Improved YOLOv9 pole-mounted transformer insulation damage identification and spraying method

Abstract

The invention discloses an improved YOLOv9 pole-mounted transformer insulation damage identification and spraying method. The method comprises the following steps: constructing a damage data set of a 10kV pole-mounted transformer; constructing and training an image recognition model based on YOLOv9 to obtain an improved pole-mounted transformer insulation damage recognition model; a video is collected through an unmanned aerial vehicle, a key frame is extracted, a pole-mounted transformer insulation damage identification model is input, and the model outputs damage position and degree information; determining the type of the nozzle, constructing a nozzle three-dimensional model by using Solidworks, and importing Workbench to optimize the grid quality; spraying simulation is conducted in Fluent, spraying parameters and unmanned aerial vehicle flight control parameters are set, and the optimal spraying working condition is determined. By combining the unmanned aerial vehicle technology and deep learning, accurate detection and efficient spraying of insulation damage of the 10kV pole-mounted transformer are achieved, the efficiency and safety of maintenance operation are remarkably improved, and the method has great significance in power grid inspection.

Description

Technical Field

[0001] The present invention relates to the technical field of insulation damage identification and spraying of pole-mounted transformers, and in particular to an improved YOLOv9 method for insulation damage identification and spraying of pole-mounted transformers. Background Art

[0002] With the rapid growth of electricity demand in recent years, the requirements for the reliability of distribution network operation are increasing. 10kV pole-mounted transformers are widely used in rural, suburban and urban distribution networks to provide stable power supply for residents and commercial users. They are usually located in the center of 10kV pole-mounted equipment and are large in size. During daily operation, the insulating sheath of the 10kV pole-mounted transformer will be damaged due to interference such as bird claws, branches, rain rust, etc., resulting in exposed metal points. These insulation damage points are mainly concentrated on the seven taps of the pole-mounted transformer. These taps are the key nodes of power connection and are also the areas most prone to electrical leakage. In order to achieve highly reliable power, the pole-mounted transformer needs to be insulated.

[0003] Traditional power equipment status detection is usually carried out by manual inspection, and the insulation maintenance method for power equipment is also mainly manual spraying and manual replacement of insulation sheaths. Manual operations are inefficient and inaccurate, prone to omissions and misjudgments, and have high safety risks. When necessary, power outages are required. With the rapid development of artificial intelligence, deep learning technology has gradually emerged. Especially in the application field of convolutional neural networks (CNN), traditional manual feature extraction methods have been gradually replaced by autonomous learning mechanisms, which greatly improves the accuracy and processing efficiency of recognition. By combining drone technology, the image data of 10kV pole-mounted transformers can be automatically collected and incorporated into the artificial intelligence platform without relying on manual inspections. The deep learning algorithm is used to intelligently process and identify these data, which can effectively detect potential problems such as insulation damage and metal exposure, provide data support for subsequent spraying, and improve the efficiency and accuracy of insulation spraying. Then, the drone spraying operation is used to carry out targeted spraying on the damaged insulation of the pole-mounted transformer after identification. By combining drone spraying technology with deep learning intelligent analysis, a new solution is provided for the detection and maintenance of insulation damage of 10kV pole-mounted transformers. Summary of the invention

[0004] 1. Technical problems to be solved:

[0005] In view of the above technical problems, the present invention provides an improved YOLOv9 method for identifying and spraying insulation damage of a pole-mounted transformer.

[0006] 2. Technical solution:

[0007] Improved YOLOv9 pole transformer insulation damage identification and spraying method, characterized by: including:

[0008] Step 1: Construct a damaged dataset of 10kV pole-mounted transformers; use a high-resolution camera to capture images of pole-mounted transformers and preprocess them to obtain damaged sample images of 0kV pole-mounted transformers, and then construct a damaged dataset;

[0009] Step 2: Build an image recognition model based on YOLOv9; optimize the loss function in the YOLOv9 model to MPDIoU, and use the PGI module of the YOLOv9 model to improve the image detection performance;

[0010] Step 3: Use the damaged data set to train the model in step 2; divide the damaged data set into 80% training set and 20% validation set, perform multiple iterations and hyperparameter adjustments, and obtain a pole-mounted transformer insulation damage identification model with improved accuracy and robustness;

[0011] Step 4: Collect videos through drones and extract key frames, input them into the insulation damage recognition model of the pole-mounted transformer, and the model outputs the damage location and degree information;

[0012] Step 5: Determine the nozzle type, build the nozzle 3D model with Solidworks, and import it into Workbench to optimize the mesh quality; perform spraying simulation in Fluent, set the spraying parameters and UAV flight control parameters, and determine the optimal spraying conditions.

