Tank bottom leakage point positioning method and system based on acoustic emission signal processing

Abstract

The application discloses a kind of based on acoustic emission signal processing's storage tank tank bottom leakage point positioning method and system, including using sensor array, acquisition multiple acoustic emission signals and carry out band-pass filtering;Signal segment is extracted using sliding time window, the rank order of each signal segment and the spireman correlation coefficient between all sensors are calculated, correlation coefficient matrix is formed and spatial feature and time domain feature are extracted, and are fused into comprehensive feature vector;Comprehensive feature vector is input into pre-trained random forest classifier, corrosion risk signal is identified and screened out;For corrosion risk signal, using the time difference of arrival of multiple sensors, with residual sum of squares as objective function, tank bottom circular region as constraint, optimal sound source coordinates are obtained by using trust region algorithm.The application realizes the high-precision identification of corrosion signal in complex acoustic field, and uses overdetermined positioning to improve the positioning accuracy of tank bottom edge region, with the advantages of reliable detection, accurate positioning and strong applicability.

Description

Technical Field

[0001] This invention relates to the field of tank corrosion detection technology, and in particular to a method and system for locating leak points at the bottom of storage tanks based on acoustic emission signal processing. Background Technology

[0002] Large oil storage tanks, as key infrastructure in the petrochemical and oil and gas storage and transportation industries, are responsible for storing flammable, explosive, toxic, and hazardous media such as crude oil, refined oil, and chemical raw materials. The tank bottom, as the critical part in contact with the foundation, is subjected to multiple effects over a long period, including media pressure, soil corrosion, and foundation settlement, making it highly susceptible to defects such as corrosion thinning, pitting, cracks, and weld leaks. Acoustic emission testing (AE), a dynamic non-destructive testing method, can detect elastic stress wave signals generated by the activity of internal defects in materials under stress. Compared with traditional testing methods, AE has unique advantages such as online detection, dynamic monitoring, full-area coverage, and high efficiency and speed. It is currently one of the few technologies that can assess the condition of the tank bottom without cleaning the tank.

[0003] Traditional acoustic emission techniques often employ simple threshold judgments or unsupervised clustering (such as K-means, DBSCAN, etc.) to coarsely distinguish signals. Clustering methods are sensitive to feature selection, lack labeled information, and struggle to stably separate true corrosion acoustic emission signals in high-noise backgrounds. They are prone to misidentifying water flow sounds or noise as corrosion events, leading to numerous false alarms. They may also miss weak true corrosion signals, severely interfering with subsequent localization and assessment. Furthermore, the inability to effectively distinguish interfering signals results in inaccurate identification results, especially when the tank contains internal media. They fail to fully utilize the spatial correlation patterns between multiple sensors and do not adequately consider the physical constraints of the tank during localization, resulting in low localization accuracy in edge areas or localization results that may extend beyond the tank bottom, lacking physical plausibility.

[0004] It is evident that existing technologies have significant shortcomings in terms of signal recognition reliability and edge region positioning accuracy in complex sound field environments, and there is an urgent need for an intelligent detection method that can achieve accurate signal classification and high-precision positioning. Summary of the Invention

[0005] Therefore, the purpose of this invention is to provide a method and system for locating leak points at the bottom of storage tanks based on acoustic emission signal processing. By extracting the Spearman correlation coefficient matrix features of multi-sensor signals and integrating traditional time-frequency features and spatial features, machine learning is used to achieve high-precision automatic differentiation of multiple signals. Overdetermined positioning algorithms are used to improve the positioning accuracy of edge areas and solve the problem of accurate leak point location.

