A method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment

Abstract

The present invention relates to an acoustic emission signal feature extraction method based on active defect monitoring of pressure-bearing equipment. The method extracts acoustic emission waveform feature parameters based on the actual parameter features of the acoustic emission signal for active defect monitoring of pressure-bearing equipment; uses the extracted feature parameters as feature vectors, and defines the distance between variables using similarity coefficients; performs hierarchical clustering, classifies the parameters one by one according to the similarity coefficients between the parameters, and obtains an acoustic emission feature parameter correlation hierarchical clustering tree; sets a correlation threshold, and retains parameters with correlation coefficients below the correlation threshold as feature vectors for cluster analysis; and performs K-means cluster analysis on the selected vectors based on the determination of the cluster analysis feature vectors. The present invention reduces the effective signals that are mistakenly deleted, and performs effective identification, thereby improving the accuracy of the prediction results of the acoustic emission signal analysis for active defect monitoring of pressure-bearing equipment.

Description

Technical Field

[0001] The invention relates to an acoustic emission signal feature extraction method based on hierarchical clustering and K-means algorithm, and in particular to an acoustic emission signal feature extraction method based on active defect monitoring of pressure-bearing equipment. Background Art

[0002] Acoustic emission signal is an elastic stress wave formed by the sudden release of strain energy from the inside of the material. It contains information about the internal damage of the material. Therefore, acoustic emission technology can be used to identify different types of damage occurring in the loaded material. In the process of online acoustic emission detection of normal pressure vertical storage tanks, due to the presence of a certain amount of interference caused by environmental and electromagnetic noise in the original acoustic emission data, the traditional clustering method is usually used to analyze and process the original data of the tank bottom plate, remove a large number of interference signals in the original data, and compare the data files before and after processing by acoustic emission correlation analysis to obtain new tank detection data. However, in the extraction of acoustic emission waveform characteristic parameters, although the amplitude and energy can well reflect the intensity characteristics of the acoustic emission signal, the duration and count cannot reflect the shape characteristics of the acoustic emission waveform; and the usual signal processing method will inevitably delete the valid signal by mistake during the process of signal noise reduction, resulting in inaccurate signal analysis. Therefore, in the judgment of damage categories, more advanced signal analysis technology is needed to identify different acoustic emission sources. Summary of the invention

[0003] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a method for extracting features of acoustic emission signals based on active defect monitoring of pressure-bearing equipment to perform effective identification, thereby improving the accuracy of analysis and prediction results of acoustic emission signals for active defect monitoring of pressure-bearing equipment.

[0004] The object of the present invention is achieved in that:

[0005] A preparation method for an acoustic emission signal feature extraction method based on active defect monitoring of pressure-bearing equipment includes the following contents:

[0006] S1: Monitor the actual parameter characteristics of acoustic emission signals for active defects in pressure-bearing equipment and extract characteristic parameters of acoustic emission waveforms;

[0007] S2: The extracted feature parameters are used as feature vectors, and the distance between variables is defined using the similarity coefficient;

[0008] S3: Perform hierarchical clustering, classify the parameters into one category one by one according to the similarity coefficients between the parameters, and obtain a hierarchical clustering tree of the correlation of acoustic emission characteristic parameters;

[0009] S4: Set a correlation threshold and retain the parameters with correlation coefficients below the correlation threshold as the feature vectors for cluster analysis;

[0010] S5: Based on the determination of the cluster analysis feature vector, a K-means cluster analysis is performed on the selected vector.

[0011] Furthermore, the extraction of acoustic emission waveform characteristic parameters in step 1 of the above scheme mainly includes:

[0012] a. Select amplitude and energy to reflect the waveform intensity characteristics of the acoustic emission signal;

[0013] b. Determine A% of the maximum amplitude as the soft threshold value, and recalculate the characteristic parameters of the waveform so that the obtained characteristic parameters reflect the shape characteristics of the acoustic emission waveform;

[0014] c. The margin factor is used to reflect the shape characteristics of the acoustic emission waveform. The margin factor expression is shown in Formula 1:

[0015]

[0016] Where N represents the number of data points in the analysis sample, x n represents the time history of the sample signal, and T represents the time of all samples;

[0017] d. Through wavelet decomposition, the wavelet characteristic energy spectrum coefficient vectors of the acoustic emission signal are calculated in batches according to the wavelet characteristic energy spectrum coefficient definition to reflect the frequency distribution characteristics of the waveform.