[0013] Furthermore, the preprocessing in step one includes eliminating invalid images with a resolution lower than a preset value, enhancing and optimizing valid insulation damage images, and marking the locations and types of damage in the samples accordingly.

[0014] Furthermore, in step 2, the PGI module of the YOLOv9 model is located at the output end of the backbone network to enhance the model's ability to capture target features. Its target capture effect on the image includes the main branch for reasoning architecture, auxiliary reversible branches, and multi-level auxiliary information.

[0015] Furthermore, the MPDIoU loss function in the image recognition model provides a more accurate loss measurement method by directly calculating the key point distance between the predicted box and the true box.

[0016] Furthermore, in step 4, the acquired human-machine video data is segmented into multiple sub-videos using a shot segmentation technique according to a preset pixel threshold, each sub-video takes the tail frame of an adjacent sub-video as the starting frame, and the starting frame and tail frame of each sub-video are input as key frames into the pole-mounted transformer insulation damage recognition model.

[0017] Furthermore, step five specifically includes the following steps:

[0018] S51: Determine the nozzle type; based on the small exposed metal area of ​​the pole-mounted transformer and the requirement of accurate spraying, determine that the nozzle for the spraying operation is a fan-shaped pressure nozzle; wherein the fan-shaped pressure nozzle includes a tubular nozzle body connected to the spraying equipment of the drone body, the tail end of the nozzle body is a spherical head surface, and the top of the spherical head surface is provided with a through hole for spraying paint; the through hole is conical from the inside to the outside, and the apex of the cone is collinear with the diameter where the apex of the spherical head surface is located;

[0019] S52: Determine the parameters of the nozzle; due to the relative cutting depth H of the fan-shaped pressure nozzle f The cutting angle α is an important parameter affecting the spray angle of UAV insulation spraying. The relative cutting depth is inversely proportional to the cutting angle. f The calculation is as follows:

[0020]

[0021] In the above formula, D is the diameter of the ball head surface; H is the height of the conical through hole; h is the thickness corresponding to the conical through hole opened at the ball head surface; R is the radius of the ball head surface; e is the distance between the vertex of the conical through hole and the center of the ball head surface;

[0022] S53: Determine the amount of spray; Analysis and determination of relevant structural parameters affecting the spray amount of fan-shaped fog nozzles According to the principles of fluid mechanics and Bernoulli's equation, the amount of spray Q is equal to the product of the liquid flow rate v and the cross-sectional area F at the water outlet of the spray hole, that is:

[0023]

[0024] In the above formula, g is the acceleration due to gravity; γ is the density of the liquid (kg / m 3 ), p is the liquid pressure, μ is the flow coefficient, and the flow coefficient μ is an important parameter for calculating the flow rate Q, and its calculation formula is as follows:

[0025]

[0026] Where: ∑μ is the sum of nozzle cross-sectional contraction coefficients; μ is the flow coefficient; the relationship between flow Q and flow coefficient μ is fitted through experimental data;

[0027] S54: According to the cutting depth H of the nozzle f , cutting angle α, eccentricity e, ball head surface diameter D and hole edge distance h parameters, build a three-dimensional model of the selected nozzle in Solidworks;

[0028] S55: Use Workbench to mesh the entire model of the pole-mounted transformer spraying. Enter the extracted basin geometry into the large mesh for meshing. Divide the mesh, click Capture Proximity in the Size Adjustment option, and set the minimum size to 2e. -3 m; Select Capture Proximity in the Size Adjustment option to automatically adjust the mesh density and adjust the mesh nodes with poor mesh quality to improve the mesh quality of the overall spray model; After generating the mesh, divide the boundary conditions and name the watershed;

[0029] S56: Change the insulation spraying parameters and conduct a simulated spraying experiment in Fluent to determine the best spraying conditions; the insulation spraying parameters include the cutting angle of the nozzle, the distance between the nozzle and the column, and the spraying angle of the nozzle; by obtaining the coverage cloud map and the trajectory of the spraying particles, the spraying coverage rate is compared to determine the best spraying condition.

