Transformer fault sound source localization method and system based on orthogonal clarity beam

Abstract

The present invention discloses a transformer fault sound source localization method and system based on orthogonal clear beams, belonging to the technical field of sound source localization and state detection of power equipment. The method comprises: arranging N acoustic sensors in an array in a three-dimensional space around the transformer to form a three-dimensional sound field measurement surface, collecting sound pressure signals and performing preprocessing; constructing a cross-spectrum matrix between the acoustic sensors based on the preprocessed sound pressure signals, and performing eigenvalue decomposition and principal component extraction on the cross-spectrum matrix; calculating the orthogonal beam output based on the cross-spectrum matrix after principal component extraction to generate a preliminary sound source map, and screening the grid points in the preliminary sound source map based on a preset sound source intensity threshold; iteratively clearing the screened sound source map using a CLEAN-SC algorithm, and outputting a sound source localization map after iterative clear beam processing when an iterative termination condition is met; marking areas with intensities higher than a set threshold in the sound source localization map, analyzing the location of the fault source, and generating a fault localization report.

Description

Technical Field

[0001] The present invention belongs to the technical field of sound source localization and status detection of electric power equipment, and particularly relates to a transformer fault sound source localization method and system based on orthogonal clarity beams. Background Art

[0002] As one of the core devices in the power system, the operating status of the transformer is particularly critical to the stability and reliability of the entire power grid. Therefore, transformer fault detection has become a research hotspot in this field. Transformer fault detection technology is constantly improving and developing. Traditional fault detection methods mainly include vibration signal detection, oil chromatography detection, spectral signal detection, gas composition detection, and acoustic signal detection. Vibration signal detection detects the operating status of the transformer by collecting its vibration signal, which can more accurately identify its operating defects. However, the disadvantage is that it requires contact detection. Acoustic signal detection can perform non-contact detection of the transformer through sound signals. It is convenient to operate and has efficient detection, but it cannot specifically locate the transformer sound source.

[0003] Compared to other detection technologies, sound source localization technology not only analyzes the transformer's operating status but also intuitively indicates the location of the transformer's sound source. It also maintains non-contact and anti-interference characteristics, offering significant advantages for transformer status monitoring. Sound source localization technology precisely locates and identifies sound sources by analyzing the sound's spectrum, waveform, and time domain characteristics. Initially widely used in the automotive and aviation sectors, it also holds great promise for monitoring the status of power equipment.

[0004] Existing sound source localization technology has problems such as low resolution and long positioning time. How to comprehensively improve the accuracy and computational efficiency of fault sound source localization based on the characteristics of transformer fault sound sources requires further research and optimization.

[0005] Prior art document 1 (CN114235366A) discloses an identification scheme based on transformer noise and vibration characteristics in the field of transformer fault identification. It collects transformer vibration noise sound pressure signals, calculates characteristic values, and uses BP neural network to perform fault diagnosis. However, it relies on vibration signal collection and analysis. In actual applications, it is susceptible to external noise interference, and the BP neural network relies on training data and has high computing resource requirements, resulting in problems such as affected recognition accuracy, diagnostic reliability, and computing efficiency.

[0006] Prior Art Document 2 ("Identification of Weak Spots in Automobile Dashboard Sound Insulation Based on CLEAN-SC Clear Beamforming" (Acoustic Technology, Vol. 34, No. 5, October 2015, Yang Yang and Chu Zhigang)) discloses a solution for identifying sound sources in automotive acoustics based on CLEAN-SC clear beamforming. By utilizing CLEAN-SC clear beamforming, this solution improves resolution, attenuates sidelobes, and more accurately identifies single and incoherent sound sources. Experiments have identified weak spots in automobile dashboard sound insulation and their causes. However, this solution suffers from numerous challenges. For example, its resolution and accuracy are poor when identifying low-frequency noise, making it difficult to meet high-precision localization requirements. Its high computational complexity creates a performance bottleneck in scenarios requiring high real-time performance. Furthermore, its applicability is unstable across different frequency ranges and in complex, dynamic sound field environments, limiting its widespread application. Summary of the Invention

[0007] In order to solve the technical problems existing in the prior art, the present invention provides a transformer fault sound source localization method and system based on orthogonal clarity beams.

[0008] The present invention adopts the following technical solutions.

