Defect monitoring method, system and device for high-voltage switch

Abstract

The invention provides a defect monitoring method, system and device for a high-voltage switch, and the method comprises the steps: collecting image information of target equipment; calling three-dimensional model data of the target equipment, determining corresponding point cloud data information through conversion, and determining weight information of corresponding features according to the point cloud data information; inputting the image information of the target equipment and the weight information of the corresponding features into a preset neural network for combination to obtain a corresponding three-dimensional model neural field; projecting the three-dimensional model neural field to a preset multi-surface projection coordinate surface to obtain picture information corresponding to each angle of the target equipment; and sequentially inputting the picture information corresponding to each angle of the target equipment into a preset defect detection model, and detecting the appearance state of the target equipment to obtain a defect monitoring result. According to the invention, comprehensive and accurate defect monitoring of the high-voltage switchgear is realized, the problem of incomplete coverage of existing routing inspection is solved, and the target defect position and defect content are accurately positioned.

Description

Technical Field

[0001] The present invention relates to the technical field of defect monitoring of high-voltage switches, and in particular to a method, system and device for defect monitoring of high-voltage switches. Background Art

[0002] In the power system sector, substation equipment maintenance and inspection are crucial. Traditional substation inspection solutions primarily utilize inspection vehicles or drones to conduct patrols targeting existing fixed-point photography targets to perform tasks such as defect detection. Existing substation inspection technologies fall into two main categories. One type involves fixed-point, timed, and targeted inspections using inspection vehicles, drones, or on-site fixed-point cameras. This type of solution detects defects by photographing equipment along fixed routes and locations and then analyzing the captured images. The other type utilizes conventional depth camera modeling and scanning solutions. This solution first scans the equipment with a depth camera to obtain three-dimensional data, then performs secondary modeling on this data to construct a three-dimensional model of the equipment. Finally, defect detection is performed based on this model.

[0003] However, existing inspection technologies have many shortcomings. Limited by factors such as inspection routes and equipment layout, inspection schemes using inspection carts, inspection drones, and on-site fixed-point cameras can only detect fixed target locations. This results in incomplete inspection target locations and makes it difficult to completely locate the target defect location points, which in turn makes the assessment of the fault point incomplete. When using conventional depth camera modeling and scanning schemes, there are problems with long scanning times and complex inspection routes. Moreover, after the scan is completed, the equipment still needs to be remodeled, which not only increases the workload, but also leads to problems such as inaccurate defect information reflection and cumbersome defect location positioning. It is not well suited for complex equipment scenarios on the substation site. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, system and device for defect monitoring of high-voltage switches, so as to solve the technical problem of how to achieve efficient and accurate defect monitoring of high-voltage switches.

[0005] In one aspect, a defect monitoring method for a high-voltage switch is provided, comprising:

[0006] Collecting image information of the target device, wherein the image information includes at least multiple groups of images in different angular directions, and each group of angular directions includes at least one image;

[0007] Retrieving the 3D model data of the target device and determining the corresponding point cloud data information through conversion, and determining the weight information of the corresponding features based on the point cloud data information; inputting the image information of the target device and the weight information of the corresponding features into a preset neural network for combination to obtain the corresponding 3D model neural field;

[0008] Project the 3D model neural field onto a preset multi-faceted projection coordinate plane to obtain image information corresponding to each angle of the target device;

[0009] The image information corresponding to each angle of the target device is sequentially input into a preset defect detection model, the appearance status of the target device is detected, and the defect monitoring results are obtained; wherein, the defect detection model is a recognition model trained by historical defect images.

[0010] Preferably, when collecting image information of the target device, a set of images of the target device taken at different shooting angles and without jitter are selected by a preset image collection device, and the images at different angles and directions are combined into the final image information.

[0011] Preferably, the method further includes inputting the point cloud data information into a preset encoding program to obtain weight information of the neural field voxel feature, and outputting the weight information of the neural field voxel feature as weight information of the corresponding feature.

[0012] Preferably, the preset neural network is used to use the weight information of the features of the point cloud data as the loss weight for evaluating the voxel feature information generated by the neural network.

