A fault diagnosis method and device, equipment and medium for a distribution transformer

Abstract

The application provides a power distribution transformer fault diagnosis method and device, equipment and medium, and belongs to the technical field of fault diagnosis. The method comprises the following steps: acquiring current image data and current acoustic data of a power distribution transformer; inputting the current image data into a pre-trained target detection model for processing to obtain position information of a potential fault area in the power distribution transformer; determining a spatial position corresponding to the potential fault area based on pixel coordinates and a preset mapping relationship between the pixel coordinates and spatial coordinates; obtaining acoustic feature information corresponding to the potential fault area based on the spatial position and the current acoustic data; and obtaining a fault diagnosis result of the power distribution transformer based on the acoustic feature information and a preset fault criterion. The power distribution transformer fault diagnosis method and device, equipment and medium provided by the application can improve the accuracy and reliability of power distribution transformer fault diagnosis.

Description

Technical Field

[0001] This application belongs to the field of fault diagnosis technology, and more specifically, it relates to a fault diagnosis method, device, equipment, and medium for distribution transformers. Background Technology

[0002] As a critical piece of equipment at the end of the power system, distribution transformers are subjected to multiple stresses from electromagnetic, thermal, and mechanical sources over long periods, making them prone to faults such as partial discharge, insulation degradation, and core loosening. Failure to identify and locate these faults promptly and accurately threatens power supply safety and equipment lifespan. Currently, fault diagnosis for distribution transformers largely relies on a single information source. However, in actual operating environments, distribution transformers often face complex factors such as electromagnetic interference, environmental noise, and multi-source signal coupling. A single information source cannot simultaneously ensure accurate fault location and reliable identification of fault characteristics. This can easily lead to misinterpreting interference signals from non-fault areas as fault features, or missing fault signals due to background noise, significantly limiting the accuracy of diagnostic results. Therefore, improving the accuracy and reliability of distribution transformer fault diagnosis remains a pressing technical problem to be solved in this field. Summary of the Invention

[0003] The purpose of this application is to provide a method, apparatus, equipment, and medium for diagnosing faults in distribution transformers, which can improve the accuracy and reliability of fault diagnosis. To achieve the above objective, the technical solution provided by this application is as follows: Firstly, a fault diagnosis method for distribution transformers is provided, including: Acquire the current image data and current acoustic data of the distribution transformer; The current image data is input into a pre-trained target detection model for processing to obtain the location information of potential fault areas in the distribution transformer; the location information includes the pixel coordinates of the potential fault areas in the current image data. Based on pixel coordinates and the preset mapping relationship between pixel coordinates and spatial coordinates, the spatial location corresponding to the potential fault area is determined. Based on spatial location and current acoustic data, and processed by a delay summation beamforming algorithm, acoustic feature information corresponding to potential fault areas is obtained. Based on acoustic feature information and preset fault criteria, the fault diagnosis results of the distribution transformer are obtained.

[0004] Secondly, a fault diagnosis device for a distribution transformer is provided, comprising: The data acquisition module is used to acquire the current image data and current acoustic data of the distribution transformer; The visual recognition module is used to input the current image data into a pre-trained target detection model for processing to obtain the location information of potential fault areas in the distribution transformer; wherein, the location information includes the pixel coordinates of the potential fault areas in the current image data; The coordinate transformation module is used to determine the spatial location of potential fault areas based on pixel coordinates and the preset mapping relationship between pixel coordinates and spatial coordinates. The acoustic processing module is used to obtain the acoustic feature information corresponding to the potential fault area based on the spatial location and current acoustic data, and through the delay summation beamforming algorithm. The fault diagnosis module is used to obtain fault diagnosis results of distribution transformers based on acoustic feature information and preset fault criteria.

[0005] Thirdly, embodiments of this application also provide an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement a fault diagnosis method for a distribution transformer provided in any possible implementation of the first aspect.

[0006] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements a fault diagnosis method for a distribution transformer provided by any possible implementation of the first aspect.

[0007] The beneficial effects of the technical solution provided in this application are as follows: This application provides a fault diagnosis method, apparatus, equipment, and medium for distribution transformers. Compared with related technologies, this application acquires current image data and current acoustic data of the distribution transformer, uses a pre-trained target detection model to quickly identify potential fault areas from the image and extract pixel coordinates, then converts them into spatial locations based on a preset mapping relationship. Using this spatial location as the focal point, a delayed summation beamforming algorithm is employed to enhance directional acoustic signals and extract features, combined with the current acoustic data. Finally, a diagnostic result is obtained based on acoustic feature information and preset fault criteria. This application first uses visual information to achieve preliminary spatial positioning of the fault area, and then uses this positioning as a guide to perform spatially selective focusing processing on the acoustic data. This effectively suppresses background noise interference from non-fault areas, improves the signal-to-noise ratio of acoustic features and the accuracy of fault identification, and achieves synergistic complementarity between visual positioning and acoustic diagnosis, thereby improving the accuracy and reliability of distribution transformer fault diagnosis. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0009] Figure 1 A flowchart illustrating a fault diagnosis method for a distribution transformer provided in an embodiment of this application; Figure 2 A structural block diagram of a fault diagnosis device for a distribution transformer provided in an embodiment of this application; Figure 3 A schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0010] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0011] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” can be implemented as “A,” or as “B,” or as “A and B.” When describing multiple (two or more) items, if the relationship between the multiple items is not explicitly defined, the multiple items can refer to one, several or all of the multiple items. For example, the description of "parameter A includes A1, A2, A3" can be implemented as parameter A includes A1 or A2 or A3, or it can be implemented as parameter A includes at least two of the three items A1, A2 and A3.

[0012] This application provides a fault diagnosis method for distribution transformers, applied to online monitoring and fault identification scenarios for dry-type distribution transformers. Due to their oil-free structure, dry-type transformers are typically deployed in indoor distribution rooms, underground utility tunnels, or prefabricated substations. Their windings, cores, and insulation components are constantly exposed to air, making them susceptible to temperature and humidity changes, vibration stress, and localized electric field concentrations, which can lead to faults such as partial discharge, core loosening, and inter-turn short circuits in the windings.

[0013] This method can be executed by an electronic device, such as... Figure 1 As shown, the method may include: S101: Obtain the current image data and current acoustic data of the distribution transformer.

[0014] In this embodiment, the current image data refers to the image information acquired by the image acquisition device at the diagnostic time, which reflects the appearance and thermal distribution of the dry-type transformer. The current acoustic data refers to the sound signals generated during the operation of the dry-type transformer acquired by the acoustic sensor array at the diagnostic time, which may include electromagnetic humming and core vibration sounds during normal operation, as well as discharge pulse sounds and abnormal mechanical collision sounds that may occur under fault conditions.

[0015] In one embodiment of this application, acquiring current image data and current acoustic data of a distribution transformer includes: Acquire visible light image data and infrared thermal imaging data of the distribution transformer acquired by a dual-spectrum image acquisition device; Based on the common feature points in visible light image data and infrared thermal imaging data, the infrared thermal imaging data and visible light image data are spatially registered to obtain the registered infrared thermal imaging data. The spatial registration is based on the pixel coordinate system of the visible light image data. The common feature points are the matching feature point pairs between the feature points in the visible light image data and the feature points in the infrared thermal imaging data whose feature descriptors have a similarity greater than a preset matching threshold. The registered infrared thermal imaging data and visible light image data are fused at the pixel level to obtain fused image data; the pixel coordinate system of the fused image data is consistent with the pixel coordinate system of the visible light image data; the fused image data is used as the current image data. The sound signals of the power distribution transformer during operation are acquired by a microphone array composed of multiple acoustic sensors to obtain the current acoustic data; the spatial coordinates of each acoustic sensor in the microphone array are pre-calibrated.