[0009] A first aspect of the present invention provides a transformer fault sound source localization method based on orthogonal clarity beams, characterized by comprising the following steps:

[0010] With the center of the transformer under test as the coordinate origin, N acoustic sensors are arranged in an array in the three-dimensional space around the transformer to form a three-dimensional sound field measurement surface, and the sound pressure signal is collected and pre-processed;

[0011] Based on the preprocessed sound pressure signal, a cross-spectral matrix between acoustic sensors is constructed, and the cross-spectral matrix is subjected to eigenvalue decomposition. The first K largest eigenvalues and their corresponding eigenvectors are selected to reconstruct the dimension-reduced cross-spectral matrix. The orthogonal beam output is calculated based on the dimension-reduced cross-spectral matrix to generate a preliminary sound source map. The grid points in the preliminary sound source map are screened using a preset sound source intensity adaptive dynamic threshold. The screened sound source map is iteratively clarified using the CLEAN-SC algorithm. When the iterative termination condition is met, the sound source localization map after iteratively clarified beam processing is output.

[0012] The area with intensity higher than the set threshold is marked in the sound source localization map, and the location of the fault source is analyzed based on the physical structure and operating status of the tested transformer to generate a fault location report.

[0013] Optionally, the preprocessing of the sound pressure signal includes:

[0014] The sound pressure signal is framed and windowed, and each frame of the time domain signal is converted into a frequency domain signal through Fourier transform to obtain the frequency domain signal vector , and perform filtering and noise reduction.

[0015] Optionally, constructing a cross-spectral matrix includes:

[0016] Based on the frequency domain signal vector obtained after preprocessing Constructing the cross-spectral matrix between acoustic sensors :

[0017]

[0018] in, is the average number of frames; is the frequency domain signal vector of the nth frame; for The conjugate transpose of

[0019] Cross-spectral matrix Perform eigenvalue decomposition to obtain the product of three matrices: an orthogonal matrix consisting of eigenvectors, a diagonal matrix whose diagonal elements are eigenvalues arranged in descending order, and the conjugate transpose of the orthogonal matrix.

[0020] From the decomposed eigenvalues, select the first 20 eigenvalues with the largest values and extract their corresponding 20 eigenvectors. Multiply the outer product matrix of each eigenvalue with its corresponding eigenvector, and perform weighted summation on all 20 outer product matrices to obtain the cross-spectral matrix after dimensionality reduction. .

[0021] Optionally, the calculating the orthogonal beam output based on the cross-spectral matrix after principal component extraction includes:

[0022] The measurement surface is discretized into a grid surface containing M grid points, and the sound source calculation plane is defined based on the grid surface;

[0023] Calculate the orthogonal beam output of each grid point in the sound source calculation plane to generate a preliminary sound source map;

[0024] The grid points in the preliminary sound source map are screened according to the preset sound source intensity threshold.

[0025] Optionally, discretizing the measurement surface into a grid surface including M grid points and generating a preliminary sound source map includes:

[0026] The measurement surface is discretized into a grid surface containing M grid points. The grid point set is , the sound source calculation plane contains M grid points, and the focusing plane is defined for each grid point. The focusing vector of the mth grid point is , ;

[0027] The orthogonal beam output at the mth grid point is:

[0028]

[0029] in, is the dimension-reduced cross-spectral matrix, for The conjugate transpose of .

[0030] Optionally, the iteratively clarifying the filtered sound source map using the CLEAN-SC algorithm includes:

[0031] The filtered sound source map is set as the initial sound source map, and the dimension-reduced cross-spectrum matrix is used as the initial value of the cross-spectrum matrix;

[0032] In the i-th iteration, the main lobe peak position is searched from the current sound source map as the strongest sound source, and the direction vector and weight vector are calculated based on this position;

[0033] According to the coherent source component contribution of the current main lobe peak, the component is removed from the cross-spectrum matrix, and the cross-spectrum matrix of the remaining sound source information is updated;

[0034] Based on the direction vector and the weight vector, the contribution of the current main lobe peak is removed from the sound source map to generate an updated sound source localization map;

[0035] storing the sharpened beam data determined by the gain factor and phase angle parameters in each iteration;

[0036] When the preset criteria are met, the iteration is terminated and the sound source localization map after iteratively clarified beam processing is output.

[0037] Optionally, the main lobe peak search and direction / weighted vector calculation include:

[0038] In the i-th iteration, in the sound source map Search for the position of the main lobe peak in :

[0039]

[0040] calculate Direction vector :

[0041]

[0042] in, It is i The search for the main lobe peak position in the iteration With the n The distance between the microphones; It is n The location of the microphones;

[0043] calculate The weight vector :

[0044] .

