i is the Lagrange multiplier, the constraint condition should satisfy When this condition is met, the following equation is obtained;

[0038]

[0039] μ i =C-δ i

[0040] Substitute the above equation into the formula, then the minimization problem can be changed to:

[0041]

[0042] The sequential minimal optimization algorithm is used to solve the quadratic programming optimization problem, and the equation has an optimal solution when the following conditions are met:

[0043]

[0044] Through the minimum optimization algorithm, the normal vector V of the hyperplane and the threshold b can be finally solved (after obtaining the values of these two parameters, the normal vector V and the threshold b are obtained, that is, the hyperplane equation H·V+b=0 is determined, the classification boundary on the coordinate axis is obtained, high RA and low AF represent one kind of damage, low RA and high AF are another kind of damage, and it can be further divided into medium RA and AF, which is another kind of damage, two or three kinds of hyperplane boundaries and data points on the hyperplane are obtained, which can well classify the damage type.

[0045] The classification of the step (41) specifically includes the following steps: the Kmeans algorithm classifies the damage into two categories, three categories, and the following is only an example of two kinds of damage in the commonly used RA-AF analysis. After Kmeans clustering of the damage, the clustering boundary is not clear, and it lacks quantitative and interpretable display, therefore, a clear and visible classification boundary is established by introducing the SVM algorithm, and a clearer trend area is divided by dividing the training set and the test set, and a more interpretable classification result is provided, the specific operation steps are as follows:

[0046] (411)Randomly select 2 data points as the center points of the initial 2 clusters; in general, high RA and low AF represent shear failure, and low RA and high AF represent tensile failure, which are usually divided into two types of damage, and the classification is divided by machine learning, and here the two common types of damage are clustered and analyzed;

[0047] (412)Calculate the distance of each data point in the data set to the center of the 2 clusters, and attribute it to the class corresponding to the cluster center with the smallest distance, and the Euclidean distance d is used to represent the distance between the ith data point [RA i , AF i ] and the cluster center point, and the calculation formula is as follows;

[0048]

[0049] (413)According to the existing data points in each cluster, recalculate to update the cluster center position N(k) according to the formula;

[0050]

[0051] Where s represents the number of data points contained in the kth cluster, i.e. the cluster center position is updated to the mean value of all data points in the cluster;

[0052] (414)Repeat the two-step operations of (412) and (413) until the cluster center is stable and no longer changes or reaches the preset number of iterations, and consider that the Kmeans clustering is completed at this time, and the original data set X becomes a data set Y with labels.

[0053] The Kmeans algorithm outputs the data set Y with classification labels, which roughly classifies the categories, and then determines the more accurate classification boundary through SVM, and uses the cross-validation method to train, test and verify the labeled data after clustering.

[0054] For example, the labeled AE signal data obtained by Kmeans clustering and its label data are combined into a database, 40% of the data in the database are randomly selected as training data for building an SVM model, and the model is tested, 20 groups of data are randomly selected from the database for model accuracy testing, and each group of random data is 20% of the database. The prediction accuracy is defined as the ratio of the number of correctly predicted data to the number of test data. The test results of the 20 groups of data are shown in Figure 2 and Figure 3 As shown, the accuracy of all groups is higher than 99.9%, indicating that the established SVM classification model can effectively separate the two types of signals.

[0055] At present, the damage type classification of RA-AF is generally artificially divided threshold, and the method is combined with unsupervised and supervised machine learning to divide the threshold category, and the RA-AF value is needed when GMM is used for cluster analysis.

[0056]

[0057] In the formula, x is a sample vector, that is, x=(RA, AF);M is a Gaussian distribution mixing number, omega k is the weight coefficient of the kth Gaussian distribution, N(x|u k ,∑ k ) is the kth Gaussian distribution, u k is the mean of the Gaussian distribution, and ∑ k is the covariance matrix.

[0058] The maximum likelihood function is used to estimate the GMM model parameters lambda=(omega k , mu k ,∑ k ), that is, the best parameters lambda are estimated to maximize the likelihood estimate value of the probability density function. The likelihood estimate value of the GMM probability density function is obtained by the following formula:

[0059]

[0060] When the data sample size is large, the probability of a single data point is very small, and after multiplication, a minimum value is obtained. Therefore, the above formula is usually in logarithmic form:

[0061]

[0062] When the GMM model takes the best parameters, the maximum likelihood estimate value of the above formula takes the maximum value. Therefore, the expectation maximization (Expectation Maximization, EM) algorithm can be used to estimate the best parameters of the GMM, so as to divide the categories.