[0013] Preferably, the number of the multi-faceted projection coordinate planes is at least eighteen.

[0014] Preferably, the multi-faceted projection coordinate planes are multiple projection coordinate planes constructed according to the model characteristics of the three-dimensional neural field of the retrieved three-dimensional model data of the target device.

[0015] Preferably, it also includes, if it is detected that the appearance state of the target device is abnormal, locating the defect position point according to the multi-faceted projection coordinates and the position information of the three-dimensional model neural field, obtaining a three-dimensional appearance patrol detection, and outputting it as a defect monitoring result.

[0016] On the other hand, a defect monitoring system for a high-voltage switch is provided, which is used to implement the defect monitoring method for a high-voltage switch, comprising:

[0017] An image acquisition module is used to acquire image information of a target device, wherein the image information includes at least a plurality of groups of images in different angular directions, and each group of angular directions includes at least one image;

[0018] An image processing module is used to retrieve the 3D model data of the target device and determine the corresponding point cloud data information through conversion, and determine the weight information of the corresponding features based on the point cloud data information; input the image information of the target device and the weight information of the corresponding features into a preset neural network for combination to obtain the corresponding 3D model neural field; project the 3D model neural field onto a preset multi-faceted projection coordinate plane to obtain image information corresponding to each angle of the target device;

[0019] The defect detection module is used to input the image information corresponding to each angle of the target device into a preset defect detection model in sequence, detect the appearance status of the target device, and obtain defect monitoring results; wherein, the defect detection model is a recognition model trained by historical defect images.

[0020] On the other hand, a defect monitoring device for a high-voltage switch is also provided, and a target device is monitored by the defect monitoring system for a high-voltage switch.

[0021] In summary, the implementation of the embodiments of the present invention has the following beneficial effects:

[0022] The defect monitoring method, system, and device for high-voltage switches provided by this invention utilize the FrugalNeRF neural field network to construct a three-dimensional detection model of the patrol target, enabling comprehensive and accurate defect monitoring of high-voltage switchgear, addressing the issue of incomplete patrol coverage in existing inspections. An octahedral neural field projection expression coordinate system is constructed, and a multi-view defect monitoring set for the three-dimensional model is established. Multi-view images are obtained by projecting the model onto this coordinate system and applied to the target defect monitoring model to accurately locate the target defect location and its content. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, other drawings obtained based on these drawings still fall within the scope of the present invention.

[0024] Figure 1 Schematic diagram of a defect monitoring method for a high-voltage switch according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in further detail below with reference to the accompanying drawings.

[0026] like Figure 1 FIG. 1 is a schematic diagram of an embodiment of a method for defect monitoring of a high-voltage switch provided by the present invention. In this embodiment, the method includes the following steps:

[0027] Step S1, collect image information of the target device, wherein the image information includes at least multiple groups of images in different angles and directions, and each group of angles and directions includes at least one image; use inspection equipment such as drones and inspection carts to take pictures from different angles of the target inspection equipment (high-voltage switch) in advance.

[0028] In a specific embodiment, when capturing image information from a target device, a preset image capture device selects a set of smooth, non-shaky images of the target device taken at different angles. These images are then combined to form the final image information. To ensure the accuracy and completeness of model construction, a minimum of six smooth, non-shaky images are selected from each angle, with at least one image captured from each set. These images form a good foundation for training.

[0029] Step S2: retrieve the three-dimensional model data of the target device and determine the corresponding point cloud data information through conversion, and determine the weight information of the corresponding features based on the point cloud data information; input the image information of the target device and the weight information of the corresponding features into a preset neural network for combination to obtain the corresponding three-dimensional model neural field; the preset neural network is used to use the weight information of the features of the point cloud data as the loss weight for evaluating the voxel feature information generated by the neural network. Retrieve the three-dimensional model data of the target device and the converted point cloud data information, encode the point cloud data information to form the weight information of the neural field voxel features. Then, the captured image set and the voxel weight information obtained by encoding the point cloud data are input into the FrugalNeRF network to construct the three-dimensional field model.