[0006] To achieve the above objectives, the present invention provides a method for locating leak points at the bottom of storage tanks based on acoustic emission signal processing, comprising the following steps: S1. Simultaneously acquire multiple channels of bottom acoustic emission signals using a sensor array, and perform bandpass filtering on each channel of the bottom acoustic emission signal; S2. Analyze the filtered and retained acoustic emission signal from the bottom of the tank according to the preset sliding time window, and extract the signal segment of each sensor. S3. Calculate the rank order of each signal segment and calculate the Spearman correlation coefficient between the rank orders of signal segments of all sensor pairs to form a symmetric correlation coefficient matrix. S4. Extract spatial features from the symmetric correlation coefficient matrix and extract temporal features from each signal segment. Concatenate the spatial features and temporal features to form a comprehensive feature vector. S5. Input the comprehensive feature vector into the pre-trained classifier, output the category to which the signal belongs in the current time window, and filter out the corrosion risk signal from it; S6. For acoustic emission events identified as corrosion risk signals, the overdetermined positioning algorithm is used to calculate the coordinates of the sound source location by utilizing the time difference of arrival of signals from all triggering sensors.

[0007] More preferably, in S1, the sensor array consists of m acoustic emission sensors arranged at equal intervals on the lower side ring of the tank wall, where m ≥ 3, used to synchronously collect acoustic emission signals from the bottom of the tank.

[0008] Further preferably, in S3, the rank order of each signal segment is calculated, and the Spearman correlation coefficient between the rank orders of signal segments of all sensor pairs is calculated to form a symmetric correlation coefficient matrix, specifically including: S301, For each sensor's corresponding signal segment The amplitudes of the N discrete sampling points contained therein are sorted, and the sampling points are assigned ranks according to the sorting order; The rank of each sample value is determined according to its position in the original signal segment. The positions in the sequence are replaced sequentially to obtain the rank order of the signal segments. , S302. Define a sensor pair (p, q), where p and q take values ​​of 1, 2, ..., m respectively; m is the total number of sensors, and when p = q, it represents the sensor itself. S303. For each sensor pair (p, q), calculate the Spearman correlation coefficient according to the following formula. , in, Representative signal segment The One element, Representative signal segment The Each element, all constitute a The symmetric correlation coefficient matrix; S304, All calculated results Arranged into an m×m symmetric matrix, it serves as the Spearman correlation coefficient matrix.

[0009] Further preferably, in S4, spatial features are extracted from the symmetric correlation coefficient matrix, the spatial features including: the average value of all elements in the Spearman correlation coefficient matrix, and the average value of the correlation coefficients of adjacent sensor pairs on both sides of the main diagonal of the matrix; in S4, temporal features are extracted from each signal segment, including: For each signal segment Calculate the root mean square (RMS) and peak amplitude (PA), and combine the RMS and PA values ​​of each signal segment to form a set of time-domain features representing the current time window. .

[0010] More preferably, in S5, the comprehensive feature vector is input to a pre-trained classifier, wherein the classifier is selected as a random forest model, and the pre-training process of the classifier includes the following steps: The comprehensive feature vector is divided into a training set and a test set; the comprehensive feature vector in the training set is accompanied by a signal category label, which includes corrosion signal and water flow signal; A random forest classification model is used for classification to obtain the mapping relationship from the comprehensive feature vector to the class label. The hyperparameters of the random forest model were optimized and adjusted using the K-fold cross-validation method. The performance of the trained and optimized random forest model was evaluated using the test set.

[0011] More preferably, in S6, the overdetermined positioning algorithm specifically includes: S601. Based on the time when each sensor receives a signal, obtain the actual time difference between each sensor and the first sensor; S602, Set the sound source coordinates as follows Based on the coordinates of the sound source and the coordinates of each sensor, the theoretical time difference between each sensor and the first sensor is calculated; S603. Construct the residual square function between the actual time difference and the theoretical time difference as the objective function; S604. Based on the minimum range of the residual square function, set the physical constraints on the sound source coordinates within the circular area at the bottom of the storage tank. S605. The minimum value of the objective function under physical constraints is obtained by using the trust region algorithm to obtain the optimal sound source coordinates.