[0063] Beneficial effects: the roller compacted concrete interface damage identification method based on acoustic emission technology has the following advantages:

[0064] The present application effectively solves the problem of inaccurate damage positioning in the prior art. Through the acoustic emission technology combined with the positioning algorithm, the damage position can be accurately calculated, which provides a reliable basis for subsequent repair. Combined with unsupervised and supervised learning algorithm, the Kmeans-SVM algorithm is used for cluster analysis of acoustic emission signal characteristics, which can effectively distinguish tensile damage and shear damage, and quantitatively evaluate the clustering results, improve the identification accuracy. The patent is suitable for monitoring the damage of roller compacted concrete structure in practical engineering, provides a powerful guarantee for engineering safety, and has a broad application prospect. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 Flowchart of the present application.

[0066] Figure 2 Allocation of model test process data based on AE data in the embodiment.

[0067] Figure 3 Test accuracy effect of the model test process based on AE data in the embodiment.

[0068] Figure 4 RA-AF clustering result graph of the style in the splitting experiment by Kmeans-SVM clustering analysis in the embodiment.

[0069] Figure 5 Damage diagram of the embodiment.

[0070] Figure 6 Kmeans-SVM clustering analysis process diagram of the present application. DETAILED DESCRIPTION

[0071] The present application will be further described below in conjunction with the drawings.

[0072] As Figures 1 to 6 shown, a roller compacted concrete interface damage identification method based on acoustic emission technology, the specific steps are as follows:

[0073] 1. Prepare a standard size roller compacted concrete cube, divide it into two layers, and ensure that the interlayer interface is clear. Arrange 4 acoustic emission probes on the surface of the cube, and optimize the probe spacing according to specific requirements. Apply a splitting load to the roller compacted concrete cube to simulate interlayer interface damage.

[0074] 2. Acoustic emission (AE) technology is a non-destructive testing method that mainly monitors the damage and deformation process of materials or structures by capturing elastic waves generated inside them. Start the acoustic emission monitoring system and record the acoustic emission signals in real time. Use acoustic emission positioning algorithms to calculate the damage location based on the acoustic emission signals captured by the sensors. The specific steps are as follows:

[0075] 2.1. Determine the TOA (AE signal arrival time at the probe) of the AE waveform propagating to the AE probe:

[0076] 2.1.1. AE waveform TOA picking

[0077] AIC(t w )=t w log(var(R w (t w ,1)))+(Tw -t w -1)log(var(R w (1+t w ,T w ))

[0078] where R w is a time series window containing the arrival time of elastic wave, T w is the end time of the time window, var is the variance function, t w varies within the time window, in the formula, it is used to describe different time within the time window, AIC value is calculated by traversing t w to determine the arrival time (TOA) of AE signal, t w =1, 2, 3…T w , R w is a discrete digital signal sequence, t w is an integer variable that varies within the time window Rw, which is the serial number of sampling points. t w is equivalent to the serial number, which is dimensionless;

[0079] 2.1.2, accurate picking of AE signal TOA

[0080]

[0081] where t is time; when the singularity u=t, the value is the Cauchy principal value of the integral. After Hilbert transformation of a time series, a time series with a phase shift of π / 2 is obtained, and then the envelope time equation is obtained:

[0082]

[0083] According to the Hilbert transformation result, the AIC calculation time window is determined, and the TOA of AE signal is the time corresponding to the minimum AIC value.

[0084] 2.2, establish TOA function equation set, and propose TOA residual equation, when the TOA residual takes the minimum value, the coordinates are estimated as the actual spatial position of AE source.