[0030] In a specific embodiment, the point cloud data information is input into a preset encoding program to obtain the weight information of the neural field voxel features, and the weight information of the neural field voxel features is output as the weight information of the corresponding features. The point cloud data encoding information is used as the loss weight for evaluating the voxel feature information generated by the FrugalNeRF network. In this way, the model building process is optimized to ensure the accurate construction of the three-dimensional model neural field of the target patrol device. The FrugalNeRF network is a network with the sparse perspective Nerf three-dimensional model construction feature, which can be used to build a high-quality three-dimensional model based on images with fewer perspectives.

[0031] Step S3, projects the three-dimensional model neural field into a preset multi-faceted projection coordinate plane, thereby obtaining image information corresponding to each angle of the target device; the number of the multi-faceted projection coordinate planes is at least eighteen. The multi-faceted projection coordinate planes are multiple projection coordinate planes constructed based on the model characteristics of the three-dimensional neural field of the retrieved three-dimensional model data of the target device. Utilizing the model characteristics of the three-dimensional neural field, an eighteen-faceted projection coordinate plane of the model is constructed. The constructed model neural field is projected into the eighteen-faceted coordinate plane to obtain image information at each angle of the device. Through this multi-faceted projection method, the appearance of the device can be fully displayed from multiple angles, providing rich data for subsequent defect detection.

[0032] Step S4, input the image information corresponding to each angle of the target device into the preset defect detection model in turn, detect the appearance status of the target device, and obtain the defect monitoring result; wherein, the defect detection model is a recognition model trained by historical defect images. If the appearance status of the target device is detected to be abnormal, the defect position point is located according to the multi-faceted projection coordinates and the position information of the three-dimensional model neural field, and a three-dimensional appearance patrol detection is obtained, and output as the defect monitoring result. The eighteen pictures obtained at each angle are input into the defect detection model in turn to detect the appearance status of the device. If abnormal appearance information is detected, the defect position point is located using the eighteen-face coordinate information and the neural field position information, thereby completing the three-dimensional appearance patrol detection of the device.

[0033] In actual applications, first of all, according to the installation location and on-site environment of the high-voltage switchgear, the shooting routes and shooting angles of the drones and inspection vehicles are reasonably planned to ensure that clear images of the equipment at different angles can be obtained. In the data processing stage, professional data processing software is used to encode and pre-process the acquired three-dimensional model data and point cloud data, and then the processed data and the captured images are input into the established FrugalNeRF network model for training and three-dimensional field model construction. After the three-dimensional model neural field is constructed, the eighteen-sided projection coordinate surface is constructed according to the preset algorithm, and the model projection and image acquisition are completed. Finally, the acquired images are input into the pre-trained defect detection model for detection and defect location. During the entire implementation process, the parameters and algorithms of each step can be optimized and adjusted according to the actual situation to improve the accuracy and efficiency of monitoring.

[0034] An embodiment of the present invention further provides a defect monitoring system for a high-voltage switch, which is used in the defect monitoring method for a high-voltage switch, comprising:

[0035] An image acquisition module is used to acquire image information of a target device, wherein the image information includes at least a plurality of groups of images in different angular directions, and each group of angular directions includes at least one image;

[0036] An image processing module is used to retrieve the 3D model data of the target device and determine the corresponding point cloud data information through conversion, and determine the weight information of the corresponding features based on the point cloud data information; input the image information of the target device and the weight information of the corresponding features into a preset neural network for combination to obtain the corresponding 3D model neural field; project the 3D model neural field onto a preset multi-faceted projection coordinate plane to obtain image information corresponding to each angle of the target device;

[0037] The defect detection module is used to input the image information corresponding to each angle of the target device into a preset defect detection model in sequence, detect the appearance status of the target device, and obtain defect monitoring results; wherein, the defect detection model is a recognition model trained by historical defect images.

[0038] An embodiment of the present invention further provides a defect monitoring device for a high-voltage switch, and a target device is monitored by the defect monitoring system for a high-voltage switch.