[0012] More preferably, the objective function is expressed by the following formula: ; ; ; in, Indicates the actual time difference. This indicates the time when the i-th sensor receives the signal. Indicates the moment when the first sensor receives a signal; Indicates the theoretical time difference. This represents the position coordinates of the i-th sensor. Indicates the coordinates of the sound source; Indicates the coordinates of the first sensor. Indicates the speed of sound.

[0013] More preferably, in S605, the step of using the trust region algorithm to solve for the minimum value of the objective function under the constraints to obtain the optimal sound source coordinates specifically includes the following steps: a: Initialization: Select initial sound source coordinates x0= The initial trust region radius Δ0 > 0, and the iterative convergence accuracy is set. e Maximum number of iterations kmax, trust region radius adjustment threshold or 1 and or 2 And 0 < or 1< or 2<1; b: Convergence check: If the current error is less than or equal to the iteration convergence precision ε or the number of iterations exceeds kmax, then terminate the iteration; output the current iteration point x. k Use the optimal sound source coordinates; otherwise, proceed to step c. c: Construct a quadratic approximation model: using the current iteration point x k The second-order Taylor expansion of a general function f(x) is used as a quadratic approximation function. , where d is the trial step size, d = (△x, △y); For the objective function in The gradient vector at that point, It is the dot product of the gradient and the step size. For the objective function in The second derivative at point; d: The actual trial step size d when the quadratic approximation model reaches its minimum value under physical constraints within the trust region and the circular region at the bottom of the storage tank.k ; e: Based on the actual trial step size d k Calculate the ratio of the actual decline to the predicted decline. , f: Determine the ratio Adjust the trust region radius according to the relationship between the threshold and the trust region radius until a ratio is found. The sound source coordinates closest to 1 are taken as the optimal sound source coordinates.

[0014] More preferably, the judgment ratio The relationship between the threshold for adjusting the trust region radius and the threshold for adjusting the trust region radius includes: If r k If x ≤ 0, then reject the trial step and let x k+1 =x k Return to step d and calculate the actual trial step size. ; If 0 <r k < or 1. Then accept the trial step xk+1=x k +d k Let the shrinkage radius Δ k+1 = / 4; like or 1≤r k ≤ or 2, then accept the trial step x k+1 =x k +d k Let Δ k+1 =Δ k ; If r k > or 2, then accept the trial step x k+1 =x k +d k and expand radius Δ k+1 =2Δ k ; Let k = k + 1, for the next iteration point x k+1 Return to step b and continue iterating.

[0015] The present invention also provides a tank bottom leak point location system based on acoustic emission signal processing, comprising: Sensor array: Used for synchronous acquisition of acoustic emission signals from the bottom of the tank; Signal preprocessing module: Used to perform bandpass filtering on the acquired multi-channel signals; Feature extraction module: Used to calculate the Spearman correlation coefficient matrix between each sensor signal segment, extract spatial features from it, extract the temporal features of each signal segment, and fuse them into a comprehensive feature vector; Classification module: Built-in pre-trained random forest classifier is used to classify the comprehensive feature vector, identify and filter out corrosion risk signals; Location module: Used to calculate the location coordinates of the sound source based on the time difference of arrival of multiple sensors for events identified as corrosion risk signals. The overdetermined location algorithm uses a least squares model of the measurement time difference and the theoretical time difference as the objective function, the circular area at the bottom of the storage tank as the constraint condition, and a trust region algorithm to solve the problem.