[0016] In this embodiment, the dual-spectrum image acquisition device refers to a composite imaging device that integrates a visible light camera and an infrared thermal imager into the same bracket or housing. The optical axes of the two imaging channels are approximately parallel after mechanical calibration, enabling the simultaneous acquisition of visible light and infrared thermal images of the target. Spatial registration refers to the process of aligning two images from different imaging channels, resolutions, or fields of view to the same pixel coordinate system through geometric transformation. Common feature points refer to image points with stable grayscale or texture characteristics that can be detected in both visible light and infrared thermal images, such as bolt holes on a dry-type transformer casing, the corners of heat dissipation fins, or the corner points of a nameplate border. Feature descriptors are mathematical codes for the grayscale distribution patterns within the neighborhood of feature points, used to measure the similarity between two feature points. Pixel-level fusion refers to the operation of combining corresponding pixels from two images into a single image according to certain fusion rules after spatial registration.

[0017] In this embodiment, the image acquisition device adopts a dual-spectrum structure. The visible light channel has a resolution of 1920×1080 pixels and a frame rate of 25 frames per second; the infrared thermal imaging channel has a resolution of 640×480 pixels and a frame rate of 25 frames per second, with a thermal sensitivity better than 50mK and a temperature measurement range of -20℃ to 300℃. The acquisition device can be installed approximately 1.5 meters in front of the dry-type transformer using an L-shaped mounting bracket. The bottom of the bracket is fixed to the floor of the power distribution room with expansion bolts to avoid image jitter caused by operational vibrations.

[0018] After acquiring the visible light image and infrared thermal image, spatial registration is first performed. The registration process is as follows: Feature points are extracted from the visible light image and infrared thermal image respectively. In this embodiment, a feature extraction method based on scale-invariant feature transform is used. Approximately 320 feature points are detected in the visible light image and approximately 210 feature points are detected in the infrared thermal image. Then, the Euclidean distance between the feature descriptors of each pair of feature points is calculated. Feature point pairs with an Euclidean distance less than a preset matching threshold are identified as common feature points. In this embodiment, the preset matching threshold is set to 0.75 times the maximum possible distance of the feature descriptor vectors, specifically 0.75 × 255 ≈ 191 (when the descriptor is a 128-dimensional vector with each dimension ranging from 0 to 255 integers, the maximum Euclidean distance is approximately...). Therefore, the threshold is set to approximately 2172 (2896 × 0.75 ≈ 2896 × 0.75). This threshold was determined statistically from a registration experiment of 50 pairs of dual-spectral images of dry-type transformers: when the threshold is 0.70 to 0.80 times the maximum distance, the overall effect of the correct matching rate and the number of matching points is optimal. After matching, this embodiment can obtain 35 to 50 pairs of common feature points. Based on these pairs of common feature points, a random sampling consensus algorithm is used to estimate the homography transformation matrix from the infrared thermal imaging pixel coordinates to the visible light image pixel coordinates. This matrix is ​​a 3×3 matrix containing rotation, scaling, translation, and perspective components. Using this homography transformation matrix, the infrared thermal imaging data is mapped pixel by pixel to the pixel coordinate system of the visible light image to obtain the registered infrared thermal imaging data.

[0019] After spatial registration, pixel-level fusion is performed. The fusion rule adopts a weighted average method: for each pixel location in the registered infrared thermal imaging data, the sum of the visible light grayscale value multiplied by a weight of 0.6 and the infrared grayscale value (normalized to the range of 0 to 255) multiplied by a weight of 0.4 is taken as the output value of that pixel location in the fused image. The weights 0.6 and 0.4 are set based on the fact that surface faults (such as carbonization and discoloration) of dry-type transformers have higher contrast in the visible light channel, while internal thermal anomalies (such as local temperature rise caused by inter-turn short circuits) are more obvious in the infrared channel. Therefore, fusion is performed with visible light as the primary source and infrared as the secondary source. The fused image data has a resolution of 1920×1080 pixels and contains 3 channels (R, G, B). The infrared information has been integrated into the grayscale values ​​of each channel. This fused image data is used as the current image data.

[0020] Acoustic data acquisition is achieved using a microphone array. In this embodiment, the microphone array consists of eight acoustic sensors, N=8. The eight acoustic sensors are mounted in a regular octagonal layout on a ring-shaped support around the dry-type transformer. The ring-shaped support has a diameter of 1.2 meters, and its center is aligned with the geometric center of the transformer. The support is fixed to the transformer base. Each acoustic sensor is a broadband condenser microphone with a frequency response range of 20Hz to 100kHz and a sensitivity of -26dBV / Pa (i.e., outputting a 50mV electrical signal at 1Pa sound pressure). The spatial coordinates of the eight sensors are calibrated using a total station during system deployment, with a calibration accuracy better than ±1 mm. The calibration results are expressed in three-dimensional coordinates. The parameters are stored in the system's calibration parameter file in the form of [the parameter file]. After the acoustic signals are converted into analog electrical signals by various sensors, they are digitized by a 24-bit precision analog-to-digital converter chip at a sampling rate of 200kHz. Each acquisition lasts for 2 seconds, resulting in an acoustic data matrix with 8 channels and 400,000 sampling points. This data matrix serves as the current acoustic data.

[0021] As can be seen from the above, this embodiment integrates the visible light texture information and thermal distribution information of the dry-type transformer surface into the same pixel coordinate system through dual-spectral fusion imaging. This provides richer input features for the target detection model, enabling it to not only identify visible faults such as carbonization and cracking, but also detect abnormal surface temperatures caused by internal local overheating, thus expanding the coverage of detectable faults. Simultaneously, the coordinates of each sensor in the microphone array are precisely calibrated using a total station, providing a reliable spatial reference for subsequent beamforming delay calculations, ensuring the accuracy of acoustic positioning from the data source.

[0022] S102: Input the current image data into a pre-trained target detection model for processing to obtain the location information of potential fault areas in the distribution transformer; wherein, the location information includes the pixel coordinates of the potential fault areas in the current image data.

[0023] In this embodiment, the target detection model refers to a deep neural network model trained on a large number of normal and faulty image samples of dry-type transformers. Its function is to identify and locate areas where faults may exist from the input image. Pixel coordinates refer to two-dimensional coordinates in the image, with the top left corner as the origin and in pixels as the unit.

[0024] In this embodiment, the target detection model can adopt a single-stage detection architecture based on convolutional neural networks, with the current image data as input. During the training phase, the target detection model has learned a large number of image features of dry-type transformers under normal conditions and various fault conditions (including but not limited to carbonization marks on the winding surface, local overheating and discoloration of the iron core, bulging and cracking of the insulation layer, and burning of the terminals). It can perform forward calculation on the input image during the inference phase and output several detection boxes, each represented by pixel coordinates.

[0025] Specifically, the position information of each detection box is expressed in the top-left pixel coordinates. and bottom right pixel coordinates The results are given in the form of a confidence score and a fault category prediction label. In this embodiment, only detection boxes with a confidence score greater than 0.6 are retained as valid detection results. If there is no detection box with a confidence score greater than 0.6 in a frame, it is determined that no potential fault area was detected in the current frame, and the next frame is analyzed.

[0026] In one embodiment of this application, the training process of the pre-trained object detection model includes: Acquire sample image data of the distribution transformer; the sample image data includes sample images under normal conditions and sample images under various fault conditions; The fault regions in the sample image data are labeled to obtain the labeling information, which includes the bounding box coordinates of the fault regions and the fault category label. Based on sample image data and annotation information, the preset initial object detection network is iteratively trained. In each iteration, the loss function value is calculated based on the prediction output and annotation information of the initial object detection network, and the network parameters of the initial object detection network are updated based on the loss function value until the loss function value converges to the preset threshold, thus obtaining the pre-trained object detection model.