[0030] Furthermore, in step S56, the selected grooving angle of the nozzle is 30°, 45° or 60°; the selected spraying angle is 0°, 15° or 30°; and the selected distance between the nozzle and the pole-mounted transformer is 0.5m, 0.65m or 1.0m.

[0031] 3. Beneficial effects:

[0032] (1) The present invention provides an improved YOLOv9 method for identifying and spraying insulation damage on a pole-mounted transformer. The method uses a high-resolution camera or drone to collect images containing insulation damage parts, ensuring that the images are clear and cover the key parts of the transformer. The images are pre-processed by denoising, contrast enhancement, and size normalization to enhance the prominence of damage features, construct a data set containing a variety of damage conditions, and significantly improve the generalization ability of the model.

[0033] (2) The present invention provides an improved YOLOv9 pole-mounted transformer insulation damage identification and spraying method, which improves the neural network model of YOLOv9, adds a programmable gradient information (PGI) module to the output end of the backbone network, and includes a main branch for reasoning and an auxiliary reversible branch to enhance the ability to capture detailed features. These improvements significantly improve the detection performance of the model in complex scenarios, especially the recognition of subtle damage features. At the same time, the loss function of YOLOv9 is optimized, and the MPDIoU loss function is adopted to improve the problems of difficult samples and bounding box regression. Compared with the traditional CIoU loss function, the MPDIoU loss function converges faster and has higher detection accuracy during detection, thereby further improving the performance of the model in complex scenarios.

[0034] (3) The present invention provides an improved YOLOv9 method for identifying and spraying insulation damage on a pole-mounted transformer. In order to adapt to the problem that the proportion of equipment in the pole-mounted transformer image taken by a drone under safety distance restrictions is relatively small, the present invention extracts key frame images, and the recognition accuracy is significantly improved compared to direct input of the original image. This improvement significantly enlarges the proportion of the target in the original image and improves the detection accuracy.

[0035] (4) The present invention selects the appropriate nozzle type based on the data and conclusions of image recognition, and uses Solidworks and ANSYS Workbench to perform detailed three-dimensional modeling and meshing of the spraying model of the nozzle and the pole-mounted transformer, optimizes the mesh quality, and ensures the accuracy and effectiveness of the spraying process. In order to verify the effect of the nozzle, the physical and chemical properties of the spraying material and the flight control parameters of the drone during the spraying operation were simulated in Fluent. By comparing the spraying coverage and spraying particle trajectory under different conditions, the optimal spraying condition was determined. This method not only improves the coverage and uniformity of the spraying, but also significantly reduces the trial and error cost in actual operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is the overall flow chart of this method;

[0037] Figure 2 It is the process of preprocessing the picture in the specific embodiment;

[0038] Figure 3 These are the regression results of two different bounding boxes of the MPDIoU loss function in a specific embodiment;

[0039] Figure 4 This is the YOLOv9 network structure diagram involved in this method;

[0040] Figure 5 It is a pole transformer insulation damage point map output by the pole transformer insulation damage identification model in a specific embodiment;

[0041] Figure 6 It is a cross-sectional view of the fan-shaped pressure nozzle in the present invention;

[0042] Figure 7 It is a three-dimensional model of the fan-shaped pressure nozzle in the present invention;

[0043] Figure 8 It is a schematic diagram of volume extraction in a specific embodiment;

[0044] Fig. 9 A schematic diagram of grid division in a specific embodiment;

[0045] Fig.10This is a diagram of spraying simulation results when the diffusion angle is 30° in a specific embodiment;

[0046] Fig.11 This is a diagram of spraying simulation results when the diffusion angle is 45° in a specific embodiment;

[0047] Fig.12 This is a diagram of spraying simulation results when the diffusion angle is 60° in a specific embodiment;

[0048] Fig.13 This is a diagram of spraying simulation results when the spraying angle is 15° in a specific embodiment;

[0049] Fig.14 This is a spraying simulation result diagram when the spraying angle is 30° in a specific embodiment;

[0050] Fig.15 This is a simulation result diagram when the spraying distance is 0.5m in a specific embodiment. DETAILED DESCRIPTION

[0051] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0052] As attached Figure 1 As shown, the improved YOLOv9 pole transformer insulation damage identification and spraying method is characterized by: including:

[0053] Step 1: Construct a damaged dataset of 10kV pole-mounted transformers; use a high-resolution camera to capture images of pole-mounted transformers and preprocess them to obtain damaged sample images of 0kV pole-mounted transformers, and then construct a damaged dataset;

[0054] Step 2: Build an image recognition model based on YOLOv9; optimize the loss function in the YOLOv9 model to MPDIoU, and use the PGI module of the YOLOv9 model to improve the image detection performance;

[0055] Step 3: Use the damaged data set to train the model in step 2; divide the damaged data set into 80% training set and 20% validation set, perform multiple iterations and hyperparameter adjustments, and obtain a pole-mounted transformer insulation damage identification model with improved accuracy and robustness;

[0056] Step 4: Collect videos through drones and extract key frames, input them into the insulation damage recognition model of the pole-mounted transformer, and the model outputs the damage location and degree information;

[0057] Step 5: Determine the nozzle type, build the nozzle 3D model with Solidworks, and import it into Workbench to optimize the mesh quality; perform spraying simulation in Fluent, set the spraying parameters and UAV flight control parameters, and determine the optimal spraying conditions. Specific embodiment:

[0059] As attached Figure 2 As shown, in this embodiment, when constructing a damaged data set, the collected pictures are high-resolution images, and the pictures are preprocessed by FSRCNN; the super-resolution reconstruction model based on convolutional neural network learns the image mapping relationship through multi-layer convolution and upsampling layers to achieve super-resolution reconstruction. The process is divided into five stages: feature extraction, compression, nonlinear mapping, expansion and deconvolution. The feature extraction layer uses a 5×5 convolution kernel, and the dimension of the output feature map depends on the number of channels of the input image. In the compression stage, a 1×1 convolution kernel is used to reduce the dimension of the feature map. In the nonlinear mapping stage, the activation function is used to capture the complex features in the image and improve the reconstruction quality. In the expansion stage, the number of network layers is increased to capture the texture and detail information of the image more carefully, further improving the reconstruction effect. In the deconvolution stage, the low-resolution image is restored to high resolution through upsampling operation. The processed high-resolution picture is added as a new sample to the existing sample library to obtain an optimized sample database of insulation damage of the pole transformer; the samples in the sample database of insulation damage of the pole transformer are labeled with Labelimg.

[0060] As attached Figure 3 As shown, the regression results of two different bounding boxes of the MPDIoU loss function in the insulation damage identification model of the pole-mounted transformer in this method. In the process of target detection and instance segmentation, the traditional bounding box regression (BBR) loss function is difficult to optimize the situation when the predicted box and the real box have the same aspect ratio but different specific sizes. The present invention introduces a better bounding box similarity measurement method, MPDloU, to solve this problem. MPDIoU is calculated based on the minimum point distance of the horizontal rectangle by comprehensively considering the overlapping area, center point distance, and width and height deviations.

[0061] During the training phase, the loss function is minimized to make the bounding box predicted by the model close to its true bounding box. Each predicted bounding box is:

[0062] B prd =[x prd y prd ω prd h prd ] -1

[0063] Minimize the following loss function to get close to the predicted box and the true box:

[0064] B gt =[x gt y gt ω gt h gt ] -1

[0065] L=min Θ L(B gt ,B prd |Θ)

[0066] Among them: B gt is the set of true bounding boxes, and Θ is the parameter of the regression depth model.

[0067] The expression of MPDIoU is as follows:

[0068]

[0069] in: Represents the sum of squared differences in the x and y coordinates of the upper left corner of the predicted box P and the true box G. Represents the sum of the squared differences between the x and y coordinates of the lower right corner of the predicted box P and the true value box G. The MPDIoU loss function is defined as:

[0070] L MPDIoU =1-f MPDIoU

[0071] MPDloU provides a more accurate loss metric by directly calculating the keypoint distance between the predicted box and the true box.

[0072] Attached Figure 3 Two different bounding box regression results are shown. The green box represents the real bounding box, and the red box represents the predicted bounding box. Comparing the two cases, the loss values ​​calculated by traditional loss functions such as G1oU, DloU, CoU, and EloU are the same, but the loss values ​​calculated by the MPDIoU method are different. This shows that in some special cases, the use of traditional loss functions may not be able to distinguish different prediction results, while MPDIoU can more accurately reflect the difference between the predicted box and the real box, highlighting its superiority in the bounding box regression problem.