[0018] Furthermore, the similarity coefficient in step 2 of the above scheme is defined as follows:

[0019] Let C ij For X i and X j The similarity coefficient between them has the following restrictions: |C ij |≤1, for all i,j; C ij =C ji , for all i, j; the absolute value of the similarity coefficient |C ij The closer to 1, the i and X j The closer, C ij The closer it is to 0, the more distant the relationship between the two.

[0020] For quantitative variables, the similarity coefficient used is X i and X j Correlation coefficient of variable X i and X j The correlation coefficient is usually r ij Indicates, here denoted as C ij ,Right now

[0021]

[0022] When C ij =1 indicates that the two variables are related. Generally, |C ij |≤1;

[0023] The similarity coefficient is used to define the distance between variables, that is,

[0024] d ij =1-|C ij |

[0025] Furthermore, in step 3 of the above scheme, the steps of hierarchical clustering are as follows:

[0026] SS3.1. Calculate the distance between n parameters and obtain the distance matrix D between samples. (0) ;

[0027] SS3.2, Initial (first step: i = 1) n samples each constitute a class, the number of classes k = n, the i-th class G i ={X (i)}(i=1,...,n); in this case, the distance between classes is the distance between samples (i.e., D (1) =D (0) ); then for sample X (i) (i=1,...,n) execute steps SS3.3 and SS3.4 of the merging process;

[0028] SS3.3, the distance matrix D obtained in step S2 (i-1) , merge the two classes with the smallest inter-class distance into a new class. At this time, the total number of classes k is reduced by 1, that is, k = n-i+1;

[0029] SS3.4. Calculate the distance between the new class and other classes to get a new distance matrix D (i) ; If the total number of classes after merging is still greater than 1, repeat steps SS3.3 and SS3.4 until the total number of classes is 1, then go to step (5);

[0030] SS3.5, draw pedigree cluster diagrams;

[0031] SS3.6. Determine the number of categories and the members of each category.

[0032] Furthermore, in step 5 of the above scheme, the steps of K-means clustering are as follows:

[0033] SS5.1. Specify the distance between samples. Artificially set three numbers: k (number of categories), C (minimum distance between categories) and R (maximum distance within categories); take the first k sample points as the condensation points;

[0034] SS5.2. Calculate the distances between the k condensation points. If the smallest distance is less than C, merge the corresponding two condensation points, use the centroid of the two points as the new condensation point, and repeat step (2) until the distances between all condensation points are greater than C.

[0035] SS5.3. Classify the remaining nk samples one by one. For each sample, calculate the distance between the sample and all the clustering points. If the minimum distance is greater than R, the sample is used as a new clustering point. If the minimum distance is less than or equal to R, the sample is classified into the class where the clustering point closest to it is located. Then recalculate the centroid of the previous class and use the centroid as the new clustering point. If the distances between clustering points are all greater than or equal to C, consider the next sample. Otherwise, use step (2) to merge and then consider the next sample, until all samples are classified.

[0036] SS5.4, classify the samples one by one according to step (3) from the beginning to the end, except that: after a sample is classified, if the classification is consistent with the original, the centroid does not need to be calculated; if the classification is different from the original, the centroids of the two categories involved must be recalculated;

[0037] If the new classification is the same as the previous one, the clustering process ends, otherwise repeat step SS5.4;

[0038] The optimal number of clusters k in k-means clustering is determined by the Dacies&Bouldin criterion, which is defined as shown in Formula 3:

[0039]

[0040] where d i and d j are the average distances within class i and class j, respectively, and D ij Represents the distance between class i and class j. When DB reaches the minimum value, the number of clusters is optimal.