[0027] In this embodiment, sample image data refers to the image dataset used to train the target detection model, including images of dry-type transformers under normal conditions and various simulated or real fault conditions. Annotation information refers to label data describing the location and category of the fault area, marked on the sample images manually or semi-automatically. Bounding box coordinates refer to the four coordinate values ​​that define the fault area in the form of a rectangle, typically the pixel coordinates of the upper left and lower right corners. Fault category labels refer to the name of the fault type corresponding to the bounding box, which in this embodiment includes five categories: "discharge traces on winding surface," "local overheating of iron core," "insulation layer damage," "terminal erosion," and "foreign object attachment." The initial target detection network refers to a deep learning model with a preset network structure that has not yet been trained for a specific task. The loss function value is a measure of the difference between the model's predicted output and the actual annotation; the smaller the value, the closer the model's prediction is to the actual annotation.

[0028] In this embodiment, a total of 1200 images were collected from the operating sites of dry-type distribution transformers at three different voltage levels (6kV, 10kV, and 35kV) under normal operating conditions. From a power equipment testing laboratory, five types of fault samples were artificially simulated: surface discharge of windings (generating discharge carbonization marks on the winding insulation surface using a high-voltage corona generator), localized overheating of the iron core (generating a temperature rise of 150°C to 280°C in a localized area of ​​the iron core using an induction heating device), insulation layer damage (creating damage of varying depths on the epoxy resin casting layer surface through mechanical scraping), terminal ablation (causing oxidation discoloration and pitting on the copper terminal surface through high-current impact), and foreign matter adhesion (placing metal debris or fibrous foreign matter on the winding surface). 600 images were collected for each type, totaling 3000 fault sample images. The total sample image data is 4200 images, each with a resolution of 1920×1080 pixels and a 24-bit true-color image format.

[0029] The annotation process combines manual annotation with semi-automatic assistance. Annotators use a bounding box tool to define the fault region on each sample image, with each fault region corresponding to a bounding box, and record the coordinates of its upper left corner. and the coordinates of the bottom right corner The images were then labeled with corresponding fault categories. For images with multiple fault areas, multiple bounding boxes were labeled for each area. After labeling, another reviewer cross-checked the results to ensure the bounding boxes matched the fault areas within 3 pixels. The final labeling file contained 4200 sample records, each including the image filename, bounding box coordinates, and fault category label.

[0030] The initial object detection network adopts a single-stage detection architecture, with a depthwise separable convolutional structure as the backbone, containing 53 convolutional layers. The input size is 640×640 pixels (the original images are scaled to this size during training). The training process is as follows: 4200 sample images are randomly divided into a training set (2940 images), a validation set (630 images), and a test set (630 images) in a ratio of 7:1.5:1.5. Training uses a mini-batch stochastic gradient descent optimization algorithm with a batch size of 16 and an initial learning rate of 0.01. Cosine annealing is used for learning rate scheduling, and the total number of training epochs is set to 300. The loss function consists of three parts: bounding box regression loss (using full intersection-over-union loss), object confidence loss (using binary cross-entropy loss), and class classification loss (using binary cross-entropy loss), which are weighted and summed in a 1:1:1 ratio. After each training round, the average precision on the validation set is calculated. When the improvement of the average precision on the validation set for 20 consecutive rounds is less than 0.001, the loss function value is determined to have converged to the preset threshold, training is stopped, and the current network parameters are saved as the pre-trained target detection model.

[0031] As can be seen from the above, this embodiment ensures the coverage and quality of training data by constructing a multi-category sample set covering typical fault types of dry-type transformers and employing a labeling strategy that combines manual annotation with cross-checking. The cosine annealing learning rate scheduling strategy enables the model to finely adjust network parameters with small step sizes in the later stages of training, avoiding oscillations near the optimal solution. This allows the model to be fully trained to detect subtle fault features on the surface of dry-type transformers (such as early discharge carbonization points and slight insulation bulges), laying the foundation for accurate localization of subsequent fault areas.

[0032] S103: Determine the spatial location of the potential fault area based on pixel coordinates and the preset mapping relationship between pixel coordinates and spatial coordinates.

[0033] In this embodiment, spatial position refers to the coordinates in real three-dimensional space obtained by restoring pixel coordinates through geometric mapping.

[0034] In one embodiment of this application, determining the spatial location corresponding to a potential fault region based on pixel coordinates and a preset mapping relationship between pixel coordinates and spatial coordinates includes: Acquire the internal and external parameters of the dual-spectrum image acquisition device; the internal parameters include the focal length and principal point coordinates, and the external parameters include the rotation matrix and translation vector of the dual-spectrum image acquisition device relative to the preset spatial coordinate system. Based on pixel coordinates, intrinsic parameters, and extrinsic parameters, construct a projection ray originating from the pixel coordinates; The intersection of the projected ray and the preset three-dimensional model of the distribution transformer is calculated, and the coordinates of the intersection point of the projected ray and the preset three-dimensional model are determined as the spatial location corresponding to the potential fault area; the spatial location is represented by three-dimensional coordinates in the preset spatial coordinate system.

[0035] In this embodiment, internal parameters refer to the optical characteristics of the image acquisition device itself, including focal length (expressed in pixels) and principal point coordinates (i.e., the position of the intersection of the optical axis and the imaging plane in the pixel coordinate system). External parameters refer to the pose parameters of the image acquisition device in three-dimensional space, including rotation matrix (a 3×3 orthogonal matrix describing the orientation of the device's coordinate axes relative to the preset spatial coordinate axes) and translation vector (a 3×1 vector describing the position of the device's optical center in the preset spatial coordinate system). The preset spatial coordinate system is a right-handed rectangular coordinate system defined artificially during the system deployment phase. In this embodiment, its origin is the geometric center of the dry-type transformer base, the X-axis points to the front of the transformer, the Y-axis points to the left side of the transformer, and the Z-axis points vertically upward, with units in millimeters. A projection ray is a semi-straight line extending from the optical center of the image acquisition device through the imaging point corresponding to the pixel coordinate into three-dimensional space. The preset three-dimensional model is the three-dimensional geometric model of the dry-type transformer, which in this embodiment is a surface mesh model composed of triangular facets.

[0036] In this embodiment, the calibration of internal and external parameters is completed during the system deployment phase. A checkerboard calibration board is used, with a checkerboard size of 9×6 squares, each square having a side length of 25 mm. The calibration board is placed in at least 20 different poses within the field of view of the image acquisition device, and a visible light image is acquired in each pose. The pixel coordinates of 54 corner points of the checkerboard in each image are extracted using a corner detection algorithm. Combined with the known 3D coordinates of the checkerboard, the Zhang Zhengyou calibration method is used to solve for the internal and external parameter matrices. In this embodiment, the calibrated internal parameter is: focal length. Pixels The pixel coordinates (slight differences in focal length in the X and Y directions are due to lens manufacturing tolerances) are (962.4, 538.7) pixels. The extrinsic parameters are: rotation matrix R and translation vector T, where the translation vector T = (1502.6, -18.3, 876.5) mm, indicating that the optical center of the image acquisition device is located in the preset spatial coordinate system at a position 1502.6 mm from the transformer center in the X direction, 18.3 mm to the left in the Y direction, and 876.5 mm above the origin in the Z direction.