[0045] Optionally, updating the cross-spectral matrix of the remaining sound source information includes:

[0046] Assume that in the i-th iteration is composed of a single coherent source component The resulting cross-spectral matrix:

[0047]

[0048] in, is the beamforming output corresponding to the main lobe peak in the i-1th iteration; is the focusing vector corresponding to a single coherent source component in the i-th iteration; for The conjugate transpose of

[0049]

[0050] according to , update the cross-spectral matrix:

[0051]

[0052] in, is the cross-spectral matrix at the i-th iteration, which is used to store the updated sound source information during the iteration process. It is the weight coefficient for adjusting the attenuation ratio of the sound source component during the iteration process.

[0053] Optionally, the fault location analysis and report generation includes:

[0054] Based on a dynamic threshold, continuous areas with intensities above the threshold are extracted from the clear sound source localization map. Potential fault sources are identified through image processing algorithms. The sound source grid coordinate system is aligned with the three-dimensional physical structure model of the transformer to establish a mapping relationship between the fault area and the actual component.

[0055] Combined with the sound source intensity distribution, direction vector, weighted vector and the physical structure of the transformer, high-intensity areas are located and optimized, and coherence analysis is performed to determine the specific location and priority of the fault source;

[0056] The output includes a location report of the fault source coordinates, intensity, associated components, and fault type. The location results are verified in combination with the transformer operating status, and repair measures are deployed.

[0057] A second aspect of the present invention provides a transformer fault sound source localization system based on orthogonal clarity beams, based on the transformer fault sound source localization method based on orthogonal clarity beams described in the first aspect of the present invention, the system comprising:

[0058] The sensor array module is used to arrange acoustic sensors in a three-dimensional array and collect transformer sound pressure signals in real time;

[0059] Signal preprocessing module, used to process the original sound pressure signal, including framing, windowing, Fourier transform and filtering noise reduction;

[0060] Cross-spectral matrix calculation module, used to construct the cross-spectral matrix and extract the principal component features;

[0061] Orthogonal beamforming module, used to generate preliminary sound source maps and screen effective grid points;

[0062] Iterative clarification module, used to iteratively optimize the sound source localization map using the OB-CLEAN-SC algorithm;

[0063] Fault analysis module, used to analyze the location and priority of fault sources in relation to transformer structure;

[0064] The report generation module is used to generate a fault source location report.

[0065] Compared with the existing technology, the beneficial effects of the present invention are embodied in:

[0066] 1. To address the problems of large computational complexity and poor real-time performance in existing technologies, the present invention uses dynamic grid compression technology to retain effective grid points and extract the principal components of the cross-spectral matrix, solving the resource waste problem caused by traditional full-grid computing, improving computing efficiency, and enabling the algorithm to run in real time in complex scenarios such as substations.

[0067] 2. To address the problems of low sound source resolution and ambiguous positioning in existing technologies, the present invention solves the problems of strong sidelobe interference and low low-frequency resolution in traditional beamforming through the orthogonal beamforming (OB) algorithm and iterative clearing (CLEAN-SC), thereby improving the spatial resolution of sound source positioning.

[0068] 3. In response to the problems of noise sensitivity and high false positive / missed positive rates in the existing technology, the present invention solves the problems of vibration signals being susceptible to electromagnetic interference and fixed thresholds having poor adaptability through non-contact acoustic detection and adaptive dynamic thresholds, thereby improving the anti-interference ability in complex noise environments and reducing the false detection rate.

[0069] 4. To address the problems in existing technologies where the results lack physical basis and deviate greatly from the actual component positions, the present invention iteratively updates the cross-spectral matrix based on the sound field propagation law and aligns it with the sound source map-transformer three-dimensional structure mapping. This solves the problem of mismatch between positioning results and actual component positions caused by ignoring physical constraints in traditional methods, thereby improving the physical rationality of fault location and its maintenance guidance value. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1Provides a schematic diagram of beamforming sound source identification space for an embodiment of the present invention;

[0071] Figure 2 Provides a flow chart of the OB-CLEAN-SC algorithm for an embodiment of the present invention;

[0072] Figure 3 A flowchart of a method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0073] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments are only for illustrating the present invention, rather than for limiting the scope of protection of the present invention.

[0074] Figure 1 It is a schematic diagram of the sound source recognition space of the algorithm of the present invention. Figure 2 This is the flow chart of the OB-CLEAN-SC algorithm.

[0075] In embodiment 1, the present invention provides a transformer fault location method based on an orthogonal clarity beamforming algorithm, such as Figure 3 As shown, the following steps are included:

[0076] Step 1: With the center of the transformer under test as the coordinate origin, N acoustic sensors are arranged in an array in the three-dimensional space around the transformer to form a three-dimensional sound field measurement surface, and the sound pressure signal is collected and pre-processed.