[0073] The YOLOv9 model used in this method is as shown in the attached Figure 4 As shown in the figure, the improved YOLOv9 core module is RepNCSPELAN, which uses the image super-resolution reconstruction method (FSRCNN) to replace the traditional downsampling, and the rest of the network structure remains unchanged. At the same time, the loss function is improved to MPDIoU, which can more accurately reflect the difference between the predicted box and the real box, and effectively solves the problem that the traditional bounding box regression loss function is poorly optimized when the aspect ratio is the same but the size is different.

[0074] In this embodiment, the video is segmented into multiple sub-videos using the shot segmentation technology according to the preset pixel threshold. Each sub-video takes the tail frame of the adjacent sub-video as the starting frame, and the starting frame and tail frame of each sub-video are used as key frames. Compared with the direct input of the original image, the key frame extraction technology can greatly improve the recognition accuracy of the insulation damage part of the pole transformer. Finally, these segmented images are input into the pole transformer insulation damage recognition model to output the insulation damage points of the equipment as shown in the attached figure. Figure 5 shown.

[0075] The cross-sectional structure diagram of the fan-shaped pressure nozzle is shown in the attached figure. Figure 6 The constructed three-dimensional model is shown in the attached Figure 7 As shown; through steps S51 to S54, the relevant parameters of the nozzle are determined for simulation.

[0076] As attached Figure 8 As shown in the figure, according to the ratio of 1:1, the volume of the fan-shaped nozzle and the column transformer is extracted, as shown in the attached figure. Fig. 9 As shown, the column transformer spray model is meshed as a whole through Workbench, the geometry of the extracted flow basin is meshed in the large mesh, the mesh is divided, and the capture proximity is clicked in the size adjustment option, and the minimum size is set to 2e -3 m. Select Capture Proximity in the Size Adjustment option to automatically adjust the mesh density and adjust the mesh nodes with poor mesh quality to improve the mesh quality of the overall spray model. After the mesh is generated, the boundary conditions are divided and the flow domain is named.

[0077] The insulating spray coating is assumed to be a sand-stone mixture with a density of 1.2 g / cm 3 , the viscosity is 1.72×10 -3 kg / (ms), the surface tension of the droplet is 0.7N / m, this kind of coating can effectively prevent oxidation and corrosion of copper and aluminum wiring; the flight control parameters for the UAV insulation spraying of the pole-mounted transformer are set as follows.

[0078] (1) Flight altitude. The height of a 10kV electric pole tower is about 10m. To ensure the insulation spraying effect, the flight altitude of the drone must not be lower than the height of the pole transformer, so that the pole transformer can be better treated with insulation spraying from the top and side. However, if the flight altitude is too high, it will cause serious loss of insulation coating and poor economic efficiency. After comprehensive consideration, the present invention proposes that the flight altitude of the drone be 10 to 12m.

[0079] (2) Flight speed. When spraying a single device, the drone should remain hovering in the air, which can better ensure the accuracy of the spraying. After the spraying is completed, the drone needs to adjust its flight posture to spray the next device. At this time, considering safety, the flight speed of the drone should not be too fast. Taking all factors into consideration, the present invention proposes that the flight speed of the drone is 0.8 to 1 m / s.

[0080] (3) Safe distance from the wire. Considering the strong electric field near the wire, the drone nozzle is moved closer to the wire through a telescopic spray boom, and the closest distance between the nozzle and the wire should be greater than 0.3m. At the same time, to ensure the flight safety of the drone, the closest distance between the drone and the wire should be greater than 0.8m.

[0081] (4) Minimum width of drone operation. In order to ensure the uniformity and effect of spraying, the drone will select the corresponding operation width when performing insulation spraying operations to reduce the frequency of repeated operations and improve work efficiency. In this embodiment, the minimum width S of drone spraying is 2m.

[0082] The nozzle diffusion angles were set to (30°, 45°, 60°), the spraying angles were set to (0°, 15°, 30°), and the distances between the nozzle and the pole-mounted transformer were set to (0.5m, 0.65m). Simulation spraying verification was performed to obtain the coverage cloud map and the trajectory of the sprayed particles. The spray coverage was compared to obtain the optimal spraying condition.