[0037] The preset 3D model of the dry-type transformer is established in the following way: the transformer is scanned from all directions using a 3D laser scanner to obtain 3D point cloud data of the transformer surface (the point cloud density is about 20 points per square centimeter). After denoising, stitching and meshing, a surface mesh model composed of about 850,000 triangular facets is generated. This model is stored with a preset spatial coordinate system as the reference coordinate system.

[0038] For the pixel coordinates representing the potential fault area The spatial location calculation process is as follows: First, the pixel coordinates are converted into normalized direction vectors in the coordinate system of the image acquisition device using the inverse of the intrinsic parameter matrix. ,in =962.4、 =538.7 is the principal point coordinate. Then, the direction vector d is transformed from the device coordinate system to the preset spatial coordinate system using the rotation matrix R, to obtain the spatial direction vector. Finally, the position of the optical center of the device in the preset spatial coordinate system is determined. Starting point For the direction, construct parameterized ray equations. Substituting the ray equation into the plane equation of each triangular facet of the preset 3D model, the parameter t is solved. If t > 0 and the intersection point is located inside the triangular facet (determined by the centroid coordinate method), then the intersection point is the point where the ray intersects the model surface. In this embodiment, for 850,000 triangular facets, a hierarchical bounding box acceleration structure is used to control the time consumption of the intersection calculation to within 2 milliseconds. The final obtained 3D coordinates of the intersection point are... This refers to the spatial location of the potential fault area. For example, during a diagnostic test, a suspected discharge area was detected on the upper part of the A-phase winding of a transformer, with representative pixel coordinates (1023, 387). The spatial location calculated above is... = mm.

[0039] As can be seen from the above, this embodiment, by combining camera calibration with 3D model intersection, accurately restores the 2D pixel coordinates on the image plane to the 3D spatial coordinates on the surface of the dry-type transformer, realizing the transformation from position on the image to position in physical space. The use of a hierarchical bounding box acceleration structure makes the intersection operation extremely time-efficient, meeting the real-time requirements of online diagnosis.

[0040] S104: Based on spatial location and current acoustic data, and processed by a delay summation beamforming algorithm, acoustic feature information corresponding to potential fault areas is obtained.

[0041] In this embodiment, the delay summation beamforming algorithm is an array signal processing algorithm. Its basic principle is: for a certain focal point in space, calculate the difference in sound wave propagation distance from that point to each sensor in the array, and apply corresponding time delay compensation to each sensor signal accordingly, so that the sound wave signals from the direction of the focal point are aligned in time. Then, the compensated signals are superimposed and summed to enhance the signal from that direction and suppress interference from other directions.

[0042] In one embodiment of this application, acoustic feature information corresponding to a potential fault region is obtained based on spatial location and current acoustic data, and processed using a delay-sum beamforming algorithm, including: Obtain the sensor spatial coordinates of each acoustic sensor in the microphone array in a preset spatial coordinate system; the microphone array includes N acoustic sensors, where N is an integer greater than or equal to 2; Based on the spatial location and the sensor spatial coordinates of each acoustic sensor, calculate the propagation distance between each acoustic sensor and the spatial location of the potential fault area. Divide each propagation distance by the preset sound wave propagation speed to obtain the delay time corresponding to each acoustic sensor; the unit of propagation distance is meters, the unit of sound wave propagation speed is meters per second, and the unit of delay time is seconds; Based on the delay time corresponding to each acoustic sensor, the acoustic signal collected by the acoustic sensor in the current acoustic data is delayed and compensated to obtain the delayed compensated acoustic signal corresponding to each acoustic sensor. The beamforming output signal is obtained by superimposing and summing the N delayed-compensated acoustic signals. Feature extraction is performed on the beamforming output signal to obtain the acoustic feature information corresponding to the potential fault area.

[0043] In this embodiment, sensor spatial coordinates refer to the three-dimensional position coordinates of each acoustic sensor in the microphone array within a preset spatial coordinate system. Propagation distance refers to the straight-line distance a sound wave travels from a potential fault location to a specific acoustic sensor. Delay time refers to the time required for a sound wave to travel the propagation distance at a preset propagation speed. Delay compensation refers to shifting the acoustic signal acquired by each sensor along the time axis based on the calculated delay time, ensuring time alignment of the sound wave signal from the focal point direction across all channels. The beamforming output signal is the single signal obtained by summing the delay-compensated signals from all channels.

[0044] In this embodiment, the specific calculation process for delay-sum beamforming is as follows. Assume the microphone array contains N=8 acoustic sensors, and the spatial coordinates of the i-th sensor are... The coordinates are read from the calibration parameter file. Let the spatial location of the potential fault area be... .

[0045] The propagation distance between the i-th sensor and the potential fault area Calculated according to the three-dimensional Euclidean distance formula: .

[0046] For example, regarding spatial location mm and coordinates of a certain sensor mm, propagation distance m.

[0047] In this embodiment, the preset sound wave propagation speed is 343 m / s, which is the speed of sound under standard atmospheric pressure and dry air conditions at 20°C. Considering the influence of the ambient temperature of the dry-type transformer, the sound speed can be corrected based on the temperature sensor readings of the power distribution room during actual deployment. The correction formula is as follows: (T is the temperature in Celsius). In this embodiment, the temperature of the power distribution room is 28℃, therefore v = 331.3 + 0.606 × 28 = 348.3 m / s. The delay time corresponding to the i-th sensor. Taking the aforementioned propagation distance as an example, =0.6743 / 348.3≈1.936×10 - 3 seconds ≈ 1936 vs.

[0048] The delay compensation is implemented as follows: Since the acoustic signal sampling rate is 200kHz and the sampling interval is 5vs, ​​while the delay time... The frequency response interval is usually not an integer multiple of the sampling interval, therefore a fractional delay is achieved using the frequency domain phase shift method. Specifically, a fast Fourier transform is performed on the 400,000-point acoustic signal of the i-th sensor to obtain the spectrum. Multiply the spectrum by the phase compensation factor Then, an inverse fast Fourier transform is performed to obtain the time-domain signal after delay compensation. The effective length of the delay-compensated signal is taken as the middle value. Points, among which The maximum delay time among the eight sensors is, in this embodiment... ≈2.87ms, corresponding to approximately 574 sampling points, therefore the effective signal length is approximately 399426 sampling points.

[0049] The beamforming output signal is obtained by summing the delayed-compensated signals from the eight channels point by point. ,in Let be the original signal from the i-th sensor. To eliminate the influence of sensitivity differences between channels, the signal from each channel is normalized by dividing it by the root mean square value of that channel before superposition.

[0050] Feature extraction is performed on the beamforming output signal to obtain acoustic feature information. In this embodiment, feature extraction includes the following: calculating the peak amplitude of the beamforming output signal over the entire 2-second time window, i.e. The beamforming output signal is subjected to a short-time Fourier transform (window length of 2048 sampling points, window shift of 512 sampling points) to obtain the time spectrum. Then, the proportion of energy in the discharge characteristic frequency band (set to 30kHz to 80kHz in this embodiment) and the mechanical abnormality characteristic frequency band (set to 500Hz to 5kHz in this embodiment) to the total spectrum energy is calculated, i.e., the spectrum energy distribution. The number of pulses in the beamforming output signal whose amplitude exceeds the preset pulse detection threshold (set to 0.3 times the peak amplitude in this embodiment) is counted and divided by the time window length of 2s to obtain the pulse repetition rate, in units of times / second.

[0051] As can be seen from the above, the delay-summation beamforming scheme utilizes the spatial sampling characteristics of the microphone array to perform acoustic focusing on spatial locations, effectively suppressing interference from other parts of the dry-type transformer and environmental noise in the distribution room. The frequency domain phase shift method achieves fractional delay compensation with sub-sampling interval accuracy, avoiding phase errors caused by rounding up integer sampling points, ensuring the focusing gain of beamforming, so that even if the intensity of the fault sound source is weak (such as the tiny sound pulses generated by early partial discharge), sufficient signal-to-noise ratio can be obtained after superposition.