[0041] The new tank inspection data obtained by the above method is used to infer the nature, approximate area and severity of the acoustic emission source, and then evaluate the monitoring status of active defects in pressure equipment and guide maintenance decisions.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] (1) The extraction method of the present invention determines the margin factor of the acoustic emission signal based on the waveform characteristics of the acoustic emission source, and at the same time determines A% of the maximum amplitude of the waveform as a soft threshold to recalculate the characteristic parameters of the waveform, so that the obtained characteristic parameters can better reflect the shape characteristics of the acoustic emission waveform, and the parameter space is easy to classify.

[0044] (2) The present invention uses a hierarchical clustering algorithm and a K-means algorithm to process acoustic emission data for monitoring active defects of pressure-bearing equipment, thereby reducing the number of valid signals that are mistakenly deleted, effectively classifying multiple different signals, and effectively identifying them, thereby improving the accuracy of the analysis and prediction results of acoustic emission signals for monitoring active defects of pressure-bearing equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0046] In order to better understand the technical solution of the present invention, the following will be described in detail in conjunction with the relevant illustrations. It should be understood that the following specific embodiments are not intended to limit the specific implementation of the technical solution of the present invention, and are only implementations that can be adopted by the technical solution of the present invention. It should be noted that the description of the positional relationship of each component in this article, such as component A being located above component B, is based on the description of the relative positions of the components in the illustration, and is not intended to limit the actual positional relationship of the components.

[0047] Embodiment 1:

[0048] See also Figure 1 , Figure 1 A flow chart of a method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment is drawn. As shown in the figure, a method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment of the present invention includes the following contents:

[0049] Step 1: Feature parameter extraction:

[0050] Through wavelet decomposition, the wavelet characteristic energy spectrum coefficient vectors of the acoustic emission signal are calculated in batches according to the wavelet characteristic energy spectrum coefficient definition to reflect the frequency distribution characteristics of the waveform;

[0051] Amplitude and energy are selected to reflect the waveform intensity characteristics of the acoustic emission signal;

[0052] Determine A% of the maximum amplitude as the soft threshold value to recalculate the characteristic parameters of the waveform so that the obtained characteristic parameters reflect the shape characteristics of the acoustic emission waveform;

[0053] The margin factor is used to reflect the shape characteristics of the acoustic emission waveform. The margin factor expression is shown in Formula 1:

[0054]

[0055] Where N represents the number of data points in the analysis sample, x n represents the time history of the sample signal, and T represents the time of all samples.

[0056] Step 2: Define the distance between variables:

[0057] Using the parameters and vectors as feature vectors to describe the acoustic emission signal,

[0058] Let C ij For X i and X j The similarity coefficient between them has the following restrictions: |C ij |≤1, for all i,j; C ij =C ji , for all i, j; the absolute value of the similarity coefficient |C ij The closer to 1, the i and X j The closer, C ij The closer it is to 0, the more distant the relationship between the two.

[0059] For quantitative variables, the similarity coefficient used is X i and X j The correlation coefficient is the cosine of the angle after the data is standardized; the variable X i and X j The correlation coefficient is usually r ij Here we denote it as C ij ,Right now

[0060]

[0061] When C ij =1 indicates that the two variables are related. Generally, |C ij |≤1.

[0062] The similarity coefficient is used to define the distance between variables, that is,

[0063] d ij =1-|C ij |or

[0064] Step 3: Hierarchical clustering:

[0065] Too many redundant parameters will cause the originally different categories to become similar. Hierarchical clustering is used to analyze the similarity between the parameters of acoustic emission. It is a method to reduce the number of categories. Therefore, hierarchical clustering is used. The steps of hierarchical clustering are as follows:

[0066] (1) Calculate the distance between n parameters and obtain the distance matrix D between samples (0) .

[0067] (2) Initially (first step: i = 1) n samples each constitute a class, the number of classes is k = n, and the i-th class G i ={X (i)}(i=1,...,n). At this time, the distance between classes is the distance between samples (i.e., D (1) =D (0) ). Then for sample X (i) (i=1,...,n) performs steps (3) and (4) of the classifying process.

[0068] (3) The distance matrix D obtained in step (2) (i＝1) , merge the two classes with the smallest inter-class distance into a new class. At this time, the total number of classes k is reduced by 1, that is, k = n-i+1.