[0083] Keep the vertical spraying angle at 0°, that is, vertical spraying. At the same time, according to the on-site construction experience, the distance between the nozzle and the column is set to 0.65m. At this time, the nozzle diffusion angle is set to 30°, 45° and 60° respectively. The simulation spraying simulation experiment is carried out in Fluent. The experimental results are shown in the attached figure. Figures 10-12 As shown. Through simulation experiments, it is found that the nozzle diffusion angle has a significant effect on the spray coverage. When the nozzle diffusion angle is 30°, the spray coverage of the pole-mounted transformer within 1s is 50.0%; when the nozzle diffusion angle is 45° and 60°, the spray coverage drops significantly, only 17.7% and 17.5%. Therefore, in this embodiment, when the nozzle diffusion angle is selected to be 30°, the insulation spraying effect is better.

[0084] Keep the nozzle diffusion angle at 30°, the distance between the nozzle and the column is 0.65m, and set different spraying angles to 0° (i.e. vertical spraying), 15° and 30°, and conduct a simulated spraying experiment in Fluent. The experimental results are shown in the attached figure. Fig.10 , Attachment Fig.13 , Attachment Fig.14As shown. Through simulation experiments, it is found that when the spraying angles are 0°, 15° and 30°, the spraying coverage of the pole-mounted transformer within 1s is 50.0%, 75.0% and 37.3% respectively. It can be concluded that when performing UAV insulation spraying, a certain spraying angle needs to be set, rather than vertical spraying directly above. A spraying angle of 15° is more appropriate.

[0085] Based on the experience of on-site construction, this embodiment sets the distance between the fan-shaped nozzle and the pole-mounted transformer to 0.65m. Since spraying is a live operation, the spraying distance needs to be greater than the safety distance of 0.3m according to the influence of the electric field near the wire. Therefore, under the condition of keeping the nozzle diffusion angle at 30° and the spraying angle at 15°, two groups of distances of 0.5m and 0.65m are set for comparison to verify whether the insulation spray coverage is efficient. The experimental results are shown in the attached figure. Fig.13 , Attachment Fig.15 As shown. Through simulation experiments, it was found that when the fixed nozzle diffusion angle was 30° and the spraying angle was 15°, and the spraying distance was 0.5m and 0.65m respectively, the spraying coverage of the pole-mounted transformer within 1s was 29.8% and 75.0% respectively. Therefore, the present invention concludes that the optimal working condition for the UAV to spray insulation on the pole-mounted transformer is when the nozzle diffusion angle is 30°, the spraying angle is 15° and the distance between the nozzle and the pole-mounted transformer is 0.65m. At this time, the spraying coverage is relatively high. Through multiple 1s point spraying, the pole-mounted transformer, especially the tap, can be effectively insulated to ensure the power supply reliability of the 10kV pole-mounted transformer.

[0086] Although the present invention has been disclosed as above in terms of preferred embodiments, they are not intended to limit the present invention. Anyone skilled in the art can make various changes or modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection defined by the claims of this application.

Claims

1. Improved YOLOv9 pole transformer insulation damage identification and spraying method, characterized by: include: Step 1: Construct a damaged dataset of 10kV pole-mounted transformers; use a high-resolution camera to capture images of pole-mounted transformers and preprocess them to obtain damaged sample images of 0kV pole-mounted transformers, and then construct a damaged dataset; Step 2: Build an image recognition model based on YOLOv9; optimize the loss function in the YOLOv9 model to MPDIoU, and use the PGI module of the YOLOv9 model to improve the image detection performance; Step 3: Use the damaged data set to train the model in step 2; divide the damaged data set into 80% training set and 20% validation set, perform multiple iterations and hyperparameter adjustments, and obtain a pole-mounted transformer insulation damage identification model with improved accuracy and robustness; Step 4: Collect videos through drones and extract key frames, input them into the insulation damage recognition model of the pole-mounted transformer, and the model outputs the damage location and degree information; Step 5: Determine the nozzle type, build the nozzle 3D model with Solidworks, and import it into Workbench to optimize the mesh quality; perform spraying simulation in Fluent, set the spraying parameters and UAV flight control parameters, and determine the optimal spraying conditions.

2. The improved YOLOv9 pole transformer insulation damage identification and spraying method according to claim 1 is characterized in that: The preprocessing in step one includes eliminating invalid images with resolutions lower than a preset value, enhancing and optimizing valid insulation damage images, and marking the locations and types of damage in the samples accordingly.