[0085] 2.2.1, AE signal positioning

[0086] Assume that the elastic wave emitted by the AE source propagates along a straight line in the concrete wing plate, and the propagation speed of the elastic wave in the concrete wing plate is v p . For the AE signal with spatial coordinates (x s , y s , z s ), the elastic wave emitted at t s propagates to the i-th AE probe (x i, y i , z i ) of TOAt i,c is:

[0087]

[0088] In the formula, there are 4 unknowns, which are the spatial coordinates (x s , y s , z s ) of the AE source and the time ts when the elastic wave is emitted by the AE source. Therefore, in order to realize the spatial positioning of the AE source, an equation group containing at least 4 elastic wave propagation time equations needs to be established, that is, at least 4 AE probes are needed. According to the actually collected AE signal TOAt i,o , the residual equation for calculating TOAt i,c and the actual TOAt i,o is established:

[0089] ri = |t i，o -t i，c | According to the residual equation, the AE source positioning problem is converted into the minimization problem of the residual equation:

[0090]

[0091] In the formula, N s is the number of AE probes, which is 4.

[0092] 3、RA-AF (RA is the rise time to amplitude ratio; AF is the ratio of signal ring count to duration) is commonly used to distinguish the damage type of concrete cracking. First, the Kmeans clustering analysis method is used to obtain the corresponding relationship between the input data and the output mode, and the data is divided into different category subsets to form a pseudo-labeled data set. Then the SVM classification algorithm is introduced to train and test the clustered data, so as to obtain a classification model suitable for the damage mode recognition of roller compacted concrete. The specific steps are as follows:

[0093] The Kmeans clustering principle is to minimize the sum of distances between data points X and their respective cluster center points, that is, the following objective function:

[0094]

[0095] The unlabeled AE dataset X = {x(l), x(2),..., x(n)} is selected as the input layer. Wherein, n is the number of AE events, and x(i) in the dataset is a two-dimensional array of [RAi, AFi]. The original RA value and AF value have obvious differences in the numerical scale, which will reduce the reliability of the clustering results, and sometimes even lead to the non-convergence of the clustering results, while the normalization of the parameters can avoid this phenomenon. In this study, the original data is normalized and pretreated before clustering using the following formula:

[0096]

[0097] Wherein, RA nori and AF nori represent the normalized RA value and AF value, respectively. The processed unlabeled AE dataset is divided into two classes, namely the stretching cluster and the shearing cluster, using the Kmeans algorithm. The specific operation steps are as follows:

[0098] (1) Randomly select 2 data points as the initial center points of the 2 clusters;

[0099] (2) Calculate the distance of each data point in the dataset to the 2 cluster centers, and attribute it to the class corresponding to the cluster center with the smallest distance. In this study, the Euclidean distance d is used to represent the distance between the i-th data point [RA i , AF i ] and the cluster center point, and the calculation formula is as follows;

[0100]

[0101] (3) According to the existing data points in each cluster, recalculate N(k) to update the cluster center position according to the formula;

[0102]

[0103] Wherein, n represents the number of data points contained in the k-th cluster. That is, the cluster center position is updated to the mean value of all data points in the cluster.

[0104] (4) Repeat the iteration of steps (2) and (3) until the clustering center is stable and no longer changes or the preset iteration number is reached. After completing the Kmeans clustering, the original dataset X becomes a labeled dataset Y. Then the quantitative separation of the two types of labeled data needs to be realized. The optimal separation hyperplane is obtained using the SVM algorithm, so as to effectively separate the sample points of the two different labeled categories in the dataset Y, and maximize the interval between the two categories. The normalized linear function of the hyperplane is: H·V+b=0. Wherein, H is the vector on the hyperplane, V is the normal vector of the hyperplane, and b is the threshold value of the support vector machine. The boundary of the hyperplane is:

[0105] y hb ·(x hb V + b) = 1

[0106] Among them, x hb ∈x is a vector on the boundary of the hyperplane, called a support vector, and y hb ∈y is the vector x hb . is the vertical distance between the hyperplane and the origin, and ||V|| is the Euclidean norm of vector v. Since the distance between the hyperplane boundary and the optimal hyperplane is Then the solution of the formula is equal to the following constraints:

[0107]

[0108] In order to avoid the non-convergence of the hyperplane solution, a non-negative relaxation coefficient, ζ, is introduced. -i ., then the formula becomes:

[0109]

[0110] Where C is a non-negative penalty coefficient. The Lagrangian functional expression of the formula is:

[0111]

[0112] Among them, δ i and μ i is the Lagrange multiplier. The constraints should satisfy When this condition is met, the following equation is obtained.

