A method and system for identifying defects in insulation winding wires of high-power transformers

Abstract

The present invention discloses a method and system for identifying defects in insulating wrapped wires applied to high-power transformers. The method comprises acquiring a target insulating wrapped wire and performing fixed-point multi-angle X-ray image acquisition on the target insulating wrapped wire to obtain a set of X-ray images of the target insulating wrapped wire to be identified; inputting the set of X-ray images to be identified into a multi-model collaborative defect recognition model so that the multi-model collaborative defect recognition model outputs defect data of the target insulating wrapped wire, wherein the defect data includes a defect type and a defect location, and the defect types include internal conductor breakage, a gap between the insulating layer and the conductor, abnormal insulating layer thickness, and insulating layer shedding; by implementing the present invention, the comprehensiveness and accuracy of defect identification of the target insulating wrapped wire can be improved, thereby ensuring the safe and stable operation of the transformer when the insulating wrapped wire is applied to a high-power transformer after defect identification.

Description

Technical Field

[0001] The present invention relates to the technical field of insulating wire defect identification, and in particular to an insulating wire defect identification method and system applied to a high-power transformer. Background Art

[0002] Transformers are essential equipment in power systems. Their primary function is to transform voltage and transmit and distribute current through the principle of electromagnetic induction. During transformer manufacturing, insulating wire, as a key component, plays a crucial role. Its use in transformers effectively prevents current leakage and short circuits, reducing the risk of transformer failure. Furthermore, its wear resistance reduces winding damage caused by mechanical stress and friction, significantly contributing to the safe and stable operation of transformers. Before applying insulating wire to transformers, especially high-power transformers, which are core power equipment in power systems, it is crucial to identify defects in the insulating wire. This ensures that the wire will not cause power fluctuations, voltage variations, or other issues once applied, ensuring safe and stable operation.

[0003] Existing defect detection methods for insulating wire primarily rely on capturing images of the wire's exterior and locating defects within these images using algorithms such as target detection. However, this defect detection method can only identify surface defects such as deformation and shedding of the insulating wire, but cannot promptly detect internal conductor fractures. Using insulating wire with internal defects in transformers can still cause power fluctuations and voltage variations, hindering safe and stable operation. Summary of the Invention

[0004] The embodiments of the present invention provide a method and system for identifying defects in insulating winding wires applied to high-power transformers, which can solve the problem that the existing technology cannot identify internal defects in insulating winding wires. On the basis of realizing the identification of the internal structure of the target insulating winding wire, the comprehensiveness and accuracy of the target insulating winding wire defect identification are improved, thereby ensuring the safe and stable operation of the transformer when the target insulating winding wire that does not have defects after defect identification is applied to the high-power transformer.

[0005] An embodiment of the present invention provides a method for identifying defects in insulation winding wires of a high-power transformer, comprising:

[0006] Acquire a target insulating wrapped wire and perform fixed-point multi-angle X-ray image acquisition on the target insulating wrapped wire to obtain a set of X-ray images of the target insulating wrapped wire to be identified; wherein the set of X-ray images to be identified includes a plurality of X-ray images of the target insulating wrapped wire to be identified taken from a plurality of angles;

[0007] Inputting the to-be-identified X-ray image set into a multi-model collaborative defect recognition model, so that the multi-model collaborative defect recognition model outputs defect data of the target insulated wrapped wire, the defect data including defect type and defect location, the defect type including internal conductor breakage, gap between the insulation layer and the conductor, abnormal insulation layer thickness, and insulation layer shedding;

[0008] When the X-ray image set to be identified is input into the multi-model collaborative defect recognition model, the first defect recognition model of the multi-model collaborative defect recognition model constructs a three-dimensional model of the target insulating wrapped wire according to the X-ray image set to be identified, and outputs first defect data according to the three-dimensional model of the target insulating wrapped wire;

[0009] The second defect recognition model of the multi-model collaborative defect recognition model outputs second defect data according to each X-ray image to be recognized;

[0010] The collaborative decision model of the multi-model collaborative defect recognition model outputs defect data of the target insulated wrapped wire based on the first defect data and the second defect data.

[0011] Furthermore, the construction of the multi-model collaborative defect recognition model includes:

[0012] Obtaining a training sample set for a multi-model collaborative defect recognition model; wherein the training sample set includes a plurality of X-ray image sample subsets of a plurality of sample insulating wrapped wires, each of the X-ray image sample subsets including a plurality of X-ray image defect samples of the current sample insulating wrapped wire taken from multiple angles, a reference three-dimensional model of the current sample insulating wrapped wire, and actual defect data of the current sample insulating wrapped wire;

[0013] Constructing an initial multi-model collaborative defect recognition model; wherein the initial multi-model collaborative defect recognition model includes an initial first defect recognition model, an initial second defect recognition model, a calibration model, a weight adjustment model and an initial collaborative decision model;

[0014] The multi-model collaborative defect recognition model is generated by performing multi-model collaborative training on the initial multi-model collaborative defect recognition model using the training sample set of the multi-model collaborative defect recognition model until the initial multi-model collaborative defect recognition model converges.

[0015] Furthermore, the multi-model collaborative training of the initial multi-model collaborative defect recognition model using the training sample set of the multi-model collaborative defect recognition model includes:

[0016] The initial multi-model collaborative defect recognition model is trained by taking the X-ray image sample subset as input and the predicted defect data of the sample insulation winding wire corresponding to the X-ray image sample subset as output.