[0016] This application discloses a method and system for locating leak points at the bottom of storage tanks based on acoustic emission signal processing. Compared with existing technologies, it has at least the following advantages: This patent incorporates the Spearman correlation coefficient matrix of multi-sensor signals into acoustic emission analysis. By fusing it with the time-frequency characteristics of the acoustic emission signals and utilizing a machine learning model, it distinguishes between corrosion sounds, water flow sounds, structural reflection sounds, and environmental noise, thus fundamentally solving the signal classification problem. Furthermore, it employs an overdetermined positioning algorithm that incorporates the physical constraints of the tank, replacing the traditional triangulation method which is sensitive to edge errors. This effectively improves the positioning accuracy and stability of the entire tank bottom area using multi-sensor data. This patent integrates intelligent screening with precise positioning, forming a complete technical solution for detecting and locating leaks at the bottom of large storage tanks. It enables online assessment and precise location of the tank bottom corrosion state without interrupting production or cleaning the tank, avoiding economic losses caused by periodic tank opening and maintenance. This approach improves economic efficiency and convenience while ensuring safety.

[0017] This application achieves precise localization of corroded areas through a three-step process of acoustic signal classification, identification, and localization. The detection results not only reveal the overall condition of the tank bottom plate but also identify the corrosion status of other internal facilities, providing crucial guidance for subsequent targeted maintenance. Furthermore, the system is also suitable for corrosion activity assessment and life extension management of aging storage tanks, providing auxiliary location for sudden minor leaks, and establishing an acoustic baseline for new or repaired tanks. The online acoustic emission detection technology for tank bottoms has a wide range of applications, low cost, and high detection efficiency, offering significant technical advantages in corrosion detection and safety assessment of large storage tanks. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing, as described in this invention.

[0019] Figure 2 This is a flowchart of the trust region algorithm in this application. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] like Figure 1 As shown, one embodiment of the present invention provides a method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing, which includes the following steps: S1. A sensor array is used to synchronously acquire multiple channels of acoustic emission signals from the tank bottom, and bandpass filtering is applied to each channel of the acoustic emission signal from the tank bottom. Further, in S1, the sensor array consists of m acoustic emission sensors (m ≥ 3) arranged at equal intervals in a ring around the lower side of the tank wall, used to synchronously acquire the acoustic emission signals from the tank bottom, obtaining m channels of raw time-domain signals. ,in Each signal is bandpass filtered to preserve the target frequency band and suppress out-of-band noise.

[0022] S2. Analyze the filtered and retained acoustic emission signal from the bottom of the tank according to the preset sliding time window, and extract the signal segment from each sensor; analyze the signal extraction time window, assuming the time window length is... The original time-domain signal after preprocessing for each channel. In the middle, extract the signal segment within the length of the time window; then... The signal segments collected by each sensor are denoted as Its length is the same as the time window length, denoted as At this point, we can obtain the signal segment of the raw time-domain signal acquired by each sensor within the time window: ; S3. Calculate the rank order of each signal segment and calculate the Spearman correlation coefficient between the rank orders of signal segments of all sensor pairs to form a symmetric correlation coefficient matrix. More preferably, in S3, the calculation of the rank order of each signal segment includes: Let the time window signal segment of the p-th sensor be... Calculate the rank order of the segment of length N. ; right All sampled values ​​are arranged in ascending order of numerical value. ; Assign ranks to signal segments in an ordered sequence, from 1 to N in ascending order.

[0023] The rank of each sample value is determined according to its position in the original signal segment. The positions in the sequence are replaced sequentially to obtain the rank order of the signal segments. ; The Spearman correlation coefficients between the rank order of signal segments from all sensor pairs are calculated, forming a symmetric correlation coefficient matrix, which specifically includes: S301, For each sensor's corresponding signal segment Sort the amplitudes of the N discrete sampling points contained in the signal segment, determine the rank of each sampling point, and calculate the rank order of each signal segment. For a detailed calculation process, please refer to the following example: Suppose a certain signal segment =[5.2,2.1,8.3,3.7] After sorting: [2.1, 3.7, 5.2, 8.3] Assign rank: [1,2,3,4] Replaced in original order: the original first value was 5.2 (rank 3), the second value was 2.1 (rank 1), the third value was 8.3 (rank 4), and the fourth value was 3.7 (rank 2); therefore, the rank order is: Rank( =[3,1,4,2] S302. Define a sensor pair (p, q), where p and q take values ​​of 1, 2, ..., m, where m is the total number of sensors; when p = q, it represents the sensor itself. S303. For each sensor pair (p, q), calculate the Spearman correlation coefficient according to the following formula. , in, Representative signal segment The One element, Representative signal segment The Each element, all constitute a The symmetric correlation coefficient matrix; S304, All calculated results Arranged into an m×m symmetric matrix, it serves as the Spearman correlation coefficient matrix.