[0052] S105: Based on acoustic feature information and preset fault criteria, obtain the fault diagnosis results of the distribution transformer.

[0053] In this embodiment, acoustic feature information is compared item by item with pre-set fault criteria. Based on the comparison results, it is determined whether the dry-type transformer is currently in a normal state, a partial discharge fault state, or a mechanical structural abnormality fault state, and the final fault diagnosis result is output. The fault criteria are a set of rules comprehensively formulated based on historical fault data of dry-type transformers, laboratory simulated fault test data, and field operation experience. For different types of faults, acoustic feature parameters exhibit different distribution patterns, and corresponding thresholds and judgment logic are set accordingly.

[0054] As can be seen from the above, this embodiment acquires the current image data and current acoustic data of the distribution transformer, uses a pre-trained target detection model to quickly identify potential fault areas from the image and extract pixel coordinates, then converts them into spatial locations based on a preset mapping relationship, and then uses this spatial location as the focal point, combined with the current acoustic data, employs a delayed summation beamforming algorithm to perform directional acoustic signal enhancement and feature extraction, and finally obtains the diagnostic result based on acoustic feature information and preset fault criteria. This embodiment first uses visual information to achieve preliminary spatial positioning of the fault area, and then uses this positioning as a guide to perform spatially selective focusing processing on the acoustic data, effectively suppressing background noise interference from non-fault areas, improving the signal-to-noise ratio of acoustic features and the accuracy of fault identification, realizing the synergistic complementarity of visual positioning and acoustic diagnosis, and improving the accuracy and reliability of distribution transformer fault diagnosis.

[0055] In one embodiment of this application, the acoustic feature information includes the peak amplitude, spectral energy distribution, and pulse repetition rate of the beamforming output signal; the spectral energy distribution is the proportion of energy of the beamforming output signal within a preset discharge characteristic frequency band and a mechanical anomaly characteristic frequency band; the pulse repetition rate is the number of times the amplitude of the beamforming output signal exceeds a preset pulse detection threshold per unit time. Based on acoustic feature information and preset fault criteria, the fault diagnosis results of the distribution transformer are obtained, including: The peak amplitude is compared with a preset amplitude threshold, the spectral energy distribution is matched with the discharge characteristic frequency band and the mechanical anomaly characteristic frequency band respectively, and the pulse repetition rate is compared with a preset repetition rate threshold. If the peak amplitude exceeds the amplitude threshold, the spectral energy distribution is concentrated in the discharge characteristic frequency band, and the pulse repetition rate exceeds the repetition rate threshold, then the fault diagnosis result is determined to be a partial discharge fault. If the peak amplitude exceeds the amplitude threshold, and the spectral energy distribution is concentrated in the mechanical abnormality characteristic frequency band, and the pulse repetition rate does not exceed the repetition rate threshold, then the fault diagnosis result is determined to be a mechanical structural abnormality fault. If the peak amplitude does not exceed the amplitude threshold, the fault diagnosis result is determined to be normal.

[0056] In this embodiment, the discharge characteristic frequency band refers to the frequency range in which the acoustic pulse signal energy generated by partial discharge in a dry-type transformer is mainly concentrated. Partial discharge in dry-type transformers is mainly air gap discharge and surface discharge, and its acoustic pulses are characterized by extremely steep rise times (nanoseconds) and extremely short durations (microseconds). The spectral energy is mainly distributed in the ultrasonic band, which is set to 30kHz to 80kHz in this embodiment. The mechanical anomaly characteristic frequency band is the frequency range in which the vibration and impact acoustic energy generated by mechanical faults in dry-type transformers, such as loose core laminations, loose clamping bolts, and displaced winding pads, is mainly concentrated. The acoustic signals of such faults are characterized by low frequency and strong persistence, which is set to 500Hz to 5kHz in this embodiment. The amplitude threshold is the peak amplitude threshold used to distinguish normal background noise from fault acoustic signals. The repetition rate threshold is the pulse frequency threshold used to distinguish between random single pulses and continuous discharge events.

[0057] In this embodiment, the specific threshold settings for the fault criterion are as follows: The amplitude threshold is set to 0.05 Pa. This threshold can be obtained by statistically analyzing the peak amplitude of background noise from 30 normally operating dry-type transformers under full load conditions. The mean peak amplitude of the beamforming output signal during normal operation is 0.018 Pa, and the standard deviation is 0.009 Pa. Taking the mean plus three times the standard deviation, approximately 0.045 Pa, and leaving a certain margin, it is set to 0.05 Pa. The repetition rate threshold can be determined based on laboratory simulated partial discharge test data. Under the condition of a simulated discharge voltage of 12 kV (1.2 times the rated voltage of the dry-type transformer 10 kV), the discharge pulse repetition rate is stable between 40 and 120 times / second. The repetition rate of occasional electromagnetic interference pulses under normal operating conditions does not exceed 5 times / second. Therefore, the repetition rate threshold in this embodiment is set to 15 times / second.

[0058] In this embodiment, the peak amplitude is... Compared with the amplitude threshold of 0.05 Pa, the proportion of energy in the discharge characteristic frequency band is... and the proportion of frequency band energy of mechanical anomalies The pulse repetition rate is compared with the preset energy concentration criterion (the energy concentration criterion is: the energy proportion of the corresponding frequency band is greater than 0.6) to determine the pulse repetition rate. Compare with the repetition rate threshold of 15 times / second.

[0059] If all three of the following conditions are met: , and , If the frequency is calculated at times per second, the fault diagnosis result is determined to be a partial discharge fault. For example, in a certain diagnosis... , , , If the frequency is 10 times per second and all of the above conditions are met, it is determined to be a partial discharge fault.

[0060] If all three of the following conditions are met: , and , If the frequency is calculated at times per second, the fault diagnosis result is determined to be a mechanical structural abnormality. For example, in a certain diagnosis... , , , If the speed is [number] times per second and all of the above conditions are met, it is determined to be an abnormal mechanical structure fault.

[0061] like Regardless of the spectral energy distribution and pulse repetition rate, the fault diagnosis result is determined to be normal.

[0062] In this embodiment, for cases where no complete combination of the above conditions is met (e.g., the peak amplitude exceeds the threshold but the spectral energy distribution is not obviously concentrated in any characteristic frequency band), the diagnostic result is marked as "suspected abnormality, manual review recommended" in this embodiment, and no definitive fault conclusion is given directly, so as to avoid misjudgment.

[0063] As can be seen from the above, this embodiment combines acoustic characteristics of three dimensions—peak amplitude, spectral energy distribution, and pulse repetition rate—to reduce the probability of misjudgment compared to discrimination methods that rely on only a single parameter. Partial discharge and mechanical anomalies in dry-type transformers are naturally distinguishable in terms of spectral distribution; the former is concentrated in the ultrasonic band, while the latter is concentrated in the low-frequency band. This physical difference is fully utilized in the criterion design, giving fault type identification a clear physical basis rather than purely empirical rules.