[0017] In each collaborative training process, the initial first defect recognition model constructs an initial three-dimensional model of the current sample insulating wrapped wire based on each X-ray image defect sample, and transmits the initial three-dimensional model to the calibration model so that the calibration model obtains a baseline three-dimensional model of the current sample insulating wrapped wire, determines a three-dimensional model construction error based on the baseline three-dimensional model of the current sample insulating wrapped wire and the initial three-dimensional model, and feeds back the three-dimensional model construction error to the initial first defect recognition model so that the initial first defect recognition model adjusts the initial three-dimensional model based on the three-dimensional model construction error, and outputs first predicted defect data based on the adjusted initial three-dimensional model;

[0018] The initial second recognition model extracts each defect feature data of each X-ray image defect sample, and outputs second predicted defect data based on each defect feature data;

[0019] The weight adjustment model determines a first prediction error of the first recognition model and a second prediction error of the second recognition model based on the first predicted defect data, the second predicted defect data, and the actual defect data, and adjusts the first weight and the second weight of the initial collaborative decision-making model based on the first prediction error, the second prediction error, and the three-dimensional model construction error; wherein the first weight is associated with the first predicted defect data, and the second weight is associated with the second predicted defect data;

[0020] The initial collaborative decision-making model outputs the predicted defect data of the current sample insulated winding wire based on the adjusted first weight, the adjusted second weight, the first predicted defect data and the second predicted defect data, and determines the collaborative decision error based on the predicted defect data and the actual defect data.

[0021] Furthermore, determining the three-dimensional model construction error based on the reference three-dimensional model of the current sample insulating wrapped wire and the initial three-dimensional model includes:

[0022] Registering the reference three-dimensional model and the initial three-dimensional model to obtain a registered reference three-dimensional model and a registered initial three-dimensional model;

[0023] Constructing feature matching geometric space based on the registered reference 3D model;

[0024] Synchronously moving the registered initial three-dimensional model and the registered reference three-dimensional model to the feature matching geometric space;

[0025] Repeating the regional feature matching operation in the feature matching geometric space until the registered reference 3D model and the registered initial 3D model have no overlapping regions in the feature matching geometric space, and determining a 3D model construction error based on the registered reference 3D model and the registered initial 3D model without overlapping regions in the feature matching geometric space;

[0026] The regional feature matching operation includes:

[0027] Constructing a plurality of spheres with radii of random values, and making the spheres randomly cover the feature matching geometric space;

[0028] For each sphere, the first Euclidean distance between each pair of registration points of the registered reference 3D model and the registered initial 3D model in the sphere coverage area is calculated;

[0029] The overlapping areas of the registered reference three-dimensional model and the registered initial three-dimensional model are eliminated according to the first Euclidean distances.

[0030] Furthermore, after eliminating the overlapping areas between the registered reference three-dimensional model and the registered initial three-dimensional model according to each of the first Euclidean distances, the method further includes:

[0031] Calculate the second Euclidean distance between the two registration points of the registered reference three-dimensional model and the registered initial three-dimensional model outside the preset distance area of the sphere coverage area;

[0032] The second Euclidean distances are fitted to obtain a current sphere radius adjustment parameter, and the current sphere radius is adjusted according to the current sphere radius adjustment parameter.

[0033] Furthermore, the method of using the X-ray image sample subset as input for the initial multi-model collaborative defect recognition model includes:

[0034] Input the initial multi-model collaborative defect recognition model in the order in which each X-ray image defect sample in the X-ray image sample subset is collected;

[0035] The initial second recognition model extracts each defect feature data of each X-ray image defect sample and outputs second predicted defect data according to each defect feature data, including:

[0036] The initial second recognition model extracts defect feature data of the first input X-ray image defect sample, and when the initial second recognition model receives the remaining X-ray image defect samples except the first one, determines the overlapping area of the current X-ray image defect sample and the previous X-ray image defect sample based on the previous X-ray image defect sample and the current X-ray image defect sample;

[0037] After removing the overlapping area between the current X-ray image defect sample and the previous X-ray image defect sample, defect feature extraction is performed on the current X-ray image defect sample to obtain defect feature data of the current X-ray image defect sample;

[0038] Second predicted defect data is output according to each defect feature data of each X-ray image defect sample.

[0039] Furthermore, the method of collecting fixed-point multi-angle X-ray images of the target insulating wrapped wire to obtain a set of to-be-identified X-ray images of the target insulating wrapped wire includes:

[0040] Place the target insulated wrapped wire on the stage of the X-ray image acquisition device and acquire X-ray images at the initial angle;

[0041] After the X-ray image acquisition at the initial angle is completed, the stage is repeatedly rotated to the preset angle and the X-ray image at the current angle is acquired until the stage rotates 360°. The X-ray image acquisition is stopped and a set of X-ray images to be identified of the target insulated wrapped wire is constructed based on the X-ray images acquired at each angle.