[0024] S4. Extract spatial features from the symmetric correlation coefficient matrix and extract temporal features from each signal segment. Concatenate the spatial features and temporal features to form a comprehensive feature vector. In S4, spatial features are extracted from the symmetric correlation coefficient matrix. This includes: the average value of all elements in the Spearman correlation coefficient matrix, and the average value of the correlation coefficients of adjacent sensor pairs on both sides of the main diagonal of the matrix.

[0025] In S4, temporal features are extracted from each signal segment, including: For each signal segment Calculate its root mean square (RMS) and peak amplitude, and form a set of time-domain features representing the current time window by combining the RMS and peak amplitude of each signal segment. .

[0026] The obtained spatial features With the obtained time-frequency features The features are concatenated to form a high-dimensional comprehensive feature vector. This integrated feature vector A classifier that takes signal feature parameters as input.

[0027] S5. Input the comprehensive feature vector into the pre-trained classifier, output the category to which the signal belongs in the current time window, and filter out the corrosion risk signal from it; More preferably, in S5, the comprehensive feature vector is input to a pre-trained classifier, wherein the classifier is selected as a random forest model, and the pre-training process of the classifier includes the following steps: The comprehensive feature vector is divided into a training set and a test set; the comprehensive feature vector in the training set is accompanied by a signal category label, which includes corrosion signal and water flow signal; A random forest classification model is used for classification to obtain the mapping relationship from the comprehensive feature vector to the class label. The hyperparameters of the random forest model were optimized and adjusted using the K-fold cross-validation method. The performance of the trained and optimized random forest model was evaluated using the test set.

[0028] The training set is used to learn and tune the parameters of the random forest classification model, while the test set is used to evaluate the model's classification performance and prevent overfitting. A random forest model is chosen as the base model for machine learning, and the initial model is trained using the training set to obtain features from the feature vectors. The mapping relationship to category labels. The random forest classification model outputs the probability distribution of the feature vector belonging to each category. . This represents the event category label corresponding to the signal in the training set. The system outputs the event with the highest probability as the judgment result for the current signal. The trained and optimized model was evaluated using a reserved independent test set. The overall classification accuracy and precision of the model were calculated. In particular, attention should be paid to the model's accuracy and false alarm rate in identifying corrosion and cracking signal sources to ensure that it meets the requirements of engineering applications.

[0029] S6. For acoustic emission events identified as corrosion risk signals, the overdetermined positioning algorithm is used to calculate the coordinates of the sound source location by utilizing the time difference of arrival of signals from all triggering sensors.

[0030] More preferably, in S6, the overdetermined positioning algorithm specifically includes: S601. Based on the time when each sensor receives a signal, obtain the actual time difference between each sensor and the first sensor; S602, Set the sound source coordinates as follows Based on the coordinates of the sound source and the coordinates of each sensor, the theoretical time difference between each sensor and the first sensor is calculated; S603. Construct the residual square function between the actual time difference and the theoretical time difference as the objective function; more preferably, the objective function is expressed by the following formula: ; ; ; ; in, Indicates the actual time difference. This indicates the time when the i-th sensor receives the signal. Indicates the moment when the first sensor receives a signal; Indicates the theoretical time difference. This represents the position coordinates of the i-th sensor. Indicates the coordinates of the sound source; Indicates the coordinates of the first sensor. Indicates the speed of sound.

