Wind wheel posture recognition system and method based on semantic segmentation of wind power wind wheel tower drum image

Abstract

The invention discloses a wind wheel posture recognition system and method based on wind power wind wheel tower drum image semantic segmentation, and the method comprises the steps: recording the operation images of a tower drum and a wind wheel in a wind wheel rotation extension plane through portable mobile shooting equipment, and carrying out the pixel point-level recognition and classification of the tower drum and the wind wheel through a special semantic segmentation model; selecting an image frame with the maximum height difference from the highest point of the edge of the tower drum to the lowest point of the wind wheel, calculating the horizontal distance from each equal height difference point to the tower drum after the image of the blade below the tower top is calculated, recording the distance data of each equal height difference point of each blade, drawing a broken line graph, and calculating the relative deviation percentage; and comparing the morphological differences of the three blades to judge the attitude balance of the wind wheel. According to the method, mechanical angle deviation and blade structural damage defects caused by improper blade zero calibration operation or failure of a zero scale can be effectively identified, and abnormal vibration of a transmission chain caused by imbalance of a wind wheel, output reduction caused by blade asynchronism, accidents caused by blade damage and the like are avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power detection, and in particular to a wind turbine rotor posture recognition system and method based on semantic segmentation of wind turbine rotor tower images. Background Art

[0002] The wind rotor is the main component of a wind turbine that absorbs and converts wind energy. When the three blades are running at different angles, the wind rotor will produce a rotation frequency impact that will affect the mechanical load life of the transmission chain and reduce the wind energy absorption and conversion efficiency; when the blades are severely damaged, it may cause a runaway, tower sweeping, or even a tower collapse accident. Therefore, a comprehensive and effective calculation and evaluation of the wind turbine rotor attitude balance can improve the wind turbine operation stability and wind energy utilization efficiency.

[0003] At present, the wind power field lacks technical methods for wind rotor attitude recognition. The only way to judge the consistency of the blade angle is to periodically calibrate the mechanical zero position of the blade and check the SCADA data of the pitch angle. No further research and analysis has been conducted on the relationship between the mechanical zero position and the blade angle data and the balance of the wind rotor attitude. The specific process of mechanical zero position calibration is to manually control the pitch of a single blade, align the blade root zero position scale line and the hub zero position scale line by visual observation, and this process is recorded as mechanical zero position calibration; then mark the angle measurement sensor such as the pitch motor encoder as 0° at this position, and this process is recorded as electrical zero position calibration. After completing the above operations, during the normal operation of the wind turbine, the pitch angle deviation of the three blades is observed through SCADA to judge the consistency of its pitch action.

[0004] If the mechanical zero position calibration process is performed improperly or the zero position scale fails, the requirement that the mechanical zero position of the three blades is completely consistent cannot be met, and the pitch angle measurement sensor is zeroed on this basis and starts measuring the angle data afterwards, which cannot reflect the actual angle deviation of the three blades.

[0005] The consistency of blade angle cannot fully and reliably reflect the balance of wind rotor attitude, and there is no technical method for wind rotor attitude recognition in the industry at this stage. The only wind turbine blade clearance recognition technology with a certain similar correlation is wind turbine blade clearance recognition technology. This type of technical solution requires the installation of a lidar or industrial camera on the cover below the nacelle or on the ground below the nacelle or on the ground far away from the side of the nacelle, and the real-time monitoring of the distance between the lowest blade tip and the tower is achieved through laser ranging or edge detection methods to avoid tower sweeping accidents. According to the industry accident case report, after a wind turbine blade flying accident occurred in a wind farm, the clearance recognition system failed to issue a warning, and the clearance measurement data could not effectively reflect the signs of blade defect accidents. This type of technical solution has the following defects: 1. Single function and limited role: It can only measure the minimum distance between the blade tip position and the tower, which can only prevent tower sweeping, and cannot effectively reflect the operating status of the wind rotor and other defects. 2. The cost of measurement equipment is high: it is necessary to purchase additional computing and analysis equipment and lidar or industrial cameras, and it is necessary to spend long-term maintenance costs to ensure the basic operation of the equipment. 3. Poor feasibility of on-site implementation: The cabin space is small and the hole in the bottom cover of the cabin needs to consider the danger of falling objects from high altitude and the sealing of the cabin when installing the measurement sensor; the ground installation of the measurement sensor needs to consider the pollution of rain, snow and dust, animal collision interference, human damage and theft, and the additional land acquisition and construction costs. 4. Excessive measurement wastes resources: The life cycle of blade structural parts is long, and the damage and degradation process is measured in years. The uninterrupted operation of the clearance identification equipment is actually in an invalid monitoring state for a long time and causes loss. Summary of the invention