[0042] Based on the above method embodiment, the present invention provides a corresponding system embodiment;

[0043] An embodiment of the present invention provides a defect recognition system for insulating winding wires applied to high-power transformers, comprising: an image data acquisition module and a defect recognition module;

[0044] The image data acquisition module is used to acquire a target insulating wrapped wire and perform fixed-point multi-angle X-ray image acquisition on the target insulating wrapped wire to obtain a set of X-ray images to be identified of the target insulating wrapped wire; wherein the set of X-ray images to be identified includes a plurality of X-ray images to be identified taken from a plurality of angles of the target insulating wrapped wire;

[0045] The defect recognition module is used to input the X-ray image set to be identified into the multi-model collaborative defect recognition model, so that the multi-model collaborative defect recognition model outputs defect data of the target insulating wrapped wire, the defect data including defect type and defect location, the defect type including internal conductor breakage, gap between the insulation layer and the conductor, abnormal insulation layer thickness and insulation layer shedding; wherein, when the X-ray image set to be identified is input into the multi-model collaborative defect recognition model, the first defect recognition model of the multi-model collaborative defect recognition model constructs a three-dimensional model of the target insulating wrapped wire according to the X-ray image set to be identified, and outputs first defect data according to the three-dimensional model of the target insulating wrapped wire; the second defect recognition model of the multi-model collaborative defect recognition model outputs second defect data according to each X-ray image to be identified; the collaborative decision model of the multi-model collaborative defect recognition model outputs the defect data of the target insulating wrapped wire according to the first defect data and the second defect data.

[0046] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for identifying defects of insulating winding wires applied to high-power transformers as described in the above-mentioned embodiment of the invention.

[0047] Another embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the method for identifying insulation winding wire defects applied to high-power transformers described in the above-mentioned embodiment of the invention.

[0048] The following beneficial effects are achieved by implementing the present invention:

[0049] The present invention provides a method and system for identifying defects in insulating wires used in high-power transformers. The defect identification method obtains a set of X-ray images of the target insulating wire to be identified based on fixed-point multi-angle shooting, and utilizes the ability of X-ray images to identify the internal structure of the target insulating wire. The method can obtain the internal structural characteristics of the target insulating wire, such as internal conductor breakage, gaps between the insulating layer and the conductor, abnormal insulating layer thickness, and insulating layer shedding defects, thereby providing a rich data basis for identifying internal and external defects of the target insulating wire. Furthermore, by inputting the X-ray image set to be identified into the multi-model collaborative defect recognition model, the first defect recognition model of the multi-model collaborative defect recognition model is used to construct a three-dimensional model of the target insulating winding wire, defects are identified based on the three-dimensional model and first defect data is output, and the second defect recognition model of the multi-model collaborative defect recognition model is used to output second defect data through two-dimensional image feature extraction. Then, the first defect data and the second defect data are integrated through the collaborative defect recognition model to output the defect data of the target insulating winding wire; the multi-model collaborative defect recognition model is used to identify defects of the target insulating winding wire from three-dimensional and two-dimensional perspectives respectively, and on the basis of identifying the internal structure of the target insulating winding wire, the comprehensiveness and accuracy of the defect identification of the target insulating winding wire are improved, thereby ensuring the safe and stable operation of the transformer when the target insulating winding wire that has no defects after defect identification is applied to a high-power transformer. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The present invention provides a flowchart of a method for identifying defects in insulation winding wires of a high-power transformer, according to an embodiment of the present invention.

[0051] Figure 2 The present invention is a schematic structural diagram of an insulation winding wire defect identification system for a high-power transformer provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0053] like Figure 1 FIG. 1 is a method for identifying insulation winding defects of a high-power transformer provided by an embodiment of the present invention, comprising:

[0054] Step S1: Acquire a target insulating wrapped wire and perform fixed-point multi-angle X-ray image acquisition on the target insulating wrapped wire to obtain a set of X-ray images of the target insulating wrapped wire to be identified; wherein the set of X-ray images to be identified includes a plurality of X-ray images of the target insulating wrapped wire to be identified taken from a plurality of angles;

[0055] Step S2: inputting the X-ray image set to be identified into a multi-model collaborative defect recognition model, so that the multi-model collaborative defect recognition model outputs defect data of the target insulated wrapped wire, the defect data including defect type and defect location, the defect type including internal conductor breakage, gap between the insulation layer and the conductor, abnormal insulation layer thickness, and insulation layer shedding;

[0056] Among them, when the X-ray image set to be identified is input into the multi-model collaborative defect recognition model, the first defect recognition model of the multi-model collaborative defect recognition model constructs a three-dimensional model of the target insulating wrapped wire based on the X-ray image set to be identified, and outputs first defect data based on the three-dimensional model of the target insulating wrapped wire; the second defect recognition model of the multi-model collaborative defect recognition model outputs second defect data based on each X-ray image to be identified; the collaborative decision model of the multi-model collaborative defect recognition model outputs the defect data of the target insulating wrapped wire based on the first defect data and the second defect data.

[0057] Specifically, a target insulating wrapped wire for which defects need to be identified is obtained, and fixed-point multi-angle X-ray image acquisition is performed on the target insulating wrapped wire to obtain multi-angle X-ray images of the target insulating wrapped wire to be identified, thereby obtaining a set of X-ray images of the target insulating wrapped wire to be identified.

[0058] In a preferred embodiment, the method of performing fixed-point multi-angle X-ray image acquisition on the target insulating wrapped wire to obtain the X-ray image set of the target insulating wrapped wire to be identified includes: placing the target insulating wrapped wire on the stage of the X-ray image acquisition device to perform X-ray image acquisition at an initial angle; after the X-ray image acquisition at the initial angle is completed, repeatedly rotating the stage by a preset angle and acquiring X-ray images at the current angle until the stage rotates 360°, stopping the X-ray image acquisition, and constructing the X-ray image set of the target insulating wrapped wire to be identified based on the X-ray images acquired at each angle.