[0064] In one embodiment of this application, after obtaining the fault diagnosis result of the distribution transformer, the method further includes: If the fault diagnosis result is a partial discharge fault or a mechanical structure abnormality fault, a spherical neighborhood region is determined with the spatial location as the center and the preset neighborhood radius as the radius. Within the spherical neighborhood region, M candidate focal points are generated according to the preset spatial step size; M is an integer greater than 1, and each candidate focal point is represented by three-dimensional coordinates in the preset spatial coordinate system. For each candidate focal point, based on the three-dimensional coordinates of the candidate focal point and the current acoustic data, the beamforming output signal corresponding to each candidate focal point is obtained through a delay summation beamforming algorithm; Calculate the signal energy value of the beamforming output signal corresponding to each candidate focal point within a preset time window; the signal energy value is the sum of the squares of the amplitudes of the beamforming output signal corresponding to the candidate focal point at each sampling point within the time window; The candidate focal point with the largest signal energy value among the M candidate focal points is determined as the accurate fault location point, and the three-dimensional coordinates of the accurate fault location point are updated in the fault diagnosis results.

[0065] In this embodiment, the spherical neighborhood region refers to a three-dimensional spherical spatial range with the spatial location as the center and a preset neighborhood radius as the radius. Candidate focal points are discrete three-dimensional coordinate points generated within the spherical neighborhood region according to a preset spatial step size. Each candidate focal point is used as the focal point for the delay-sum beamforming algorithm and calculated sequentially. The signal energy value is the sum of the squares of the amplitudes of the beamforming output signal at each sampling point within a preset time window, reflecting the sound source intensity in the direction of that focal point. The precise fault location point is the point with the largest signal energy value among all candidate focal points; its three-dimensional coordinates are considered the estimate closest to the actual fault sound source location. The neighborhood radius and spatial step size are two parameters controlling the precise location search range and search density.

[0066] In this embodiment, when the fault diagnosis result is a partial discharge fault or a mechanical structure abnormality fault, a precise fault location process is triggered. This is done by determining the spatial location. Centered on the sphere, with the neighborhood radius Define a spherical neighborhood region based on the radius. Neighborhood radius. The value is set to 80mm. The basis for this value is that the spatial position obtained by intersecting the three-dimensional model is limited by the accuracy of the model and the calibration error. Its positioning deviation is usually in the range of 30mm to 60mm. Taking 80mm can ensure that the real fault sound source falls into the spherical neighborhood with a high probability.

[0067] Within the spherical neighborhood, according to the preset spatial step size Candidate focal points are generated using the M-square method. The generation method involves establishing a local Cartesian coordinate system with the center of the sphere as the origin. Coordinate values ​​are taken in the X, Y, and Z directions at 10-millimeter intervals, and all grid points within 80mm of the center are selected. Calculations show that the number of candidate focal points M generated in this embodiment is approximately 2145. The spatial step size of 10mm is based on the fact that the actual sound source area of ​​partial discharge in dry-type transformers is typically between 5mm and 20mm. A step size of 10mm achieves a balance between positioning accuracy and computational complexity. If the step size is too small (e.g., 5mm), the number of candidate points will increase to approximately 17,000, increasing computation time by about 8 times, while the improvement in positioning accuracy is limited.

[0068] For each candidate focal point The delayed summation beamforming calculation is performed: based on the three-dimensional coordinates of the candidate focal point and the spatial coordinates of the eight acoustic sensors, the propagation distance and delay time are calculated. Delay compensation and summation are then applied to the current acoustic data to obtain the beamforming output signal corresponding to the candidate focal point. To reduce computational load, in this embodiment, the beamforming calculation is performed only on the first 0.5 seconds (100,000 sampling points) of the 2 seconds of acoustic data, because the fault sound pulse occurs continuously in time, and 0.5 seconds of data is sufficient to reflect the sound source intensity.

[0069] After obtaining the beamforming output signal for each candidate focal point, the signal energy value is calculated within a preset time window. In this embodiment, the time window is set to 0.5s (i.e., consistent with the data segment used for beamforming), and the signal energy value... ,in Let be the amplitude of the beamforming output signal of the j-th candidate focal point at the t-th sampling point.

[0070] Compare the signal energy values ​​of M=2145 candidate focal points, and determine the candidate focal point with the highest signal energy value as the precise fault location point. For example, in a certain diagnosis, the initial spatial location was... mm, after spherical neighborhood search, the candidate focal point with the highest signal energy value is located at mm, its signal energy value is The signal energy value at the center of the sphere is The energy increase of approximately 73% indicates that the actual sound source did indeed deviate from the initial estimated location. The three-dimensional coordinates of the precise fault location point were then determined. The mm update is added to the fault diagnosis results, and the final output diagnosis report includes the fault type (partial discharge fault) and precise location coordinates.

[0071] As can be seen from the above, this embodiment constructs a spherical neighborhood centered on the initial location point and performs a fine-grid scan. It then uses a delayed summation beamforming algorithm to traverse and calculate the signal energy of each candidate point, thereby pinpointing the precise fault location point corresponding to the maximum energy value. This coarse-to-fine secondary location strategy effectively overcomes the geometric errors inherent in single-model intersection, improves the spatial resolution and accuracy of fault location, and achieves a leap from regional-level troubleshooting to precise point-level location. This provides accurate spatial guidance for subsequent targeted maintenance and repair, significantly reducing operation and maintenance costs.

[0072] Based on the same principle as the fault diagnosis method for distribution transformers provided in the embodiments of this application, the embodiments of this application also provide a fault diagnosis device for distribution transformers, such as... Figure 2As shown, the fault diagnosis device 20 for a distribution transformer may specifically include: a data acquisition module 21, a visual recognition module 22, a coordinate transformation module 23, an acoustic processing module 24, and a fault diagnosis module 25. The data acquisition module 21 is used to acquire the current image data and current acoustic data of the distribution transformer; The visual recognition module 22 is used to input the current image data into a pre-trained target detection model for processing to obtain the location information of potential fault areas in the distribution transformer; wherein, the location information includes the pixel coordinates of the potential fault areas in the current image data; The coordinate transformation module 23 is used to determine the spatial location of the potential fault area based on the pixel coordinates and the preset mapping relationship between pixel coordinates and spatial coordinates. The acoustic processing module 24 is used to obtain acoustic feature information corresponding to potential fault areas by processing spatial location and current acoustic data through a delay summation beamforming algorithm. The fault diagnosis module 25 is used to obtain the fault diagnosis results of the distribution transformer based on acoustic feature information and preset fault criteria.

[0073] In one embodiment of this application, the data acquisition module 21 is specifically used for: Acquire visible light image data and infrared thermal imaging data of the distribution transformer acquired by a dual-spectrum image acquisition device; Based on the common feature points in visible light image data and infrared thermal imaging data, the infrared thermal imaging data and visible light image data are spatially registered to obtain the registered infrared thermal imaging data. The spatial registration is based on the pixel coordinate system of the visible light image data. The common feature points are the matching feature point pairs between the feature points in the visible light image data and the feature points in the infrared thermal imaging data whose feature descriptors have a similarity greater than a preset matching threshold. The registered infrared thermal imaging data and visible light image data are fused at the pixel level to obtain fused image data; the pixel coordinate system of the fused image data is consistent with the pixel coordinate system of the visible light image data; the fused image data is used as the current image data. The sound signals of the power distribution transformer during operation are acquired by a microphone array composed of multiple acoustic sensors to obtain the current acoustic data; the spatial coordinates of each acoustic sensor in the microphone array are pre-calibrated.

[0074] In one embodiment of this application, the training process of the pre-trained object detection model includes: Acquire sample image data of the distribution transformer; the sample image data includes sample images under normal conditions and sample images under various fault conditions; The fault regions in the sample image data are labeled to obtain the labeling information, which includes the bounding box coordinates of the fault regions and the fault category label. Based on sample image data and annotation information, the preset initial object detection network is iteratively trained. In each iteration, the loss function value is calculated based on the prediction output and annotation information of the initial object detection network, and the network parameters of the initial object detection network are updated based on the loss function value until the loss function value converges to the preset threshold, thus obtaining the pre-trained object detection model.