[0077] Preferably, in step 1, the coordinate vector of the nth acoustic sensor is , the coordinate vector is a three-dimensional coordinate;

[0078] Preferably, the preprocessing of the sound pressure signal includes:

[0079] The sound pressure signal is framed and windowed, and each frame of the time domain signal is converted into a frequency domain signal through Fourier transform to obtain the frequency domain signal vector , and perform filtering and noise reduction.

[0080] Specifically, to locate the sound source of a transformer fault, N sensors must first be arranged in an array within the sound field, forming a measurement plane for collecting sound pressure data near the transformer. The coordinates of each sensor are represented by a three-dimensional coordinate vector, with the origin set at the center of the transformer. This coordinate system uses a three-dimensional Cartesian coordinate system, with the center of the transformer as the reference point. The location of each sensor within the transformer is defined as a three-dimensional coordinate (x, y, z). The number of sensors, N, is typically greater than the number of sound sources. This is because, in practical applications, the sensor array can cover multiple angles of the transformer, providing more comprehensive sound data collection. To ensure accurate sound source localization, the measurement plane must be perpendicular to an axis of the sensor array plane. This plane is defined as the sound source focusing grid point plane. The location of the focusing grid point plane is determined based on the actual transformer structure and the possible location of the fault source. In practical transformer applications, the sensor array should be placed in the transformer's core area or other fault-prone areas, typically around the transformer's control panel, terminal blocks, or equipment joints. These areas are prone to mechanical vibration or electrical faults, which can generate acoustic signals. The sensor array should be positioned to capture all possible sound source signals while suppressing ambient noise. The setting of the sound source focusing grid point plane needs to be flexibly adjusted according to the specific transformer fault detection requirements and the possible sound source location to ensure that the detection accuracy of the fault sound source can be maximized.

[0081] Step 2: construct a cross-spectrum matrix between acoustic sensors based on the preprocessed sound pressure signal, and perform eigenvalue decomposition and principal component extraction on the cross-spectrum matrix.

[0082] Preferably, in step 2, constructing the cross-spectral matrix includes:

[0083] Based on the frequency domain signal vector obtained after preprocessing Constructing the cross-spectral matrix between acoustic sensors :

[0084]

[0085] in, is the average number of frames, used to smooth random noise; is the frequency domain signal vector of the nth frame; for The conjugate transpose of .

[0086] Preferably, in step 2, performing eigenvalue decomposition and principal component extraction on the cross-spectral matrix includes:

[0087] The diagonal elements of the cross-spectrum matrix are the eigenvalues arranged in order from large to small, and the largest first 20 eigenvalues are selected to form the cross-spectrum matrix .

[0088] Specifically, the cross-spectral matrix Perform eigenvalue decomposition to obtain the product of three matrices: an orthogonal matrix consisting of eigenvectors, a diagonal matrix whose diagonal elements are the eigenvalues in descending order, and the conjugate transpose of the orthogonal matrix:

[0089]

[0090] in, The diagonal elements of are the eigenvalues in descending order. From the decomposed eigenvalues, select the top 20 eigenvalues with the largest values. , and extract its corresponding 20 feature vectors , multiply the outer product matrix of each eigenvalue and its corresponding eigenvector, and perform weighted summation on all 20 outer product matrices to obtain the cross-spectral matrix after dimensionality reduction :

[0091] .

[0092] Step 3: Calculate the orthogonal beam output based on the cross-spectral matrix after principal component extraction to generate a preliminary sound source map, and screen the grid points in the preliminary sound source map based on a preset sound source intensity threshold.

[0093] Preferably, in step 3, calculating the orthogonal beam output based on the cross-spectral matrix after principal component extraction includes:

[0094] Step 3.1: discretize the measurement surface into a grid surface containing M grid points, and define the sound source calculation plane based on the grid surface;

[0095] Specifically, the measurement surface is discretized into a grid surface containing M grid points, and the grid point set is expressed as , the sound source calculation plane contains M grid points, and the focusing plane is defined for each grid point. The focusing vector of the mth grid point is , ;

[0096] Step 3.2, calculate the orthogonal beam output of each grid point in the sound source calculation plane to generate a preliminary sound source map;

[0097] Specifically, the orthogonal beam output of the mth grid point is calculated as:

[0098]

[0099] in, is the cross-spectral matrix obtained by principal component extraction in step 2, for The conjugate transpose of

[0100] Step 3.3, screening the grid points in the preliminary sound source map according to a preset sound source intensity threshold;

[0101] Specifically, according to the preset sound source intensity threshold Filter the grid points, the filtering conditions are:

[0102]

[0103] Retain the sound source intensity above the threshold The grid points of are taken as valid grid points;

[0104] The orthogonal beam output of the grid points that do not meet the screening conditions is assigned to , ;

[0105] It should be noted that the sound source intensity threshold The value range is usually set between 10% and 30% of the lower limit of the signal dynamic range, and the sound source intensity threshold The size of OB-CLEAN-SC has an impact on the sound source intensity threshold. When the sound source intensity threshold is larger, the main lobe width will become narrower, thereby improving the spatial resolution, but it may cause some low-energy effective signals to be ignored; when the sound source intensity threshold is When the value is smaller, the main lobe width becomes wider, which can capture more low-energy signals, but it will reduce the spatial resolution and may cause the side lobe suppression effect to deteriorate; therefore, the sound source intensity threshold The setting of the RF receiver needs to strike a balance between the main lobe width and the side lobe suppression capability to ensure the accurate positioning of the fault sound source.

[0106] Further preferably, the sound source intensity threshold It can be dynamically adjusted by the adaptive threshold algorithm based on the statistical characteristics of the sound source map. When screening the grid points of the preliminary sound source map, the median and interquartile range of the intensity values of all grid points are calculated, and the dynamic threshold formula is used to calculate the median and interquartile range of the intensity values of all grid points. ,in is the median of the grid point intensity values, is the interquartile range of intensity values at all grid points, and the coefficient It can be 1.5, covering 95% of the normal distribution data; represents the baseline intensity of background noise, Reflects the noise fluctuation range. When the fault signal intensity is significantly higher than the background noise (IQR increases), Automatically raises to focus on valid signals.

[0107] Step 4: Use the CLEAN-SC algorithm to iteratively clarify the filtered sound source map. When the iteration termination condition is met, the clear sound source localization map generated after the iterative clarity beam processing is output.

[0108] Preferably, the iterative clearing process of the filtered sound source map using the CLEAN-SC algorithm includes:

[0109] (1) Initialize the sound source map

[0110] The filtered sound source map is used as the initial sound source map , the initial value of the cross-spectral matrix is .

[0111] (2) Main lobe peak search and direction / weight vector calculation

[0112] In the i-th iteration, in the sound source map Search for the position of the main lobe peak in :

[0113]

[0114] This position is the position of the strongest sound source in the current iteration;

[0115] calculate Direction vector :

[0116]

[0117] in, It is i The search for the main lobe peak position in the iteration With the n microphones X n the distance between them; It is n The position of the microphone; k is the wave number;

[0118] calculate The weight vector :

[0119] .

[0120] (3) Cross-spectral matrix update

[0121] Remove the contribution of the current main lobe peak from the cross-spectral matrix and update the remaining sound source information;

[0122] The updating of the cross-spectral matrix of the residual sound source information includes:

[0123] The CLEAN-SC algorithm assumes that in the i-th iteration is composed of a single coherent source component The cross-spectral matrix caused by

[0124]

[0125] in, is the beamforming output corresponding to the main lobe peak in the i-1th iteration; is the focusing vector corresponding to a single coherent source component in the i-th iteration; for The conjugate transpose of

[0126]

[0127] according to , update the cross-spectral matrix:

[0128]

[0129] in, is the cross-spectral matrix at the i-th iteration, which is used to store the updated sound source information during the iteration process. It is the weight coefficient for adjusting the attenuation ratio of the sound source component during the iteration process.

[0130] (4) Update the sound source localization map based on the direction vector and weight vector

[0131] Subtract in the i-1 iteration Contribution of the peak source: In the i-1 iteration, although we have not yet calculated the result of the i+1 iteration, we can gradually subtract the contribution of the identified sound source through the design idea of the algorithm. Specifically, in each iteration, the algorithm will calculate the sound source information identified in the current iteration (i.e., the i iteration) and update the sound source map based on this information. In the i-1 iteration, the contribution of the identified sound source will be retained and gradually eliminated by optimizing the sound source map in the i iteration. For the part that has not yet entered the i+1 iteration, although it is not directly involved in the calculation, based on the sound source map of the previous round (i.e., the result of the i-1 iteration), the approximate range of the current sound source position can be estimated, and by calculating the relative influence of the current sound source and the known sound source, the contribution of the identified sound source can be subtracted. Reasonable speculation and gradual elimination of unnecessary contributions ensure the stable convergence of the algorithm. Therefore, those that are not affected by this peak for:

[0132]

[0133] in, is the weighted vector of the grid points that meet the screening conditions, then:

[0134]

[0135]

[0136] Where, The direction vector of the grid points that meet the filtering conditions.