[0039] It should be noted that the systems and devices described in the above embodiments correspond to the methods described in the above embodiments. Therefore, the parts of the systems and devices described in the above embodiments that are not described in detail can be obtained by referring to the contents of the methods described in the above embodiments, and will not be repeated here.

[0040] In summary, the implementation of the embodiments of the present invention has the following beneficial effects:

[0041] The defect monitoring method, system, and device for high-voltage switches provided by this invention utilize the FrugalNeRF neural field network to construct a three-dimensional detection model of the patrol target, enabling comprehensive and accurate defect monitoring of high-voltage switchgear, addressing the issue of incomplete patrol coverage in existing inspections. An octahedral neural field projection expression coordinate system is constructed, and a multi-view defect monitoring set for the three-dimensional model is established. Multi-view images are obtained by projecting the model onto this coordinate system and applied to the target defect monitoring model to accurately locate the target defect location and its content.

[0042] The above disclosure is merely a preferred embodiment of the present invention and certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A defect monitoring method for a high voltage switch, characterized in that: include: Collecting image information of the target device, wherein the image information includes at least multiple groups of images in different angular directions, and each group of angular directions includes at least one image; Retrieving the 3D model data of the target device and determining the corresponding point cloud data information through conversion, and determining the weight information of the corresponding features based on the point cloud data information; inputting the image information of the target device and the weight information of the corresponding features into a preset neural network for combination to obtain the corresponding 3D model neural field; Project the 3D model neural field onto a preset multi-faceted projection coordinate plane to obtain image information corresponding to each angle of the target device; The image information corresponding to each angle of the target device is sequentially input into a preset defect detection model, the appearance status of the target device is detected, and the defect monitoring results are obtained; wherein, the defect detection model is a recognition model trained by historical defect images.

2. The method according to claim 1, wherein When collecting image information of the target device, a set of images of the target device taken at different shooting angles and without jitter are selected through a preset image acquisition device, and the images at different angles and directions are combined into the final image information.

3. The method according to claim 1, wherein It also includes inputting the point cloud data information into a preset encoding program to obtain the weight information of the neural field voxel feature, and outputting the weight information of the neural field voxel feature as the weight information of the corresponding feature.

4. The method according to claim 3, wherein The preset neural network is used to use the weight information of the features of the point cloud data as the loss weight for evaluating the voxel feature information generated by the neural network.

5. The method according to claim 1, wherein The number of the multi-faceted projection coordinate surfaces is at least eighteen.

6. The method according to claim 5, wherein The multi-faceted projection coordinate planes are multiple projection coordinate planes constructed based on the model characteristics of the three-dimensional neural field of the retrieved three-dimensional model data of the target device.

7. The method according to claim 1, wherein It also includes, if the appearance state of the target device is detected to be abnormal, locating the defect position point according to the multi-faceted projection coordinates and the position information of the three-dimensional model neural field, obtaining a three-dimensional appearance patrol detection, and outputting it as a defect monitoring result.

8. A defect monitoring system for a high voltage switch, used to implement the method according to any one of claims 1 to 7, characterized in that: include, An image acquisition module is used to acquire image information of a target device, wherein the image information includes at least a plurality of groups of images in different angular directions, and each group of angular directions includes at least one image; An image processing module is used to retrieve the 3D model data of the target device and determine the corresponding point cloud data information through conversion, and determine the weight information of the corresponding features based on the point cloud data information; input the image information of the target device and the weight information of the corresponding features into a preset neural network for combination to obtain the corresponding 3D model neural field; project the 3D model neural field onto a preset multi-faceted projection coordinate plane to obtain image information corresponding to each angle of the target device; The defect detection module is used to input the image information corresponding to each angle of the target device into a preset defect detection model in sequence, detect the appearance status of the target device, and obtain defect monitoring results; wherein, the defect detection model is a recognition model trained by historical defect images.

9. A defect monitoring device for a high voltage switch, characterized in that: The target device is monitored by the defect monitoring system for a high-voltage switch according to claim 8 .

Images