[0075] In one embodiment of this application, the coordinate transformation module 23 is specifically used for: Acquire the internal and external parameters of the dual-spectrum image acquisition device; the internal parameters include the focal length and principal point coordinates, and the external parameters include the rotation matrix and translation vector of the dual-spectrum image acquisition device relative to the preset spatial coordinate system. Based on pixel coordinates, intrinsic parameters, and extrinsic parameters, construct a projection ray originating from the pixel coordinates; The intersection of the projected ray and the preset three-dimensional model of the distribution transformer is calculated, and the coordinates of the intersection point of the projected ray and the preset three-dimensional model are determined as the spatial location corresponding to the potential fault area; the spatial location is represented by three-dimensional coordinates in the preset spatial coordinate system.

[0076] In one embodiment of this application, the acoustic processing module 24 is specifically used for: Obtain the sensor spatial coordinates of each acoustic sensor in the microphone array in a preset spatial coordinate system; the microphone array includes N acoustic sensors, where N is an integer greater than or equal to 2; Based on the spatial location and the sensor spatial coordinates of each acoustic sensor, calculate the propagation distance between each acoustic sensor and the spatial location of the potential fault area. Divide each propagation distance by the preset sound wave propagation speed to obtain the delay time corresponding to each acoustic sensor; the unit of propagation distance is meters, the unit of sound wave propagation speed is meters per second, and the unit of delay time is seconds; Based on the delay time corresponding to each acoustic sensor, delay compensation is performed on the acoustic signal collected by that acoustic sensor in the current acoustic data to obtain the delay-compensated acoustic signal corresponding to each acoustic sensor. The beamforming output signal is obtained by superimposing and summing the N delayed-compensated acoustic signals. Feature extraction is performed on the beamforming output signal to obtain the acoustic feature information corresponding to the potential fault area.

[0077] In one embodiment of this application, the acoustic feature information includes the peak amplitude, spectral energy distribution, and pulse repetition rate of the beamforming output signal; the spectral energy distribution is the proportion of energy of the beamforming output signal within a preset discharge characteristic frequency band and a mechanical anomaly characteristic frequency band; the pulse repetition rate is the number of times the amplitude of the beamforming output signal exceeds a preset pulse detection threshold per unit time. Fault diagnosis module 25 is specifically used for: The peak amplitude is compared with a preset amplitude threshold, the spectral energy distribution is matched with the discharge characteristic frequency band and the mechanical anomaly characteristic frequency band respectively, and the pulse repetition rate is compared with a preset repetition rate threshold. If the peak amplitude exceeds the amplitude threshold, the spectral energy distribution is concentrated in the discharge characteristic frequency band, and the pulse repetition rate exceeds the repetition rate threshold, then the fault diagnosis result is determined to be a partial discharge fault. If the peak amplitude exceeds the amplitude threshold, and the spectral energy distribution is concentrated in the mechanical abnormality characteristic frequency band, and the pulse repetition rate does not exceed the repetition rate threshold, then the fault diagnosis result is determined to be a mechanical structural abnormality fault. If the peak amplitude does not exceed the amplitude threshold, the fault diagnosis result is determined to be normal.

[0078] In one embodiment of this application, a fault diagnosis device 20 for a distribution transformer further includes: an update module, used for: If the fault diagnosis result is a partial discharge fault or a mechanical structure abnormality fault, a spherical neighborhood region is determined with the spatial location as the center and the preset neighborhood radius as the radius. Within the spherical neighborhood region, M candidate focal points are generated according to the preset spatial step size. M is an integer greater than 1, and each candidate focal point is represented by three-dimensional coordinates in the preset spatial coordinate system. For each candidate focal point, based on the three-dimensional coordinates of the candidate focal point and the current acoustic data, the beamforming output signal corresponding to each candidate focal point is obtained through a delay summation beamforming algorithm; Calculate the signal energy value of the beamforming output signal corresponding to each candidate focal point within a preset time window; the signal energy value is the sum of the squares of the amplitudes of the beamforming output signal corresponding to the candidate focal point at each sampling point within the time window; The candidate focal point with the largest signal energy value among the M candidate focal points is determined as the accurate fault location point, and the three-dimensional coordinates of the accurate fault location point are updated in the fault diagnosis results.

[0079] The apparatus in this application embodiment can execute the method provided in this application embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0080] Figure 3 A schematic diagram of the structure of an electronic device to which this application embodiment applies is shown, such as... Figure 3 As shown, the electronic device can be used to implement the methods provided in any embodiment of this application.

[0081] like Figure 3 As shown, the electronic device 300 may primarily include at least one processor 301. Figure 3 The diagram shows components such as a memory 302, a communication module 303, and an input / output interface 304. Optionally, these components can be connected and communicate with each other via a bus 305. It should be noted that... Figure 3 The structure of the electronic device 300 shown is merely illustrative and does not constitute a limitation on the electronic devices to which the methods provided in the embodiments of this application are applicable.

[0082] The memory 302 can be used to store operating systems and applications, etc. The applications may include computer programs that implement the methods shown in the embodiments of this application when invoked by the processor 301, and may also include programs for implementing other functions or services. The memory 302 may be ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices that can store information and computer programs, or it may be EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disk storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.

[0083] Processor 301 is connected to memory 302 via bus 305 and implements corresponding functions by calling the application programs stored in memory 302. Processor 301 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0084] Electronic device 300 can connect to a network via communication module 303 (which may include, but is not limited to, components such as a network interface) to communicate with other devices (such as user terminals or servers) through the network and achieve data interaction, such as sending data to or receiving data from other devices. Communication module 303 may include wired network interfaces and / or wireless network interfaces, meaning the communication module may include at least one of wired or wireless communication modules.

[0085] The electronic device 300 can connect to necessary input / output devices, such as a keyboard and display device, via the input / output interface 304. The electronic device 300 itself may have a display device, and other external display devices can also be connected via the input / output interface 304. Optionally, a storage device, such as a hard drive, can also be connected via the input / output interface 304 to store data from the electronic device 300, retrieve data from the storage device, or store data from the storage device in the memory 302. It is understood that the input / output interface 304 can be a wired interface or a wireless interface. Depending on the actual application scenario, the device connected to the input / output interface 304 can be a component of the electronic device 300 or an external device connected to the electronic device 300 when needed.

[0086] The bus 305 used to connect the components may include a path for transmitting information between the components. The bus 305 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Depending on its function, the bus 305 may be divided into an address bus, a data bus, a control bus, etc.

[0087] Optionally, for the solution provided in the embodiments of this application, the memory 302 can be used to store a computer program that executes the solution of this application, and the processor 301 runs the computer program. When the processor 301 runs the computer program, it implements the operation of the method or apparatus provided in the embodiments of this application.

[0088] Based on the same principle as the method provided in the embodiments of this application, the embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the corresponding content of the aforementioned method embodiments.

[0089] This application also provides a computer program product, which includes a computer program that, when executed by a processor, can implement the corresponding content of the aforementioned method embodiments.

[0090] It should be noted that the terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown in the figures or text descriptions.

[0091] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0092] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0093] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.

Claims

1. A fault diagnosis method for a distribution transformer, characterized in that, include: Acquire the current image data and current acoustic data of the distribution transformer; The current image data is input into a pre-trained target detection model for processing to obtain the location information of the potential fault area in the distribution transformer; wherein, the location information includes the pixel coordinates of the potential fault area in the current image data; Based on the pixel coordinates and the preset mapping relationship between pixel coordinates and spatial coordinates, the spatial location corresponding to the potential fault area is determined. Based on the spatial location and the current acoustic data, and processed by a delay summation beamforming algorithm, the acoustic feature information corresponding to the potential fault area is obtained. Based on the acoustic feature information and the preset fault criteria, the fault diagnosis result of the distribution transformer is obtained.