[0137] (5) Store the clear beam obtained in each iteration

[0138] Store the sharpened beam obtained at each iteration :

[0139]

[0140] Where, and are the key parameters, among which Usually represents the gain factor, which is used to adjust the strength of the signal. Its value is dynamically adjusted according to the characteristics of the signal, the noise environment, and the algorithm goal. It is usually set by minimizing the error or using an adaptive algorithm. It is the phase angle, which represents the phase characteristics of the signal. Its value is based on the relative position of the sensor array and the direction of signal arrival. It is usually determined by measuring the phase difference of the signal, and its value range is generally . and The specific value of can be optimized through experiments or signal processing methods (such as minimum mean square error) to ensure accurate signal alignment and efficient beamforming.

[0141] (6) Iteration termination condition

[0142] According to the criteria , to determine whether the algorithm meets the conditions for iterative termination, the final output result is:

[0143]

[0144] It is worth noting that the OB-CLEAN-SC algorithm removes irrelevant noise in each round of calculation through multiple iterations and optimization of the sound source map, and gradually improves the accuracy of sound source positioning based on the characteristics and positions of known signal sources. In the final output sound source positioning map, grid points located near the sound source location will be displayed as high intensity values, indicating that there may be fault sources in these areas. In this way, based on the spatial distribution of intensity values, the system can clearly display the fault source in the transformer and help operation and maintenance personnel accurately locate the specific location of the fault source. In this way, OB-CLEAN-SC not only improves the accuracy of fault sound source positioning, but also can effectively distinguish between the main fault sound source and the secondary noise source, avoid misjudgment and missed judgment, and provide reliable technical support for subsequent maintenance work.

[0145] Step 5: In the clear sound source localization map generated in step 4, mark the area where the intensity is higher than the set threshold, analyze the location of the fault source in combination with the physical structure and operating status of the transformer under test, and generate a fault location report.

[0146] Preferably, the step 5 includes:

[0147] Step 5.1, identify the potential fault source location by analyzing the high-intensity area in the sound source map;

[0148] Areas of higher intensity usually correspond to fault sources inside or around the transformer;

[0149] For example, a dynamic threshold is set, usually 20% to 40% of the maximum intensity in the sound source map, which can be adjusted according to the noise background of the transformer; continuous areas with intensity greater than the threshold in the sound source map are automatically identified through image processing algorithms (such as region growing method and morphological filtering); the grid coordinate system of the sound source map is aligned with the three-dimensional CAD model or physical structure diagram of the transformer under test, and a mapping relationship between the sound source grid points and the actual components is established to determine the actual components corresponding to the high-intensity areas (such as windings, cores, bushings, etc.).

[0150] Step 5.2: Determine the specific location of the fault source based on the intensity distribution and direction vectors in the sound source map, combined with the physical structure of the transformer. For each grid point with high intensity, calculate its associated direction vector and weight vector to further improve positioning accuracy.

[0151] For example, for each high-intensity grid point, its direction vector and weight vector are calculated. Based on the calculation results, coherence analysis is performed on adjacent high-intensity areas to determine whether the fault source is diffuse or independent. Then, they are sorted by intensity, the processing priority is clarified, and the scope of the fault source is narrowed.

[0152] Step 5.3: Generate a fault source location report. The report includes the coordinates of the sound source, the sound source intensity, the possible fault type and its location, to help maintenance personnel quickly identify the fault source and repair it.

[0153] Specifically, the precise location of the fault source is determined based on the high-intensity areas in the sound source map. This location is further confirmed based on the transformer's operating status and structural characteristics. The localization results are verified to ensure the accuracy of the identified fault source location. If necessary, subsequent remedial measures are deployed.