[0031] S604. Based on the minimum range of the residual square function, set physical constraints on the sound source coordinates within the circular area at the bottom of the storage tank; the minimum range of the objective function is limited: the coordinates of the sound emission source must be inside the tank. Therefore, additional conditions are needed. r is the radius of the storage tank, and the center point of the storage tank is the origin.

[0032] S605. The minimum value of the objective function under physical constraints is obtained by using the trust region algorithm to obtain the optimal sound source coordinates.

[0033] More preferably, in S605, the step of using the trust region algorithm to solve for the minimum value of the objective function under the constraints to obtain the optimal sound source coordinates specifically includes the following steps: a: Initialization: Select initial sound source coordinates x0= The initial trust region radius Δ0 > 0, and the iterative convergence accuracy is set. e Maximum number of iterations kmax, trust region radius adjustment threshold or 1 and or 2, And 0 < or 1< or 2 < 1; b: Convergence check: If the current error is less than or equal to the iteration convergence precision ε or the number of iterations exceeds kmax, then terminate the iteration; output the current iteration point x. k Use the optimal sound source coordinates; otherwise, proceed to step c. c: Construct a quadratic approximation model: using the current iteration point x k The second-order Taylor expansion of a general function f(x) is used as a quadratic approximation function. , where d is the trial step size, d = (△x, △y); For the objective function in The gradient vector at that point, It is the dot product of the gradient and the step size. For the objective function in The second derivative at point; d: The actual trial step size d when the quadratic approximation model reaches its minimum value under physical constraints within the trust region and the circular region at the bottom of the storage tank. k ; e: Based on the actual trial step size d k Calculate the ratio of the actual decline to the predicted decline. , f: Determine the ratio Adjust the trust region radius according to the relationship between the threshold and the trust region radius until a ratio is found. The sound source coordinates closest to 1 are taken as the optimal sound source coordinates.

[0034] like Figure 2 As shown, more preferably, the judgment ratio The relationship between the trust region radius adjustment threshold and the trust region radius adjustment threshold is used to adjust the trust region radius, including the following steps: First, set the iteration count k=0, the initial value x0, and the error threshold. eh, Let the initial trust region radius be: Δ0 = Solve for the trial step size ,like ≤ e Then the process ends; otherwise, the ratio is calculated. Determine the ratio Relationship with the trust region radius adjustment threshold: Set the trust region radius adjustment threshold or1 = 0.25 or 2 = 0.75; If r k If x ≤ 0, then reject the trial step and let x k+1 =x k Return to step d and calculate the actual trial step size. ; If 0 <r k < or 1. Then accept the trial step xk+1=x k +d k Let the shrinkage radius Δ k+1 = / 4; like or 1≤r k ≤ or 2, then accept the trial step x k+1 =x k +d k Let Δ k+1 =Δ k ; If r k > or 2, then accept the trial step x k+1 =x k +d k and expand radius Δ k+1 =2Δ k ; Let k = k + 1, for the next iteration point x k+1 Return to step b and continue iterating.

[0035] The present invention also provides a tank bottom leak point location system based on acoustic emission signal processing, for implementing the above-mentioned location method, comprising: Sensor array: used for synchronous acquisition of acoustic emission signals from the bottom of the tank; Signal preprocessing module: used to perform bandpass filtering on the acquired multi-channel signals; Feature extraction module: used to calculate the Spearman correlation coefficient matrix between each sensor signal segment, extract spatial features from it, extract the temporal features of each signal segment, and fuse them into a comprehensive feature vector; Classification module: Built-in pre-trained random forest classifier is used to classify the comprehensive feature vector, identify and filter out corrosion risk signals; Location module: Used to calculate the location coordinates of the sound source based on the time difference of arrival of multiple sensors for events identified as corrosion risk signals. The overdetermined location algorithm uses a least squares model of the measurement time difference and the theoretical time difference as the objective function, the circular area at the bottom of the storage tank as the constraint condition, and a trust region algorithm to solve the problem.