[0059] Specifically, after obtaining the target insulating wrapped wire, the target insulating wrapped wire is placed on the stage of the X-ray image acquisition device. After determining that the target insulating wrapped wire is placed stably, the target insulating wrapped wire is X-ray imaged by a fixed-point X-ray image acquisition device to obtain an X-ray image of the target insulating wrapped wire at an initial angle. Further, after the stage is rotated by a preset angle (such as 30 degrees, 60 degrees or 90 degrees), and after each rotation by the preset angle, an X-ray image at the current angle is acquired until the stage rotates 360°. At this time, it is considered that the X-ray images of the target insulating wrapped wire at all angles have been acquired. These acquired X-ray images are integrated into a set of X-ray images to be identified of the target insulating wrapped wire in the acquisition order.

[0060] For step S2, the above-collected X-ray image set to be identified is input into the multi-model collaborative defect recognition model, and each X-ray image to be identified of the target insulating wire is processed by the multi-model collaborative defect recognition model to output the defect type and defect location of the target insulating wire. In the present invention, the defect types mainly include four categories: internal conductor breakage of the insulating wire, gap between the insulating layer and the conductor of the insulating wire, abnormal insulation thickness of the insulating wire, and insulation shedding of the insulating wire. Among them, internal conductor breakage, gap between the insulating layer and the conductor, and abnormal insulation thickness are internal structural abnormalities that need to be determined by looking through the internal structure, and insulation shedding is a surface abnormality that can be determined by external structural features. When the X-ray image set to be identified is input into the multi-model collaborative defect recognition model, the first defect recognition model included in the multi-model collaborative defect recognition model constructs a three-dimensional model of the target insulating wire based on the input data, and outputs first defect data based on the three-dimensional model. The second defect recognition model outputs second defect data, and the collaborative decision-making model outputs the defect data of the target insulating wire based on the first defect data and the second defect data.

[0061] In a preferred embodiment, the construction of the multi-model collaborative defect recognition model includes: obtaining a training sample set of the multi-model collaborative defect recognition model; wherein the training sample set includes several X-ray image sample subsets of several sample insulating windings, each of the X-ray image sample subsets includes several X-ray image defect samples of the current sample insulating winding from multiple angles, a baseline three-dimensional model of the current sample insulating winding and real defect data of the current sample insulating winding; constructing an initial multi-model collaborative defect recognition model; wherein the initial multi-model collaborative defect recognition model includes an initial first defect recognition model, an initial second defect recognition model, a calibration model, a weight adjustment model and an initial collaborative decision model; performing multi-model collaborative training on the initial multi-model collaborative defect recognition model using the training sample set of the multi-model collaborative defect recognition model until the initial multi-model collaborative defect recognition model converges, thereby generating the multi-model collaborative defect recognition model.

[0062] Specifically, the construction of the multi-model collaborative defect recognition model mainly includes three steps: training sample acquisition, initial model construction and model training and tuning.

[0063] First, several sample insulating wires are obtained as sample objects. It should be noted that in order to be able to identify insulating wires of multiple specifications, when obtaining sample insulating wires, sample insulating wires corresponding to each specification can be obtained based on the specifications of the insulating wires for which defects need to be identified. For each sample insulating wire, different defect types are constructed and arranged at different locations, and the locations and types of each defect are marked on the sample insulating wire as the real defect data of the insulating wire. The sample insulating wire is fixed on the stage of the X-ray image acquisition device, and several radiographic images of the sample insulating wire taken at different angles are acquired using the same acquisition method as the X-ray image acquisition in step S1, and these images are used as several X-ray image defect samples of the sample insulating wire. In addition, a baseline three-dimensional model of the sample insulating wire during production is obtained, and a subset of X-ray image samples of the current sample insulating wire is constructed using this data. Preferably, for sample insulating wires of the same specification, multiple subsets of X-ray image samples are constructed by changing the defect type and defect location, thereby forming a training sample set for the multi-model collaborative defect recognition model.

[0064] Secondly, an initial multi-model collaborative defect recognition model is constructed, which is composed of an initial first defect recognition model, an initial second defect recognition model, a calibration model, a weight adjustment model, and an initial collaborative decision model. The initial multi-model collaborative defect recognition model is multi-model collaboratively trained using a training sample set of the multi-model collaborative defect recognition model until the initial multi-model collaborative defect recognition model converges, such as when the loss function of the initial multi-model collaborative defect recognition model is minimized or when the maximum number of training times is reached, thereby generating the multi-model collaborative defect recognition model.

[0065] In a preferred embodiment, the multi-model collaborative training of the initial multi-model collaborative defect recognition model using the training sample set of the multi-model collaborative defect recognition model includes:

[0066] The initial multi-model collaborative defect recognition model is trained by taking the X-ray image sample subset as input and the predicted defect data of the sample insulation winding wire corresponding to the X-ray image sample subset as output.

[0067] During each collaborative training process, the initial first defect recognition model constructs an initial three-dimensional model of the current sample insulating wire based on each X-ray image defect sample and transmits the initial three-dimensional model to the calibration model. The calibration model obtains a baseline three-dimensional model of the current sample insulating wire, determines a three-dimensional model construction error based on the baseline three-dimensional model of the current sample insulating wire and the initial three-dimensional model, and feeds the three-dimensional model construction error back to the initial first defect recognition model. The initial first defect recognition model adjusts the initial three-dimensional model based on the three-dimensional model construction error. After the initial three-dimensional model is adjusted, defect features are extracted from the adjusted initial three-dimensional model, and first predicted defect data is determined based on the extracted defect features.