[0154] It should be noted that each grid point in the sound source localization map represents the intensity of a sound source, and areas of higher intensity typically indicate the location of the fault source. Areas of higher intensity can be identified through automatic recognition or manual inspection by maintenance personnel. These areas typically correspond to potential fault locations within the transformer. The precise location of high-intensity areas often occurs in key transformer components, such as terminal blocks, transformer joints, or other potentially faulty components. In the sound source localization map, areas of lower intensity may represent irrelevant noise or background signals. Therefore, a suitable threshold γ is required to filter out meaningful signals. By selecting a reasonable threshold, only sound sources with intensities above the threshold are retained for further analysis. This screening step effectively prevents irrelevant noise signals from affecting localization accuracy and reduces computational complexity. After filtering out high-intensity areas, the next step is to fine-tune the sound source distribution in these areas. Using the specific intensity of each grid point in the sound source localization map, maintenance personnel can pinpoint the approximate location of the fault source. Furthermore, by combining the physical structure and operating status of the transformer, it is possible to determine whether the sound source originates from a potential fault point. During the localization process, interference from ambient noise or other irrelevant sources may be encountered. By further clarifying the sound source map, redundant signals and noise are suppressed, further improving the reliability of the localization results. During each iteration, the algorithm optimizes the location of the identified sound source. Repeated iterations further improve the accuracy of the sound source map. Each iteration removes the contribution of identified sound sources and adjusts the location of the remaining sound sources, ultimately refining the location of the fault source. After multiple iterations, the final sound source map clearly identifies the fault source location inside or around the transformer. Areas of high intensity reflect the most likely fault source, and these locations are key areas for fault repair and maintenance. The final sound source map not only provides a precise spatial location, but once the fault source is located, the system can generate a detailed fault location report. This report includes the specific location of the fault source, its intensity, how it compares to background noise, and other possible fault-related information.

[0155] In embodiment 2 of the present invention, a transformer fault sound source localization system based on orthogonal clear beams is provided. Based on the transformer fault sound source localization method based on orthogonal clear beams described in embodiment 1 of the present invention, the system includes:

[0156] The sensor array module is used to arrange acoustic sensors in a three-dimensional array and collect transformer sound pressure signals in real time;

[0157] Signal preprocessing module, used to process the original sound pressure signal, including framing, windowing, Fourier transform and filtering noise reduction;

[0158] Cross-spectral matrix calculation module, used to construct the cross-spectral matrix and extract the principal component features;

[0159] Orthogonal beamforming module, used to generate preliminary sound source maps and screen effective grid points;

[0160] Iterative clarification module, used to iteratively optimize the sound source localization map using the OB-CLEAN-SC algorithm;

[0161] Fault analysis module, used to analyze the location and priority of fault sources in relation to transformer structure;

[0162] The report generation module is used to generate a fault source location report.

[0163] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A transformer fault sound source localization method based on orthogonal clarity beam, characterized in that: The steps include: With the center of the transformer under test as the coordinate origin, N acoustic sensors are arranged in an array in the three-dimensional space around the transformer to form a three-dimensional sound field measurement surface, and the sound pressure signal is collected and pre-processed; Based on the preprocessed sound pressure signal, the cross-spectral matrix between the acoustic sensors is constructed, and the eigenvalue decomposition of the cross-spectral matrix is performed. The first K largest eigenvalues and their corresponding eigenvectors are selected to reconstruct the dimension-reduced cross-spectral matrix. The orthogonal beam output is calculated by combining the dimension-reduced cross-spectral matrix to generate a preliminary sound source map. The grid points in the preliminary sound source map are screened by combining the preset sound source intensity adaptive dynamic threshold. The screened sound source map is iteratively clarified using the CLEAN-SC algorithm. When the iterative termination condition is met, the sound source localization map after iteratively clarified beam processing is output. The area with intensity higher than the set threshold is marked in the sound source localization map, and the location of the fault source is analyzed based on the physical structure and operating status of the tested transformer to generate a fault location report.

2. The transformer fault sound source localization method based on orthogonal clarity beam according to claim 1 is characterized by: The collecting of the sound pressure signal and preprocessing thereof include: The sound pressure signal is framed and windowed, and each frame of the time domain signal is converted into a frequency domain signal through Fourier transform to obtain the frequency domain signal vector , and perform filtering and noise reduction.

3. The transformer fault sound source localization method based on orthogonal clarity beam according to claim 2 is characterized by: The constructing of a cross-spectrum matrix between acoustic sensors based on the preprocessed sound pressure signal includes: Based on the frequency domain signal vector obtained after preprocessing Constructing the cross-spectral matrix between acoustic sensors : in, is the average number of frames; is the frequency domain signal vector of the nth frame; for The conjugate transpose of Cross-spectral matrix Perform eigenvalue decomposition to obtain the product of three matrices: an orthogonal matrix consisting of eigenvectors, a diagonal matrix whose diagonal elements are eigenvalues arranged in descending order, and the conjugate transpose of the orthogonal matrix. From the decomposed eigenvalues, select the first 20 eigenvalues with the largest values and extract their corresponding 20 eigenvectors. Multiply the outer product matrix of each eigenvalue with its corresponding eigenvector, and perform weighted summation on all 20 outer product matrices to obtain the cross-spectral matrix after dimensionality reduction. .