[0069] (4) Calculate the distance between the new class and other classes to obtain a new distance matrix D (i) If the total number of classes after merging is still greater than 1, repeat steps (3) and (4) until the total number of classes is 1, then go to step (5).

[0070] (5) Draw a phylogenetic clustering diagram.

[0071] (6) Determine the number of categories and the members of each category.

[0072] By using hierarchical clustering, the parameters are classified into one category one by one according to the similarity coefficients between the parameters, and a hierarchical clustering tree of the correlation of acoustic emission characteristic parameters is obtained.

[0073] Step 4: Set relevant thresholds:

[0074] The parameters calculated above are clustered hierarchically to obtain the acoustic emission characteristic parameter correlation hierarchical clustering tree. A correlation threshold is set, and the parameters below the correlation coefficient are retained as the characteristic vectors of cluster analysis.

[0075] Step 5: K-means cluster analysis:

[0076] The similarity of each parameter of acoustic emission data is analyzed by hierarchical clustering, and the selected feature vector is used as the parameter of K-means clustering for cluster analysis. The steps are as follows:

[0077] (1) Specify the distance between samples. Artificially set three numbers: k (number of categories), C (minimum distance between categories), and R (maximum distance within categories); take the first k sample points as the clustering points.

[0078] (2) Calculate the distances between the k condensation points. If the smallest distance is less than C, merge the corresponding two condensation points and use the centroid of the two points as the new condensation point. Repeat step (2) until the distances between all condensation points are greater than C.

[0079] (3) Classify the remaining nk samples one by one. For each sample, calculate the distance between the sample and all the clustering points. If the minimum distance is greater than R, the sample is used as a new clustering point. If the minimum distance is less than or equal to R, the sample is classified into the class where the clustering point closest to it is located. Then recalculate the centroid of the previous class and use the centroid as the new clustering point. If the distances between clustering points are all greater than or equal to C, consider the next sample. Otherwise, use step (2) to merge and then consider the next sample, until all samples are classified.

[0080] (4) Classify the samples one by one according to step (3) from beginning to end. The difference is: after a sample is classified, if the classification is consistent with the original one, the centroid does not need to be calculated; if the classification is different from the original one, the centroids of the two categories involved must be recalculated.

[0081] If the new classification is the same as the previous one, the clustering process ends, otherwise repeat step (4).

[0082] When performing k-means clustering, the optimal number of clusters for cluster analysis is determined by the Dacies & Bouldin criterion (a clustering algorithm with an evaluation metric proposed by David L. Davis and Donald Bouldin), which is defined as shown in Formula 3:

[0083]

[0084] where d i and d j are the average distances within class i and class j, respectively, and D ij Represents the distance between class i and class j. When DB reaches the minimum value, the number of clusters is optimal.

[0085] The above are only specific application examples of the present invention and do not constitute any limitation on the protection scope of the present invention. Any technical solution formed by equivalent transformation or equivalent replacement shall fall within the protection scope of the present invention.

Claims

1. A method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment, characterized in that: Includes the following: S1. Monitor the actual parameter characteristics of acoustic emission signals for active defects in pressure-bearing equipment and extract characteristic parameters of acoustic emission waveforms; Extraction of acoustic emission waveform characteristic parameters includes the following: Amplitude and energy are selected to reflect the intensity characteristics of acoustic emission signals; Determine A% of the maximum amplitude as the soft threshold value, and recalculate the characteristic parameters of the waveform so that the obtained characteristic parameters reflect the shape characteristics of the acoustic emission waveform; The margin factor is used to reflect the shape characteristics of the acoustic emission waveform; Through wavelet decomposition, the wavelet characteristic energy spectrum coefficient vectors of the acoustic emission signal are calculated in batches according to the wavelet characteristic energy spectrum coefficient definition to reflect the waveform frequency distribution characteristics; S2, taking the extracted feature parameters as feature vectors and using the similarity coefficient to define the distance between variables; S3, performing hierarchical clustering, classifying the parameters into one category one by one according to the similarity coefficients between the parameters, and obtaining a hierarchical clustering tree of the correlation of acoustic emission characteristic parameters; S4, set a threshold, and retain the parameters with correlation coefficients below the threshold as the feature vectors for cluster analysis; S5. Based on the determination of the cluster analysis feature vector, a K-means cluster analysis is performed on the selected vector.