[0068] In a preferred embodiment, the method of determining the three-dimensional model construction error based on the reference three-dimensional model of the current sample insulated wrapped wire and the initial three-dimensional model includes: aligning the reference three-dimensional model and the initial three-dimensional model to obtain the aligned reference three-dimensional model and the aligned initial three-dimensional model; constructing a feature matching geometric space based on the aligned reference three-dimensional model; synchronously moving the aligned initial three-dimensional model and the aligned reference three-dimensional model to the feature matching geometric space; repeatedly performing regional feature matching operations in the feature matching geometric space until the aligned reference three-dimensional model and the aligned initial three-dimensional model in the feature matching geometric space are aligned. The initial three-dimensional model after registration has no repeated areas, and the three-dimensional model construction error is determined based on the registered reference three-dimensional model and the registered initial three-dimensional model without repeated areas in the feature matching geometric space; wherein, the regional feature matching operation includes: constructing a number of spheres with random radii, and making each of the spheres randomly cover the feature matching geometric space; for each sphere, calculating the first Euclidean distance between the pairwise registration points of the registered reference three-dimensional model and the registered initial three-dimensional model in the sphere coverage area; and eliminating the repeated areas of the registered reference three-dimensional model and the registered initial three-dimensional model according to each of the first Euclidean distances.

[0069] In a supplementary embodiment, after eliminating the repeated areas of the aligned reference three-dimensional model and the aligned initial three-dimensional model according to each of the first Euclidean distances, it also includes: calculating the second Euclidean distances between the aligned reference three-dimensional model and the aligned initial three-dimensional model that are outside the preset distance area of the sphere coverage area; fitting each second Euclidean distance to obtain the current sphere radius adjustment parameter, and adjusting the current sphere radius according to the current sphere radius adjustment parameter.

[0070] Specifically, the reference 3D model and the initial 3D model are registered using the ICP (Iterative Closest Point) algorithm to obtain a registered reference 3D model and a registered initial 3D model. Based on the registered reference 3D model, a feature matching geometric space is constructed that completely encloses the registered reference 3D model and leaves a certain amount of residual space after enclosing. The geometry of this feature matching geometric space can be determined based on the geometry of the registered reference 3D model. In the present invention, the reference 3D model of the sample insulating wrapped wire primarily consists of two upper and lower cylinders with larger diameters and a central cylinder with a smaller diameter. Therefore, when constructing the feature matching geometric space, it can be constructed as a cylinder. After the feature matching geometric space is constructed, the registered initial 3D model and the registered reference 3D model are synchronously moved into the feature matching geometric space. This synchronous movement ensures that the registration effect is not affected. A regional feature matching operation is performed in the feature matching geometric space. This regional feature matching operation primarily removes overlapping regions between the registered reference 3D model and the registered initial 3D model. The regional feature matching operation specifically includes: obtaining a random value that does not exceed the value threshold, and constructing several initial spheres with the random value as the radius; wherein the value threshold is the minimum value of the length, width, height and height of the reference three-dimensional model. The constructed spheres are randomly covered in the feature matching geometric space. For each sphere, the first Euclidean distance between the two registration points of the registered reference three-dimensional model and the registered initial three-dimensional model in the coverage area of the sphere is calculated. If the first Euclidean distance between the two registration points is equal to zero or almost zero, it is considered that the two registration points basically coincide, that is, the registration points corresponding to the repeated area, and they are eliminated at this time; if the first Euclidean distance between the two registration points is not zero, it is considered that the two registration points do not coincide and there is a deviation, and they are not eliminated at this time. Furthermore, a second Euclidean distance is calculated between each pair of registration points outside a preset distance area from the edge of the sphere. This second Euclidean distance is the same as the first Euclidean distance. Based on this second Euclidean distance, whether the pair of registration points overlap is determined. Furthermore, whether there is any overlapping area within the area expanded outward by the preset distance from the current sphere's coverage area is determined. If not, the sphere radius is reduced the next time duplicate areas are eliminated. If so, a fitting is performed based on each non-zero second Euclidean distance to obtain a fitting result. The sphere radius is then increased based on this fitting result the next time duplicate areas are eliminated. This second Euclidean distance allows for dynamic adjustment of the sphere radius, improving the efficiency of dynamic duplicate area elimination.

[0071] It should be noted that the calculation of the Euclidean distance between the two registration points when multiple spheres are eliminated is a multi-threaded parallel calculation process, that is, it can be understood that each calculation unit only processes the Euclidean distance calculation corresponding to one sphere, and multiple calculation units calculate in parallel to achieve the effect of improving the efficiency of dynamic elimination of duplicate areas.

[0072] The initial second recognition model extracts each defect feature data of each X-ray image defect sample, and outputs second predicted defect data according to each defect feature data.

[0073] In a preferred embodiment, the method of using a subset of X-ray image samples as input for the initial multi-model collaborative defect recognition model includes: inputting the initial multi-model collaborative defect recognition model in the order of collection of each X-ray image defect sample in the X-ray image sample subset; the initial second recognition model extracts each defect feature data of each X-ray image defect sample, and outputs second predicted defect data based on each defect feature data, including: the initial second recognition model extracts the defect feature data of the first input X-ray image defect sample, and when the initial second recognition model receives the remaining X-ray image defect samples except the first one, determines the repeated area between the current X-ray image defect sample and the previous X-ray image defect sample based on the previous X-ray image defect sample and the current X-ray image defect sample; after removing the repeated area between the current X-ray image defect sample and the previous X-ray image defect sample, extracts the defect feature of the current X-ray image defect sample to obtain the defect feature data of the current X-ray image defect sample; and outputs the second predicted defect data based on the defect feature data of each X-ray image defect sample.