[0113]

[0114] μ i =C-δ i

[0115] Substituting the above equation into the formula, the minimization problem can be transformed into:

[0116]

[0117] The sequential minimum optimization algorithm is used to solve the quadratic programming optimization problem. The equation has an optimal solution when the following conditions are met:

[0118]

[0119] This algorithm transforms a large quadratic programming problem into a series of possible minimum quadratic programming problems, requiring only solving two randomly selected Lagrange multipliers for optimization. The minimum optimization algorithm ultimately solves the hyperplane normal vector V and threshold b.

[0120] 4. Taking the embodiment as an example, the labeled AE signal data obtained by Kmeans clustering and its label data set are combined into a database. 40% of the data in the database are randomly selected as training data for SVM model construction, and the established model is tested. 20 groups of data are randomly selected from the database for model accuracy test, and the size of each group of random data is 20% of the database. The prediction accuracy is defined as the ratio of the number of correctly predicted data in each group to the number of test data. The test results of 20 groups of data are as follows: Figure 2 and Figure 3 As shown in the figure, the accuracy of all groups is higher than 99.9%, indicating that the established SVM classification model can achieve effective separation of the two types of signals.

[0121] Figure 4 and Figure 5 The RA-AF clustering and damage location results of the sample in the embodiment in the splitting tensile test after SVM analysis are shown in the figure. It can be seen from the figure that the damage position and type are accurately located and divided.

[0122] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for identifying interface damage of roller compacted concrete based on acoustic emission technology, characterized in that: The following steps are involved: (1) Prepare a standard size RCC cube, divided into two layers, and arrange at least two acoustic emission probes on each of the two opposite surfaces of the cube. The four probes should not be in the same plane and should be arranged in a staggered manner. (2) Apply force to the cube, and the acoustic emission probe receives the sound wave signal; (3) Determine the location of concrete damage by receiving the acoustic wave signal through the acoustic emission probe; (4) The RA-AF value of the acoustic emission probe is used to determine the type of concrete cracking damage, where RA is the ratio of rise time to amplitude; AF is the ratio of signal ring count to duration.

2. The method for identifying RCC interface damage based on acoustic emission technology according to claim 1 is characterized in that: The step (3) comprises the following steps: (31) Picking up the AE waveform TOA of the acoustic emission probe AIC(t w )=t w log(var(R w (t w ,1)))+(T w -t w -1)log(var(R w (1+t w ,T w ))) Where R w is the time series window containing the arrival time of elastic waves, T w is the end time of the time window, var is the variance function, t w Indicates the time change value within the time window, t w =1, 2, 3…T w ; (32) Accurately pick up the AE signal TOA, Where t is time, R(u) and R(t) are both time series windows containing the arrival time of the elastic wave. When the singular point u = t, the value is the Cauchy principal value of the integral. After performing the Hilbert transform on a time series, a time series with a phase shift of π / 2 is obtained, and then the envelope time equation is obtained: The AIC calculation time window is determined based on the Hilbert change result, and the time corresponding to the minimum AIC value is the TOA of the AE signal; (33) A TOA function equation group is established, and a TOA residual equation is proposed. When the TOA residual takes the minimum value, the estimated coordinate is the actual spatial position of the AE source.

3. The method for identifying RCC interface damage based on acoustic emission technology according to claim 1 is characterized in that: The step (33) is implemented by the following method: Assume that the elastic wave emitted by the AE source propagates in a straight line in the concrete wing plate, and the propagation speed of the elastic wave in the concrete wing plate is v p , for the space coordinate (x s ,y s ,z s ) of the AE signal, at t s The elastic wave emitted at time propagates to the i-th AE probe (x i ,y i ,z i ) of TOAt i,c for: t i,c is the calculation time when the i-th acoustic emission probe receives the elastic wave. In the formula, there are 4 unknowns, which are the spatial coordinates of the AE source (x s ,y s ,z s ) and the time t at which the AE source emits the elastic wave s , the spatial coordinates of the AE source can be solved by establishing a propagation time equation group based on any four AE probes, and the AE signal TOAt actually collected in the experiment i,o , establish calculation TOAt i,c and actual TOAt i,o The residual equation is: r i =∣t i,o -t i,c ∣ r i is the TOA residual of the i-th probe, t i,o is the actual time when the i-th probe receives the elastic wave. According to the residual equation, the AE source positioning problem is transformed into a minimization problem of the residual equation: Where N s is the number of AE probes.