[0006] In view of the lack of wind turbine attitude recognition technology and clearance recognition and other related technical solutions in the wind power field, which have single functions, high costs, poor feasibility, and excessive measurements, a wind turbine attitude recognition system and method based on semantic segmentation of wind turbine tower images is provided. The technical solutions adopted by the present invention are as follows:

[0007] S1. Collect images of the wind rotor tower: Open the shooting function interface of the wind rotor attitude recognition mobile phone software, and shoot a 30-second video image of the entire tower and the complete blades under the tower top in the state of wind rotor rotation on the ground within the wind rotor rotation extension plane. During the shooting process, always keep the tower and the wind rotor in the corresponding reserved dotted line frame. Slight shaking of the mobile phone camera will not affect the recognition result; or operate a drone to hover in the air within the wind rotor rotation extension plane to shoot a 30-second video image of the entire tower and the complete blades under the tower top in the state of wind rotor rotation. Slight shaking of the drone camera will also not affect the recognition result; you can also use other portable mobile shooting devices to complete the shooting and export and upload it to the mobile phone software;

[0008] S2. Establish an image semantic segmentation model: select 20 images of different wind rotor postures for each background interference (such as white clouds, mountain peaks, and strong light) in the video as processing objects, use Labelme software to calibrate the outer edges of the tower and the wind rotor in the image and obtain their labeled data sets, and divide the data sets into training sets and test sets in a 9:1 ratio and complete the training through yolov11-Segmentation to obtain the wind rotor tower image semantic segmentation model and its weight coefficient file;

[0009] S3, selecting a target image frame: using an image semantic segmentation model to compare the captured blade tower video frame by frame, calculating the height difference between the highest pixel point of the tower and the lowest pixel point of the wind rotor in each frame, and selecting the image frame with the largest height difference as the target image frame;

[0010] S4, calculate the horizontal interval of each equal height difference point: in each target image frame, calculate the number of horizontal spacing pixels between each equal height difference point on the trailing edge line segment from the root to the tip of each blade below the tower top (the initial value is 5 points, and the setting value can be changed according to user needs) and each equal height point on the leading edge line segment of the tower image;

[0011] S5. Analyze the wind rotor attitude and output the results: record the time series data of the spacing between each contour difference point in each target image frame and draw a line graph, calculate the relative deviation percentage of the time series data of the spacing between each contour difference point; it is normal when the relative deviation percentage of the horizontal interval of each contour difference point is less than 3%. When it reaches 3% to 8%, the wind rotor tower operation video should be taken again to identify the wind rotor attitude and strengthen the monitoring of the unit's operating status. When it exceeds 8%, the on-site personnel can find the difference in the rotation status of the three blades of the wind rotor through naked eye observation and should immediately stop the machine and climb the tower to check the mechanical zero position consistency of the three blades and the damage of the blade structural parts; store the wind rotor attitude recognition results and the date of execution of this operation to facilitate users to check historical data and analyze the trend of wind rotor attitude changes.

[0012] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0013] 1. Comprehensively and intuitively reflects the balance of the wind rotor attitude. It fills the gap in the technical field of wind rotor operation status evaluation, solves the defects of single function, high cost and poor feasibility of technical solutions such as blade angle consistency and clearance identification, and can effectively identify mechanical angle deviation and blade structural damage defects caused by improper blade zeroing operation or zero scale failure.

[0014] 2. Easy to implement and extremely low cost. There is no need to open holes in the cabin or install any fixed measurement sensors on the ground. A mobile portable camera (such as a personal mobile phone) can be used to capture the image of the tower wind rotor, saving equipment and construction costs. The wind rotor posture recognition is achieved at almost "zero cost", which has excellent promotion value.