2. According to claim 1, a method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment is characterized in that: The margin factor expression is shown in Formula 1: Where N represents the number of data points in the analysis sample, x n represents the time history of the sample signal, and T represents the time of all samples.

3. The method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment according to claim 1 is characterized in that: The similarity coefficient in step S2 is defined as follows: Let C ij For X i and X j The similarity coefficient between them has the following restrictions: |C ij |≤1, for all i,j; C ij =C ji , for all i,j; Absolute value of similarity coefficient |C ij The closer to 1, the i and X j The closer, C ij The closer it is to 0, the more distant the relationship between the two. For quantitative variables, the similarity coefficient used is X i and X j The correlation coefficient of The similarity coefficient is used to define the distance between variables, that is, d ij =1-|C ij |or 4. The method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment according to claim 3 is characterized in that: The correlation coefficient is the cosine of the angle after the data is standardized; the variable X i and X j The correlation coefficient is usually r ij Indicated by C ij ,Right now When C ij =1 indicates that the two variables are related. Generally, |C ij |≤1.

5. The method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment according to claim 1 is characterized in that: The hierarchical clustering in step S3 includes the following: SS3.

1. Calculate the distance between n parameters and obtain the distance matrix D between samples. (0) ; SS3.2, Initial (first step: i = 1) n samples each constitute a class, the number of classes k = n, the i-th class G i ={X (i) }(i=1,...,n); at this time, the distance between classes is the distance between samples, that is, D (1) =D (0) ; Then for sample X (i) (i=1,...,n) execute steps SS3.3 and SS3.4 of the merging process; SS3.3, the distance matrix D obtained in step SS3.2 (i-1) , merge the two classes with the smallest inter-class distance into a new class; at this time, the total number of classes k is reduced by 1 class, that is, k = n-i+1; SS3.

4. Calculate the distance between the new class and other classes to get a new distance matrix D (i) , if the total number of classes after merging is still greater than 1, repeat steps SS3.3 and SS3.4 until the total number of classes is 1, then go to step SS3.5; SS3.5, draw pedigree cluster diagrams; SS3.

6. Determine the number of categories and the members of each category.

6. The method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment according to claim 1 is characterized in that: The K-means clustering in step S5 includes the following contents: SS5.

1. Specify the distance between samples and determine three numbers: k is the number of categories, C is the minimum distance between categories, and R is the maximum distance within a category; take the first k sample points as the condensation points; SS5.2, calculate the distances between the k condensation points. If the smallest distance is less than C, merge the corresponding two condensation points, use the centroid of the two points as the new condensation point, and repeat step SS5.2 until the distances between all condensation points are greater than C; SS5.3, classify the remaining nk samples one by one, and for each sample, calculate the distance between the sample and all the condensation points. If the minimum distance is greater than R, the sample is used as a new condensation point; if the minimum distance is less than or equal to R, the sample is classified into the class where the condensation point closest to it is located; then recalculate the centroid of the class and use the centroid as the new condensation point; if the distances between condensation points are all greater than or equal to C, consider the next sample, otherwise merge them using step SS5.2 and then consider the next sample, until all samples are classified; SS5.4, classify the samples one by one according to step SS5.3 from the beginning to the end. The difference is: after a sample is classified, if the classification is consistent with the original, the centroid does not need to be calculated; if the classification is different from the original, the centroids of the two categories involved must be recalculated; if the new classification is the same as the previous one, the clustering process ends, otherwise repeat step SS5.4; The optimal number of clusters k in k-means clustering is determined by the Dacies&Bouldin criterion.

7. The method for extracting acoustic emission signal features based on active defect monitoring of pressure-bearing equipment according to claim 6 is characterized in that: The Dacies & Bouldin criterion is defined as shown in Formula 3: where d i and d j are the average distances within class i and class j, respectively, and D ij Represents the distance between class i and class j. When DB reaches the minimum value, the number of clusters is optimal.

Images