4. The transformer fault sound source localization method based on orthogonal clarity beam according to claim 3 is characterized by: The calculating of the orthogonal beam output in combination with the dimension-reduced cross-spectral matrix includes: The measurement surface is discretized into a grid surface containing M grid points, and the sound source calculation plane is defined based on the grid surface; Calculate the orthogonal beam output of each grid point in the sound source calculation plane to generate a preliminary sound source map; The grid points in the preliminary sound source map are screened according to the preset sound source intensity threshold.

5. The transformer fault sound source localization method based on orthogonal clarity beam according to claim 4 is characterized in that: Discretizing the measurement surface into a grid surface including M grid points and generating a preliminary sound source map comprises: The measurement surface is discretized into a grid surface containing M grid points. The grid point set is , the sound source calculation plane contains M grid points, and the focusing plane is defined for each grid point. The focusing vector of the mth grid point is , ; The orthogonal beam output at the mth grid point is: in, is the dimension-reduced cross-spectral matrix, for The conjugate transpose of .

6. The transformer fault sound source localization method based on orthogonal clarity beam according to claim 5 is characterized by: The iterative clarification processing of the filtered sound source image using the CLEAN-SC algorithm includes: The filtered sound source map is set as the initial sound source map, and the dimension-reduced cross-spectrum matrix is used as the initial value of the cross-spectrum matrix; In the i-th iteration, the main lobe peak position is searched from the current sound source map as the strongest sound source, and the direction vector and weight vector are calculated based on this position; According to the coherent source component contribution of the current main lobe peak, the component is removed from the cross-spectrum matrix, and the cross-spectrum matrix of the remaining sound source information is updated; Based on the direction vector and the weight vector, the contribution of the current main lobe peak is removed from the sound source map to generate an updated sound source localization map; storing the sharpened beam data determined by the gain factor and phase angle parameters in each iteration; When the preset criteria are met, the iteration is terminated and the sound source localization map after iteratively clarified beam processing is output.

7. The transformer fault sound source localization method based on orthogonal clarity beam according to claim 6 is characterized by: The main lobe peak search and direction / weight vector calculation include: In the i-th iteration, in the sound source map Search for the position of the main lobe peak in : calculate Direction vector : in, is the search main lobe peak position for the i-th iteration The distance to the nth microphone; is the position of the nth microphone; calculate The weight vector : 。 8. The transformer fault sound source localization method based on orthogonal clarity beam according to claim 7 is characterized in that: The updating of the cross-spectral matrix of the residual sound source information includes: Assume that in the i-th iteration is composed of a single coherent source component The resulting cross-spectral matrix: in, is the beamforming output corresponding to the main lobe peak in the i-1th iteration; is the focusing vector corresponding to a single coherent source component in the i-th iteration; for The conjugate transpose of according to , update the cross-spectral matrix: in, is the cross-spectral matrix at the i-th iteration, which is used to store the updated sound source information during the iteration process. It is the weight coefficient for adjusting the attenuation ratio of the sound source component during the iteration process.

9. The transformer fault sound source localization method based on orthogonal clarity beam according to claim 8 is characterized in that: Analyzing the location of the fault source based on the physical structure and operating status of the transformer under test and generating a fault location report includes: Based on a dynamic threshold, continuous areas with intensities above the threshold are extracted from the clear sound source localization map. Potential fault sources are identified through image processing algorithms. The sound source grid coordinate system is aligned with the three-dimensional physical structure model of the transformer to establish a mapping relationship between the fault area and the actual component. Combined with the sound source intensity distribution, direction vector, weighted vector, and physical structure of the transformer, the system performs location optimization and coherence analysis on areas with intensity above the threshold to determine the location and priority of the fault source. The output includes a location report of the fault source coordinates, intensity, associated components, and fault type. The location results are verified in combination with the transformer operating status, and repair measures are deployed.

10. A transformer fault sound source localization system based on orthogonal clear beams, based on a transformer fault sound source localization method based on orthogonal clear beams according to any one of claims 1 to 9, characterized in that: The system includes: The sensor array module is used to arrange acoustic sensors in a three-dimensional array and collect transformer sound pressure signals in real time; Signal preprocessing module, used to process the original sound pressure signal, including framing, windowing, Fourier transform and filtering noise reduction; Cross-spectral matrix calculation module, used to construct the cross-spectral matrix and extract the principal component features; Orthogonal beamforming module, used to generate preliminary sound source maps and screen effective grid points; Iterative clarification module, used to iteratively optimize the sound source localization map using the OB-CLEAN-SC algorithm; Fault analysis module, used to analyze the location and priority of fault sources in relation to transformer structure; The report generation module is used to generate a fault source location report.

Images