[0074] Specifically, when a subset of X-ray image samples is input into the initial multi-model collaborative defect recognition model, it must be input in the order of the shooting time. The initial second recognition model is built based on the YOLO algorithm, which is used to extract defect features of the input X-ray image defect samples one by one, and then output the second predicted defect data based on the defect features. However, if the preset rotation angle is small during shooting, such as 15 degrees or 30 degrees each time, after the first image is shot, there will be a certain overlapping area between each subsequent image and the previous image. If feature recognition and extraction are also performed on the overlapping area, the same feature will be extracted multiple times, resulting in repeated confusion of the final features, which affects the accuracy of the judgment. The initial second recognition model directly extracts defect features when the first X-ray image defect sample is input; when the remaining X-ray image defect samples except the first one are input into the initial second recognition model, the overlapping area between the current X-ray image defect sample and the previous X-ray image defect sample is determined according to the previous X-ray image defect sample and the current X-ray image defect sample, and the overlapping area is removed. The defect feature extraction is performed on the remaining area after removing the overlapping area to obtain the defect feature data of the current X-ray image defect sample; the second predicted defect data is output according to the defect feature data of each X-ray image defect sample.

[0075] The weight adjustment model determines the first prediction error of the first identification model and the second prediction error of the second identification model based on the first predicted defect data, the second predicted defect data and the actual defect data, and adjusts the first weight and second weight of the initial collaborative decision-making model based on the first prediction error, the second prediction error and the three-dimensional model construction error; wherein, the first weight is associated with the first predicted defect data, and the second weight is associated with the second predicted defect data.

[0076] The initial collaborative decision-making model outputs the predicted defect data of the current sample insulated winding wire based on the adjusted first weight, the adjusted second weight, the first predicted defect data and the second predicted defect data, and determines the collaborative decision error based on the predicted defect data and the actual defect data.

[0077] Specifically, the weight adjustment model determines the first prediction error of the first identification model based on the first predicted defect data and the actual defect data; and determines the second prediction error of the second identification model based on the second predicted defect data and the actual defect data. The first weight parameter is adjusted according to the first prediction error and the three-dimensional model construction error, and the second weight parameter is adjusted according to the second prediction error. Among them, the first weight is the proportion parameter of the first predicted defect data considered by the initial collaborative decision model, and the second weight is the proportion parameter of the second predicted defect data considered by the initial collaborative decision model. Exemplarily, the decision formula of the initial collaborative decision model can be expressed as:

[0078]

[0079] in, represents the output of the initial collaborative decision-making model, i.e., the predicted defect data; represents the first weight; represents first predicted defect data; represents the second weight; Indicates the second predicted defect data.

[0080] After the first and second weights are adjusted, the initial collaborative decision-making model outputs predicted defect data for the current sample insulated wrapped wire based on the adjusted first and second weights, the first and second predicted defect data. A collaborative decision error is determined based on the predicted defect data and the actual defect data, and a multi-model collaborative defect recognition model is generated when the collaborative decision error is minimized.

[0081] Preferably, the loss function of the initial multi-model collaborative defect recognition model can also be determined based on the collaborative decision error, the first prediction error, the second prediction error and the three-dimensional model construction error. When each error is less than the preset minimum error, the multi-model collaborative defect recognition model is generated.

[0082] Based on the above method embodiments, the present invention provides corresponding system embodiments.

[0083] like Figure 2 As shown, an embodiment of the present invention provides an insulation winding wire defect recognition system for a high-power transformer, comprising: an image data acquisition module and a defect recognition module;

[0084] The image data acquisition module is used to acquire a target insulating wrapped wire and perform fixed-point multi-angle X-ray image acquisition on the target insulating wrapped wire to obtain a set of X-ray images to be identified of the target insulating wrapped wire; wherein the set of X-ray images to be identified includes a plurality of X-ray images to be identified taken from a plurality of angles of the target insulating wrapped wire;

[0085] The defect recognition module is used to input the X-ray image set to be identified into the multi-model collaborative defect recognition model, so that the multi-model collaborative defect recognition model outputs defect data of the target insulating wrapped wire, the defect data including defect type and defect location, the defect type including internal conductor breakage, gap between the insulation layer and the conductor, abnormal insulation layer thickness and insulation layer shedding; wherein, when the X-ray image set to be identified is input into the multi-model collaborative defect recognition model, the first defect recognition model of the multi-model collaborative defect recognition model constructs a three-dimensional model of the target insulating wrapped wire according to the X-ray image set to be identified, and outputs first defect data according to the three-dimensional model of the target insulating wrapped wire; the second defect recognition model of the multi-model collaborative defect recognition model outputs second defect data according to each X-ray image to be identified; the collaborative decision model of the multi-model collaborative defect recognition model outputs the defect data of the target insulating wrapped wire according to the first defect data and the second defect data.

[0086] It should be noted that the system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the drawings of the system embodiment provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which can be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0087] Those skilled in the art will clearly understand that for the sake of convenience and brevity, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0088] Based on the above method embodiment, the present invention provides a corresponding terminal device embodiment.