[0036] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing, characterized in that, Includes the following steps: S1. Simultaneously acquire multiple channels of bottom acoustic emission signals using a sensor array, and perform bandpass filtering on each channel of the bottom acoustic emission signal; S2. Analyze the filtered and retained acoustic emission signal from the bottom of the tank according to the preset sliding time window, and extract the signal segment of each sensor. S3. Calculate the rank order of each signal segment and calculate the Spearman correlation coefficient between the rank orders of signal segments of all sensor pairs to form a symmetric correlation coefficient matrix. S4. Extract spatial features from the symmetric correlation coefficient matrix and extract temporal features from each signal segment. Concatenate the spatial features and temporal features to form a comprehensive feature vector. S5. Input the comprehensive feature vector into the pre-trained classifier, output the category to which the signal belongs in the current time window, and filter out the corrosion risk signal from it; S6. For acoustic emission events identified as corrosion risk signals, the overdetermined positioning algorithm is used to calculate the coordinates of the sound source location by utilizing the time difference of arrival of signals from all triggering sensors.

2. The method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing according to claim 1, characterized in that, In S1, the sensor array consists of m acoustic emission sensors arranged at equal intervals on the lower side ring of the tank wall, where m ≥ 3, used to synchronously collect acoustic emission signals from the bottom of the tank.

3. The method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing according to claim 1, characterized in that, In S3, the rank order of each signal segment is calculated, and the Spearman correlation coefficients between the rank orders of signal segments for all sensor pairs are calculated, forming a symmetric correlation coefficient matrix, specifically including: S301, For each sensor's corresponding signal segment The amplitudes of the N discrete sampling points contained therein are sorted, and the sampling points are assigned ranks according to the sorting order; The rank of each sample value is determined according to its position in the original signal segment. The positions in the sequence are replaced sequentially to obtain the rank order of the signal segments. ; S302. Define a sensor pair (p, q), where p and q take values ​​of 1, 2, ..., m respectively; m is the total number of sensors, and when p = q, it represents the sensor itself. S303. For each sensor pair (p, q), calculate the Spearman correlation coefficient according to the following formula. in, Representative signal segment The One element, Representative signal segment The Each element, all constitute a The symmetric correlation coefficient matrix; S304, All calculated results Arranged into an m×m symmetric matrix, it serves as the Spearman correlation coefficient matrix.

4. The method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing according to claim 3, characterized in that, In S4, spatial features are extracted from the symmetric correlation coefficient matrix. These spatial features include: the average value of all elements in the Spearman correlation coefficient matrix and the average value of the correlation coefficients between adjacent sensor pairs on both sides of the main diagonal of the matrix. In S4, temporal features are extracted from each signal segment, including: For each signal segment Calculate the root mean square (RMS) and peak amplitude (PA), and combine the RMS and PA values ​​of each signal segment to form a set of time-domain features representing the current time window. .

5. The method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing according to claim 1, characterized in that, In S5, the comprehensive feature vector is input to the pre-trained classifier, which is a random forest model. The pre-training process of the classifier includes the following steps: The comprehensive feature vector is divided into a training set and a test set; the comprehensive feature vector in the training set is accompanied by a signal category label, which includes corrosion signal and water flow signal; The random forest classification model is used for classification to obtain the mapping relationship from the comprehensive feature vector to the category label; The hyperparameters of the random forest model were optimized and adjusted using the K-fold cross-validation method. The performance of the trained and optimized random forest model was evaluated using the test set.

6. The method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing according to claim 1, characterized in that, In S6, the over-determined localization algorithm specifically includes: S601. Based on the time when each sensor receives a signal, obtain the actual time difference between each sensor and the first sensor; S602, Set the sound source coordinates as follows Based on the coordinates of the sound source and the coordinates of each sensor, the theoretical time difference between each sensor and the first sensor is calculated; S603. Construct the residual square function between the actual time difference and the theoretical time difference as the objective function; S604. Based on the minimum range of the residual square function, set the physical constraints on the sound source coordinates within the circular area at the bottom of the storage tank. S605. The minimum value of the objective function under physical constraints is obtained by using the trust region algorithm to obtain the optimal sound source coordinates.