[0015] 3. Simple operation and efficient recognition. Use your mobile phone to shoot a video according to the dotted frame reserved in the software shooting interface to complete the image acquisition. After importing it into the wind wheel posture recognition mobile phone software and waiting for the image processing to complete, you can get the recognition result. According to the relative deviation percentage results between the equal height difference points of the blades, you can analyze and judge the problems of the wind wheel and implement corresponding management and inspection measures.

[0016] 4. Strong anti-interference ability and accurate recognition. In view of the environmental background interference images in the process of wind turbine attitude image recognition, such as white clouds, sunshine, cables and other wind turbines, this type of image frames are specially added as training data sets to improve the recognition accuracy and anti-interference ability of the image semantic segmentation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is the shooting interface of the wind rotor attitude recognition software. (The software presets two shooting schemes, namely, the wind rotor is on the left and right of the nacelle; the tower indicator frame and the wind rotor indicator frame are preset in the interface. When shooting, the personnel should stand in the extended plane of the wind rotor rotation and use a portable mobile shooting device to shoot, ensuring that the image of the rotating wind rotor is controlled within the wind rotor indicator frame as much as possible, and the image of the tower is controlled within the tower indicator frame as much as possible. Slight shaking when shooting with handheld will not affect the recognition effect.)

[0018] Figure 2 This is a schematic diagram for creating a tower and wind rotor image segmentation dataset. Figure 2 Use Labelme software to label the tower and wind rotor with polygonal labels and save them in a txt dataset format dedicated to yolo. This step is required for the training of the image semantic segmentation model, and the photographer does not participate in this step.

[0019] Figure 3 is a schematic diagram for explaining how to select a target image frame in a video. Figure 3 In order to use the image semantic segmentation model to compare the wind turbine tower video frame by frame, the height difference between the highest point of the tower and the lowest point of the wind turbine in each frame is calculated, and the image frame with the largest height difference is selected as the target image frame.

[0020] Figure 4 This is a schematic diagram to illustrate the horizontal distances between the various height difference points between wind turbine towers. Figure 4Schematic diagram of the process of calculating the horizontal interval of each height difference point: In each target image frame, calculate the number of horizontal pixel points between each height difference point on the trailing edge line segment from the root to the tip of each blade below the tower top (the initial value is 5 points, and the setting value can be changed according to user needs) and each height point on the leading edge line segment of the tower image.

[0021] Figure 5 This is a schematic diagram illustrating the target image frames of the three blades of the wind wheel and the horizontal intervals of their respective height difference points. Figure 5 The image of the three blades of the wind wheel at a specific position and the horizontal intervals of the equal height difference points in the same rotation cycle are shown.

[0022] Figure 6 This is a schematic diagram of the wind rotor balance evaluation results given by the wind rotor attitude recognition software after completing the analysis and calculation. Figure 6 It is a line graph of the identification data of the blade height difference points in each frame and a table of the relative deviation results of the blade height difference points. ).

Claims

1. A method for wind wheel attitude recognition, characterized in that: include: S1. Collecting wind wheel images, where the wind wheel images include multiple video frames; S2. Establishing an image semantic segmentation model according to the multiple video frames; S3, analyzing the posture of the wind wheel using the image semantic segmentation model to obtain an analysis result; S4. Output the analysis result.

2. The method according to claim 1, characterized in that The establishing of an image semantic segmentation model according to the multiple video frames comprises: Select n video frames with different wind rotor postures from the multiple video frames as processing images, calibrate the outer edge of the tower and the outer edge of the wind rotor in the image and obtain a labeled data set, divide the labeled data set into a training set and a test set, where n is an integer greater than or equal to 2; The target image model is trained and tested based on the training set and the test set respectively to obtain the image semantic segmentation model.