[0089] An embodiment of the present invention provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements a method for identifying defects in insulating winding wires applied to high-power transformers as described in any one of the present inventions.

[0090] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0091] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the terminal device and connects various parts of the entire terminal device using various interfaces and lines.

[0092] The memory can be used to store the computer program. The processor implements the various functions of the terminal device by running or executing the computer program stored in the memory and accessing the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store an operating system, at least one application required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0093] Based on the above method embodiment, the present invention provides a corresponding storage medium embodiment.

[0094] An embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute any one of the methods for identifying defects in insulating winding wires applied to high-power transformers described in the present invention.

[0095] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include: any entity or system capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium.

[0096] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying defects in insulation winding wires of high-power transformers, characterized in that: include: Acquire a target insulating wrapped wire and perform fixed-point multi-angle X-ray image acquisition on the target insulating wrapped wire to obtain a set of X-ray images of the target insulating wrapped wire to be identified; wherein the set of X-ray images to be identified includes a plurality of X-ray images of the target insulating wrapped wire to be identified taken from a plurality of angles; Inputting the to-be-identified X-ray image set into a multi-model collaborative defect recognition model, so that the multi-model collaborative defect recognition model outputs defect data of the target insulated wrapped wire, the defect data including defect type and defect location, the defect type including internal conductor breakage, gap between the insulation layer and the conductor, abnormal insulation layer thickness, and insulation layer shedding; When the set of X-ray images to be identified is input into the multi-model collaborative defect recognition model, the first defect recognition model of the multi-model collaborative defect recognition model constructs a three-dimensional model of the target insulating wrapped wire based on the set of X-ray images to be identified, and outputs first defect data based on the three-dimensional model of the target insulating wrapped wire; the second defect recognition model of the multi-model collaborative defect recognition model outputs second defect data based on each X-ray image to be identified; and the collaborative decision-making model of the multi-model collaborative defect recognition model outputs defect data of the target insulating wrapped wire based on the first defect data and the second defect data; The construction of the multi-model collaborative defect recognition model includes: Obtaining a training sample set for a multi-model collaborative defect recognition model; wherein the training sample set includes a plurality of X-ray image sample subsets of a plurality of sample insulating wrapped wires, each of the X-ray image sample subsets including a plurality of X-ray image defect samples of the current sample insulating wrapped wire taken from multiple angles, a reference three-dimensional model of the current sample insulating wrapped wire, and actual defect data of the current sample insulating wrapped wire; Constructing an initial multi-model collaborative defect recognition model; wherein the initial multi-model collaborative defect recognition model includes an initial first defect recognition model, an initial second defect recognition model, a calibration model, a weight adjustment model and an initial collaborative decision model; The multi-model collaborative defect recognition model is generated by performing multi-model collaborative training on the initial multi-model collaborative defect recognition model using the training sample set of the multi-model collaborative defect recognition model until the initial multi-model collaborative defect recognition model converges.

2. The method for identifying defects in insulation winding wires of a high-power transformer according to claim 1, wherein: The multi-model collaborative training of the initial multi-model collaborative defect recognition model using the training sample set of the multi-model collaborative defect recognition model includes: The initial multi-model collaborative defect recognition model is trained by taking the X-ray image sample subset as input and the predicted defect data of the sample insulation winding wire corresponding to the X-ray image sample subset as output. In each collaborative training process, the initial first defect recognition model constructs an initial three-dimensional model of the current sample insulating wrapped wire based on each X-ray image defect sample, and transmits the initial three-dimensional model to the calibration model so that the calibration model obtains a baseline three-dimensional model of the current sample insulating wrapped wire, determines a three-dimensional model construction error based on the baseline three-dimensional model of the current sample insulating wrapped wire and the initial three-dimensional model, and feeds back the three-dimensional model construction error to the initial first defect recognition model so that the initial first defect recognition model adjusts the initial three-dimensional model based on the three-dimensional model construction error, and outputs first predicted defect data based on the adjusted initial three-dimensional model; The initial second defect recognition model extracts each defect feature data of each X-ray image defect sample, and outputs second predicted defect data based on each defect feature data; The weight adjustment model determines a first prediction error of the first recognition model and a second prediction error of the second recognition model based on the first predicted defect data, the second predicted defect data, and the actual defect data, and adjusts the first weight and the second weight of the initial collaborative decision-making model based on the first prediction error, the second prediction error, and the three-dimensional model construction error; wherein the first weight is associated with the first predicted defect data, and the second weight is associated with the second predicted defect data; The initial collaborative decision-making model outputs the predicted defect data of the current sample insulated winding wire based on the adjusted first weight, the adjusted second weight, the first predicted defect data and the second predicted defect data, and determines the collaborative decision error based on the predicted defect data and the actual defect data.