7. The method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing according to claim 6, characterized in that, The objective function is expressed by the following formula: ； ； in, Indicates the actual time difference. This indicates the time when the i-th sensor receives the signal. Indicates the moment when the first sensor receives a signal; Indicates the theoretical time difference. This represents the position coordinates of the i-th sensor. Indicates the coordinates of the sound source; Indicates the coordinates of the first sensor. Indicates the speed of sound.

8. The method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing according to claim 6, characterized in that, In S605, the step of using the trust region algorithm to solve for the minimum value of the objective function under the constraints to obtain the optimal sound source coordinates specifically includes the following steps: a: Initialization: Select initial sound source coordinates x0= The initial trust region radius Δ0 > 0, and the iterative convergence accuracy is set. ε Maximum number of iterations kmax, trust region radius adjustment threshold η 1 and η2 ， And 0 < η 1< η 2 < 1; b: Convergence check: If the current error is less than or equal to the iteration convergence precision ε or the number of iterations exceeds kmax, then terminate the iteration; output the current iteration point x. k Use the optimal sound source coordinates; otherwise, proceed to step c. c: Construct a quadratic approximation model: using the current iteration point x k The second-order Taylor expansion of a general function f(x) is used as a quadratic approximation function. , where d is the trial step size, d = (△x, △y); For the objective function in The gradient vector at that point, It is the dot product of the gradient and the step size. For the objective function in The second derivative at that point; d: The actual trial step size d when the quadratic approximation function reaches its minimum value under physical constraints within the trust region and the circular region at the bottom of the storage tank. k ; e: Based on the actual trial step size d k Calculate the ratio of the actual decline to the predicted decline. ; f: Determine the ratio Adjust the trust region radius according to the relationship between the threshold and the trust region radius until a ratio is found. The sound source coordinates closest to 1 are taken as the optimal sound source coordinates.

9. The method for locating leak points at the bottom of a storage tank based on acoustic emission signal processing according to claim 8, characterized in that, The judgment ratio The relationship between the trust region radius adjustment threshold and the trust region radius is used to adjust the trust region radius. include: If r k If x ≤ 0, then reject the trial step and let x k+1 =x k Return to step d and calculate the actual trial step size; If 0 <r k < η 1. Then accept the trial step, let x k+1 =x k +d k Let the radius Δ k+1 = / 4; like η 1≤r k ≤ η 2. Then accept the trial step x. k+1 =x k +d k Let the radius Δ k+1 =Δ k ; If r k > η 2, then accept the trial step x k+1 =x k +d k and expand radius Δ k+1 =2Δ k ; Let k = k + 1, for the next iteration point x k+1 Return to step b and continue iterating.

10. A tank bottom leak location system based on acoustic emission signal processing, characterized in that, include: Sensor array: Used for synchronous acquisition of acoustic emission signals from the bottom of the tank; Signal preprocessing module: Used to perform bandpass filtering on the acquired multi-channel signals; Feature extraction module: Used to calculate the Spearman correlation coefficient matrix between each sensor signal segment, extract spatial features from it, extract the temporal features of each signal segment, and fuse them into a comprehensive feature vector; Classification module: Built-in pre-trained random forest classifier is used to classify the comprehensive feature vector, identify and filter out corrosion risk signals; Location module: Used to calculate the location coordinates of the sound source based on the time difference of arrival of multiple sensors for events identified as corrosion risk signals. The overdetermined location algorithm uses the least squares model of the measurement time difference and the theoretical time difference as the objective function, the circular area at the bottom of the storage tank as the constraint condition, and the trust region algorithm to solve the problem.

Images