3. The method according to claim 1 or 2, characterized in that: The using the image semantic segmentation model to analyze the posture of the wind wheel and outputting the analysis result includes: The image semantic segmentation model is used to compare the captured blade tower video frame by frame, calculate the height difference between the highest pixel point of the tower and the lowest pixel point of the wind wheel in each frame, and select the image frame with the largest height difference as the target image frame; In the target image frame, the number of horizontal spacing pixels between each equal height difference point on the trailing edge line segment from the blade root to the blade tip image of each blade below the tower top and each equal height point on the leading edge line segment of the tower image is calculated; Record the time series data of the spacing between each height difference point in each target image frame and draw a line graph, and calculate the relative deviation percentage of the time series data of the spacing between each height difference point; The relative deviation percentage is analyzed to obtain the analyzed result.

4. The method according to claim 3, characterized in that If the relative deviation percentage is less than or equal to the first threshold, the wind wheel is normal; If the relative deviation percentage is greater than a first threshold and the relative deviation percentage is less than a second threshold, the wind wheel is abnormal, and the second threshold is greater than the first threshold; If the relative deviation percentage is greater than the second threshold, the wind wheel is damaged.

5. The method according to claim 4, characterized in that The method further comprises: When the relative deviation percentage is greater than the first threshold and the relative deviation percentage is less than the second threshold, the wind rotor tower operation video is filmed again to identify the wind rotor posture and strengthen the monitoring of the wind turbine operation status.

6. The method according to any one of claims 1 to 5, characterized in that The collecting of wind wheel images comprises: The wind wheel image is collected through the shooting function interface of the wind wheel posture recognition mobile phone software; or, The wind wheel image is collected by a drone.

7. A device for wind wheel attitude recognition, characterized in that: include: A processor, the processor is used to collect wind rotor images, the tower wind rotor images include multiple video frames; establish an image semantic segmentation model according to the multiple video frames; use the image semantic segmentation model to analyze the posture of the wind rotor to obtain an analysis result; An output interface is used to output the analyzed result.

8. The device according to claim 7, characterized in that The processor is specifically used for: Select n video frames with different wind rotor postures from the multiple video frames as processing images, calibrate the outer edge of the tower and the outer edge of the wind rotor in the image and obtain a labeled data set, divide the labeled data set into a training set and a test set, where n is an integer greater than or equal to 2; The image semantic segmentation model is obtained by completing training based on the training set and the test set through yolov11-Segmentation.

9. The device according to claim 7 or 8, characterized in that The processor is specifically used for: The image semantic segmentation model is used to compare the captured blade tower video frame by frame, calculate the height difference between the highest pixel point of the tower and the lowest pixel point of the wind wheel in each frame, and select the image frame with the largest height difference as the target image frame; In the target image frame, the number of horizontal spacing pixels between each equal height difference point on the trailing edge line segment from the blade root to the blade tip image of each blade below the tower top and each equal height point on the leading edge line segment of the tower image is calculated; Record the time series data of the spacing between each height difference point in each target image frame and draw a line graph, and calculate the relative deviation percentage of the time series data of the spacing between each height difference point; The relative deviation percentage is analyzed to obtain the analyzed result.

10. The device according to claim 9, characterized in that If the relative deviation percentage is less than or equal to the first threshold, the wind wheel is normal; If the relative deviation percentage is greater than a first threshold and the relative deviation percentage is less than a second threshold, the wind wheel is abnormal, and the second threshold is greater than the first threshold; If the relative deviation percentage is greater than the second threshold, the wind wheel is damaged.

11. The device according to claim 10, characterized in that The processor is further configured to: When the relative deviation percentage is greater than the first threshold and the relative deviation percentage is less than the second threshold, the wind rotor tower operation video is filmed again to identify the wind rotor posture and strengthen the monitoring of the unit operation status.

12. The device according to any one of claims 7 to 11, characterized in that The processor is arranged on the mobile phone, and is used to collect the wind wheel image through the shooting function interface of the wind wheel posture recognition mobile phone software; or, The processor is arranged on the drone, and is used for collecting the wind wheel image through the drone.

13. A computer-readable storage medium, characterized in that: The storage medium stores a computer program or instruction. When the computer program or instruction is executed by a computer, the method according to any one of claims 1 to 6 is implemented.

14. A computer program product, characterized in that It comprises a computer program or an instruction, which, when executed by a wind wheel attitude recognition device, implements the method according to any one of claims 1 to 6.

Images