3. The method for identifying defects in insulation winding wires used in high-power transformers according to claim 2, wherein: Determining the three-dimensional model construction error based on the reference three-dimensional model of the current sample insulated wrapped wire and the initial three-dimensional model includes: Registering the reference three-dimensional model and the initial three-dimensional model to obtain a registered reference three-dimensional model and a registered initial three-dimensional model; Constructing feature matching geometric space based on the registered reference 3D model; Synchronously moving the registered initial three-dimensional model and the registered reference three-dimensional model to the feature matching geometric space; Repeating the regional feature matching operation in the feature matching geometric space until the registered reference 3D model and the registered initial 3D model have no overlapping regions in the feature matching geometric space, and determining a 3D model construction error based on the registered reference 3D model and the registered initial 3D model without overlapping regions in the feature matching geometric space; The regional feature matching operation includes: Constructing a plurality of spheres with radii of random values, and making the spheres randomly cover the feature matching geometric space; For each sphere, the first Euclidean distance between each pair of registration points of the registered reference 3D model and the registered initial 3D model in the sphere coverage area is calculated; The overlapping areas of the registered reference three-dimensional model and the registered initial three-dimensional model are eliminated according to the first Euclidean distances.

4. The method for identifying defects in insulation winding wires used in high-power transformers according to claim 3, wherein: After eliminating the overlapping areas of the registered reference three-dimensional model and the registered initial three-dimensional model according to each of the first Euclidean distances, the method further includes: Calculate the second Euclidean distance between the two registration points of the registered reference three-dimensional model and the registered initial three-dimensional model outside the preset distance area of the sphere coverage area; The second Euclidean distances are fitted to obtain a current sphere radius adjustment parameter, and the current sphere radius is adjusted according to the current sphere radius adjustment parameter.

5. The method for identifying defects in insulation winding wires of a high-power transformer according to claim 2, wherein: The method of using the X-ray image sample subset as the input of the initial multi-model collaborative defect recognition model includes: Input the initial multi-model collaborative defect recognition model in the order in which each X-ray image defect sample in the X-ray image sample subset is collected; The initial second defect recognition model extracts each defect feature data of each X-ray image defect sample and outputs second predicted defect data according to each defect feature data, including: The initial second defect recognition model extracts defect feature data of the first input X-ray image defect sample, and when the initial second defect recognition model receives the remaining X-ray image defect samples except the first one, determines the overlapping area of the current X-ray image defect sample and the previous X-ray image defect sample based on the previous X-ray image defect sample and the current X-ray image defect sample; After removing the overlapping area between the current X-ray image defect sample and the previous X-ray image defect sample, defect feature extraction is performed on the current X-ray image defect sample to obtain defect feature data of the current X-ray image defect sample; Second predicted defect data is output according to each defect feature data of each X-ray image defect sample.

6. The method for identifying defects in insulation winding wires of a high-power transformer according to claim 1, wherein: The method of collecting fixed-point multi-angle X-ray images of the target insulating wrapped wire to obtain a set of X-ray images of the target insulating wrapped wire to be identified includes: Place the target insulated wrapped wire on the stage of the X-ray image acquisition device and acquire X-ray images at the initial angle; After the X-ray image acquisition at the initial angle is completed, the stage is repeatedly rotated to the preset angle and the X-ray image at the current angle is acquired until the stage rotates 360°. The X-ray image acquisition is stopped and a set of X-ray images to be identified of the target insulated wrapped wire is constructed based on the X-ray images acquired at each angle.

7. A system for identifying defects in insulation winding wires of high-power transformers, characterized in that: include: Image data acquisition module and defect recognition module; The image data acquisition module is used to acquire a target insulating wrapped wire and perform fixed-point multi-angle X-ray image acquisition on the target insulating wrapped wire to obtain a set of X-ray images to be identified of the target insulating wrapped wire; wherein the set of X-ray images to be identified includes a plurality of X-ray images to be identified taken from a plurality of angles of the target insulating wrapped wire; The defect recognition module is used to input the X-ray image set to be identified into the multi-model collaborative defect recognition model, so that the multi-model collaborative defect recognition model outputs defect data of the target insulating wrapped wire, the defect data including defect type and defect location, the defect type including internal conductor breakage, gap between the insulation layer and the conductor, abnormal insulation layer thickness and insulation layer shedding; wherein, when the X-ray image set to be identified is input into the multi-model collaborative defect recognition model, the first defect recognition model of the multi-model collaborative defect recognition model constructs a three-dimensional model of the target insulating wrapped wire according to the X-ray image set to be identified, and outputs first defect data according to the three-dimensional model of the target insulating wrapped wire; the second defect recognition model of the multi-model collaborative defect recognition model outputs second defect data according to each X-ray image to be identified; the collaborative decision model of the multi-model collaborative defect recognition model outputs the target insulating wrapped wire according to the first defect data and the second defect data. defect data; the construction of the multi-model collaborative defect recognition model includes: obtaining a training sample set of the multi-model collaborative defect recognition model; wherein, the training sample set includes several X-ray image sample subsets of several sample insulation windings, each of the X-ray image sample subsets includes several X-ray image defect samples of the current sample insulation winding from multiple angles, a baseline three-dimensional model of the current sample insulation winding and the real defect data of the current sample insulation winding; constructing an initial multi-model collaborative defect recognition model; wherein, the initial multi-model collaborative defect recognition model includes an initial first defect recognition model, an initial second defect recognition model, a calibration model, a weight adjustment model and an initial collaborative decision model; performing multi-model collaborative training on the initial multi-model collaborative defect recognition model with the training sample set of the multi-model collaborative defect recognition model until the initial multi-model collaborative defect recognition model converges, and generating the multi-model collaborative defect recognition model.

8. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for identifying defects of insulating winding wires applied to a high-power transformer as described in any one of claims 1 to 6 is implemented.

9. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute the method for identifying defects of insulating winding wires applied to high-power transformers as described in any one of claims 1 to 6.

Images