3. The improved YOLOv9 pole transformer insulation damage identification and spraying method according to claim 2 is characterized in that: In step 2, the PGI module of the YOLOv9 model is located at the output end of the backbone network to enhance the model's ability to capture target features. Its target capture function for the image includes the main branch for reasoning architecture, auxiliary reversible branches, and multi-level auxiliary information.

4. The improved YOLOv9 pole transformer insulation damage identification and spraying method according to claim 3 is characterized in that: The MPDIoU loss function in the image recognition model provides a more accurate loss measurement method by directly calculating the key point distance between the predicted box and the true box.

5. The improved YOLOv9 pole transformer insulation damage identification and spraying method according to claim 4 is characterized in that: In step 4, the acquired human-machine video data is segmented into multiple sub-videos using the lens segmentation technology according to the preset pixel threshold. Each sub-video takes the tail frame of the adjacent sub-video as the starting frame, and the starting frame and tail frame of each sub-video are input as key frames into the pole-mounted transformer insulation damage recognition model.

6. The improved YOLOv9 pole transformer insulation damage identification and spraying method according to claim 1 is characterized in that: Step 5 specifically includes the following steps: S51: Determine the nozzle type; based on the small exposed metal area of ​​the pole-mounted transformer and the requirement of accurate spraying, determine that the nozzle for the spraying operation is a fan-shaped pressure nozzle; wherein the fan-shaped pressure nozzle includes a tubular nozzle body connected to the spraying equipment of the drone body, the tail end of the nozzle body is a spherical head surface, and the top of the spherical head surface is provided with a through hole for spraying paint; the through hole is conical from the inside to the outside, and the apex of the cone is collinear with the diameter where the apex of the spherical head surface is located; S52: Determine the parameters of the nozzle; due to the relative cutting depth H of the fan-shaped pressure nozzle f The cutting angle α is an important parameter affecting the spray angle of UAV insulation spraying. The relative cutting depth is inversely proportional to the cutting angle. f The calculation is as follows: In the above formula, D is the diameter of the ball head surface; H is the height of the conical through hole; h is the thickness of the conical through hole at the ball head surface; R is the radius of the ball head surface; e is the distance between the vertex of the conical through hole and the center of the ball head surface, i.e., the eccentricity; S53: Determine the amount of spray; Analysis and determination of relevant structural parameters affecting the spray amount of fan-shaped fog nozzles According to the principles of fluid mechanics and Bernoulli's equation, the amount of spray Q is equal to the product of the liquid flow rate v and the cross-sectional area F at the water outlet of the spray hole, that is: In the above formula, g is the acceleration due to gravity; γ is the density of the liquid (kg / m 3 ), p is the liquid pressure, μ is the flow coefficient; The flow coefficient μ is an important parameter for calculating the flow rate Q, and its calculation formula is as follows: Where: ∑μ is the sum of nozzle cross-sectional contraction coefficients; μ is the flow coefficient; the relationship between flow Q and flow coefficient μ is fitted through experimental data; S54: According to the cutting depth H of the nozzle f , cutting angle α, eccentricity e, ball head surface diameter D and hole edge distance h parameters, build a three-dimensional model of the selected nozzle in Solidworks; S55: Use Workbench to mesh the entire model of the pole-mounted transformer spraying. Enter the extracted basin geometry into the large mesh for meshing. Divide the mesh, click Capture Proximity in the Size Adjustment option, and set the minimum size to 2e. -3 m; Select Capture Proximity in the Size Adjustment option to automatically adjust the mesh density and adjust the mesh nodes with poor mesh quality to improve the mesh quality of the overall spray model; After generating the mesh, divide the boundary conditions and name the watershed; S56: Change the insulation spraying parameters and conduct a simulated spraying experiment in Fluent to determine the best spraying conditions; the insulation spraying parameters include the cutting angle of the nozzle, the distance between the nozzle and the column, and the spraying angle of the nozzle; by obtaining the coverage cloud map and the trajectory of the spraying particles, the spraying coverage rate is compared to determine the best spraying condition.

7. The improved YOLOv9 pole transformer insulation damage identification and spraying method according to claim 6 is characterized in that: In step S56, the selected grooving angle of the nozzle is 30°, 45° or 60°; the selected spraying angle is 0°, 15° or 30°; the selected distance between the nozzle and the pole-mounted transformer is 0.5m, 0.65m or 1.0m.

Images