2. The fault diagnosis method for a distribution transformer as described in claim 1, characterized in that, The acquisition of the current image data and current acoustic data of the distribution transformer includes: Acquire visible light image data and infrared thermal imaging data of the distribution transformer acquired by a dual-spectrum image acquisition device; Based on the common feature points in the visible light image data and the infrared thermal imaging data, the infrared thermal imaging data and the visible light image data are spatially registered to obtain registered infrared thermal imaging data; the spatial registration is based on the pixel coordinate system of the visible light image data; the common feature points are the matching feature point pairs between the feature points in the visible light image data and the feature points in the infrared thermal imaging data whose feature descriptors have a similarity greater than a preset matching threshold. The registered infrared thermal imaging data and the visible light image data are fused at the pixel level to obtain fused image data; the pixel coordinate system of the fused image data is consistent with the pixel coordinate system of the visible light image data; the fused image data is used as the current image data. The sound signals of the power distribution transformer during operation are acquired by a microphone array composed of multiple acoustic sensors to obtain the current acoustic data; the spatial coordinates of each acoustic sensor in the microphone array are pre-calibrated.

3. The fault diagnosis method for a distribution transformer as described in claim 1, characterized in that, The training process of the pre-trained target detection model includes: Acquire sample image data of the distribution transformer; wherein, the sample image data includes sample images under normal conditions and sample images under various fault conditions; The fault regions in the sample image data are labeled to obtain labeling information; the labeling information includes the bounding box coordinates of the fault regions and fault category labels. Based on the sample image data and the annotation information, the preset initial target detection network is iteratively trained. In each iteration, the loss function value is calculated based on the prediction output of the initial target detection network and the annotation information, and the network parameters of the initial target detection network are updated based on the loss function value until the loss function value converges to a preset threshold, thereby obtaining the pre-trained target detection model.

4. The fault diagnosis method for a distribution transformer as described in claim 2, characterized in that, Determining the spatial location corresponding to the potential fault area based on the pixel coordinates and a preset mapping relationship between pixel coordinates and spatial coordinates includes: The internal and external parameters of the dual-spectrum image acquisition device are obtained; the internal parameters include focal length and principal point coordinates, and the external parameters include the rotation matrix and translation vector of the dual-spectrum image acquisition device relative to a preset spatial coordinate system. Based on the pixel coordinates, the internal parameters, and the external parameters, a projection ray originating from the pixel coordinates is constructed; The intersection operation is performed between the projected ray and the preset three-dimensional model of the distribution transformer, and the coordinates of the intersection point of the projected ray and the preset three-dimensional model are determined as the spatial location corresponding to the potential fault area; the spatial location is represented by three-dimensional coordinates in the preset spatial coordinate system.

5. The fault diagnosis method for a distribution transformer as described in claim 2, characterized in that, The acoustic feature information corresponding to the potential fault region is obtained by processing the spatial location and the current acoustic data using a delay-sum beamforming algorithm, including: Obtain the sensor spatial coordinates of each acoustic sensor in the microphone array in a preset spatial coordinate system; the microphone array includes N acoustic sensors, where N is an integer greater than or equal to 2; Based on the spatial location and the sensor spatial coordinates of each acoustic sensor, calculate the propagation distance between each acoustic sensor and the spatial location of the potential fault area; Divide each propagation distance by a preset sound wave propagation speed to obtain the delay time corresponding to each acoustic sensor; the unit of propagation distance is meters, the unit of sound wave propagation speed is meters per second, and the unit of delay time is seconds; Based on the delay time corresponding to each acoustic sensor, delay compensation is performed on the acoustic signal collected by the acoustic sensor in the current acoustic data to obtain the delay-compensated acoustic signal corresponding to each acoustic sensor. The beamforming output signal is obtained by superimposing and summing the N delayed-compensated acoustic signals. Feature extraction is performed on the beamforming output signal to obtain the acoustic feature information corresponding to the potential fault region.

6. The fault diagnosis method for a distribution transformer as described in claim 5, characterized in that, The acoustic feature information includes the peak amplitude, spectral energy distribution, and pulse repetition rate of the beamforming output signal; the spectral energy distribution is the proportion of energy of the beamforming output signal within a preset discharge characteristic frequency band and a mechanical anomaly characteristic frequency band; the pulse repetition rate is the number of times the amplitude of the beamforming output signal exceeds a preset pulse detection threshold per unit time. The fault diagnosis result of the distribution transformer obtained based on the acoustic feature information and preset fault criteria includes: The peak amplitude is compared with a preset amplitude threshold, the spectral energy distribution is matched with the discharge characteristic frequency band and the mechanical abnormality characteristic frequency band respectively, and the pulse repetition rate is compared with a preset repetition rate threshold. If the peak amplitude exceeds the amplitude threshold, the spectral energy distribution is concentrated in the discharge characteristic frequency band, and the pulse repetition rate exceeds the repetition rate threshold, then the fault diagnosis result is determined to be a partial discharge fault. If the peak amplitude exceeds the amplitude threshold, the spectral energy distribution is concentrated in the mechanical abnormality characteristic frequency band, and the pulse repetition rate does not exceed the repetition rate threshold, then the fault diagnosis result is determined to be a mechanical structural abnormality fault. If the peak amplitude does not exceed the amplitude threshold, the fault diagnosis result is determined to be normal.

7. The fault diagnosis method for a distribution transformer as described in claim 6, characterized in that, After obtaining the fault diagnosis results of the distribution transformer, the following steps are also included: If the fault diagnosis result is the partial discharge fault or the mechanical structure abnormality fault, then a spherical neighborhood region is determined with the spatial location as the center and the preset neighborhood radius as the radius, and M candidate focal points are generated within the spherical neighborhood region according to the preset spatial step size; M is an integer greater than 1, and each candidate focal point is represented by three-dimensional coordinates in the preset spatial coordinate system; For each candidate focal point, based on the three-dimensional coordinates of the candidate focal point and the current acoustic data, and through the delay summation beamforming algorithm, the beamforming output signal corresponding to each candidate focal point is obtained; Calculate the signal energy value of the beamforming output signal corresponding to each candidate focal point within a preset time window; the signal energy value is the sum of the squares of the amplitudes of the beamforming output signal corresponding to the candidate focal point at each sampling point within the time window; The candidate focal point with the largest signal energy value among the M candidate focal points is determined as the precise fault location point, and the three-dimensional coordinates of the precise fault location point are updated in the fault diagnosis result.

8. A fault diagnosis device for a distribution transformer, characterized in that, include: The data acquisition module is used to acquire the current image data and current acoustic data of the distribution transformer; A visual recognition module is used to input the current image data into a pre-trained target detection model for processing to obtain the location information of the potential fault area in the distribution transformer; wherein, the location information includes the pixel coordinates of the potential fault area in the current image data; The coordinate transformation module is used to determine the spatial location corresponding to the potential fault area based on the pixel coordinates and the preset mapping relationship between pixel coordinates and spatial coordinates. The acoustic processing module is used to obtain the acoustic feature information corresponding to the potential fault area by processing the spatial location and the current acoustic data through a delay summation beamforming algorithm. The fault diagnosis module is used to obtain the fault diagnosis result of the distribution transformer based on the acoustic feature information and preset fault criteria.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes a fault diagnosis method for a distribution transformer according to any one of claims 1 to 7 when running the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements a fault diagnosis method for a distribution transformer as described in any one of claims 1 to 7.

Images