4. The method for identifying RCC interface damage based on acoustic emission technology according to claim 1 is characterized in that: The step (4) comprises the following steps: (41) The correspondence between input data and output patterns is obtained through the Kmeans clustering analysis method, and the data is divided into different category subsets to form a pseudo-label dataset Y; (42) The SVM classification algorithm is then introduced to train and test the clustered data, thereby obtaining a classification model suitable for RCC damage pattern recognition.

5. The method for identifying RCC interface damage based on acoustic emission technology according to claim 4 is characterized in that: The step (41) comprises the following steps: The Kmeans clustering principle is to minimize the distance between the data point X and the center point μ of each cluster. (k) The sum of the distances between , that is, the following objective function: X: Unlabeled acoustic emission dataset, X = {x(1), x(2), ..., x(n)}, x(i): Feature vector of the i-th acoustic emission event, x(i) = [RA i ,AF i ], n: total number of acoustic emission events, μ (k) : The coordinates of the center point of the kth cluster, k is the cluster index, and the unlabeled acoustic emission dataset X = {x(1), x(2)…, x(n)} is selected as the input layer, μ (k) is the coordinate of the center point of each classification cluster, where n is the number of acoustic emission events, and x(i) in the data set is [RA i , AF i ]Two-dimensional array, RA i AF is the comparison value of the rise time and amplitude of each probe, i The ratio of the ring count to the duration of each probe signal is obtained. The acoustic emission probe collects the acoustic signal characteristics, obtains its characteristic parameters, and calculates its corresponding ratio. Before clustering, the raw data is normalized and preprocessed using the following formula: Among them, RA nori and AF nori The RA value and AF value after normalization are characterized respectively. The Kmeans algorithm is used to divide the processed unlabeled AE dataset into two categories, namely stretching clusters and cutting clusters. After completing Kmeans clustering, the original dataset X becomes a labeled dataset Y.

6. The method for identifying RCC interface damage based on acoustic emission technology according to claim 4 is characterized in that: The step (42) comprises the following steps: the normalized linear function of the hyperplane is: H·V+b=0, wherein H is a vector on the hyperplane, V is a normal vector of the hyperplane, b is a threshold of the support vector machine, and the boundary of the hyperplane is: y sv (x sv V + b) = 1, where x sv ∈X is a vector on the boundary of the hyperplane, specifically a data point on the boundary, called a support vector, y sv ∈Y is the vector x sv Tags, is the vertical distance between the hyperplane and the origin, ||V|| is the Euclidean norm of vector v, and since the distance between the hyperplane boundary and the optimal hyperplane is Then the solution of the formula is equal to the following constraints: Where x i Is a vector of X, corresponding to the coordinate vector of the first data point, y i It is a vector of Y, corresponding to the classification label of the nth data point; In order to avoid the non-convergence of the hyperplane solution, a non-negative relaxation coefficient ζ- i and penalty coefficient C, the formula is changed to: The Lagrange functional expression of the formula is: C: penalty coefficient, used to balance classification error and maximize interval; ζ i : non-negative relaxation coefficient, allowing a small number of data points to be misclassified; y i : The label of the i-th data point, y i ∈{+1,-1}; where δ i and μ i is a Lagrange multiplier, and the constraints should satisfy When this condition is met, the following equation is obtained; Substituting the above equation into the formula, the minimization problem can be transformed into: The sequential minimum optimization algorithm is used to solve the quadratic programming optimization problem. The equation has an optimal solution when the following conditions are met: The minimum optimization algorithm can ultimately solve the normal vector V and threshold b of the hyperplane, determine the hyperplane equation H·V+b=0, and obtain the classification boundary on the coordinate axis. The minimum optimization algorithm solves the normal vector V and threshold b of the hyperplane, which are used to determine the hyperplane equation H·V+b=0, thereby forming a classification boundary on the coordinate plane. High RA and low AF represent shear damage, while low RA and high AF represent tensile damage.

Images