Visual detection device and method for looseness of tower bolt of wind driven generator

Abstract

The invention discloses a tower bolt looseness visual detection device and method for a wind driven generator, and the device can effectively absorb the vibration from the horizontal direction through a horizontal damping spring, and can effectively absorb the vibration from the vertical direction through a vertical damping spring. Vibration and shaking between the camera base and the camera body are reduced and controlled through the mass block, the sling and the damping assembly, and therefore the video shooting module can obtain clearer images. According to the tower drum bolt looseness visual detection method, the problem of single feature misinformation can be solved through a double-threshold cooperative judgment strategy in a complex and changeable fan operation environment, the detection precision is greatly improved, whether the bolt is loosened or not is accurately recognized, and the detection efficiency is improved. A firm and powerful support is provided for timely carrying out bolt loosening repair work and guaranteeing continuous and stable operation of the fan, the unnecessary manual inspection problem is avoided, the labor cost is reduced, and the manual inspection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of image communication technology, and in particular to a device and method for visually detecting loose tower bolts of a wind turbine. Background Art

[0002] With the increasing awareness of environmental protection, clean energy has developed rapidly. Among them, wind power generation has attracted much attention as a clean energy source. Wind power generation relies on wind turbines to convert wind energy into mechanical work, which drives the rotor to rotate and ultimately outputs alternating current.

[0003] As the support system for the entire wind turbine, the tower of a wind turbine plays a crucial role in ensuring its safe and reliable operation. The tower bears the weight of the nacelle and blades, as well as the horizontal wind load. Due to the time-varying nature of wind speed, wind turbines operate under alternating load conditions. As operating time increases, the tower's connecting bolts, subjected to alternating stresses, are susceptible to fatigue failure. Failure to detect these defects during scheduled inspections can have serious consequences. Therefore, the stability of the tower's bolted connections directly impacts the safe operation of the wind turbine.

[0004] Currently, wind turbine tower bolt loosening detection generally relies on manual inspections, which is not only inefficient but also lacks real-time performance. Consequently, a small number of more intelligent wind turbines are beginning to use sensors (such as vibration sensors) to monitor tower bolt loosening. However, since detection accuracy can only be guaranteed with a large number of sensors, the installation cost is extremely high.

[0005] In addition, there is usually large vibration during the operation of wind turbines. At this time, the bolts are in a non-static state, and there is a large deviation between the center point and angle of the initial position and the vibration position. The data is unreliable. If only the existing visual recognition technology is used to judge whether the bolts are loose, the frequency of false alarms is high, especially when a non-loose bolt is mistakenly reported as loose, which will lead to an unnecessary manual inspection, which is time-consuming and labor-intensive.

[0006] Therefore, there is an urgent need to develop a high-precision, real-time, and low-cost tower bolt loosening detection device and method. Summary of the Invention

[0007] In order to solve the technical problems of low precision, poor real-time performance and high cost in existing tower bolt loosening detection devices and methods, the present invention provides a wind turbine tower bolt loosening visual detection device and method.

[0008] The technical solution is as follows: The first aspect of the present application relates to a visual detection device for loose tower bolts of a wind turbine, comprising a camera base and a camera body arranged at the bottom of the camera base, a video shooting module being installed on the camera body, an upper mounting seat being fixedly installed at the bottom of the camera base, a top of the camera body being fixedly installed at the lower mounting seat, a middle mounting seat being arranged between the upper mounting seat and the lower mounting seat, a mounting seat through-hole being opened on the middle mounting seat, at least three axes along the center axis of the mounting seat through-hole are passed between the upper mounting seat and the middle mounting seat. The camera base is connected by evenly distributed horizontal vibration-damping springs, each of which extends horizontally perpendicular to the radial direction of the mounting seat through-hole. The middle mounting seat and the lower mounting seat are connected by at least three vertical vibration-damping springs evenly distributed along the central axis of the mounting seat through-hole, and each vertical vibration-damping spring extends in the vertical direction. A mass block is suspended at the bottom of the camera base by a sling, and the sling extends downward along the central axis of the mounting seat through-hole. The mass block is supported on the upper surface of the lower mounting seat by at least three groups of damping components evenly distributed along its circumference.

[0009] The above-mentioned visual inspection device for loose bolts on the tower of a wind turbine was installed inside the tower of the wind turbine, achieving the following technical effects: 1. A 360° panoramic monitoring image of all tower bolts can be obtained through the video shooting module. The 360° panoramic monitoring image can be used to accurately identify whether the bolts are loose using the tower bolt loosening visual detection method; 2. Horizontal vibration damping springs can effectively absorb circumferential vibrations from the horizontal direction, while vertical vibration damping springs can effectively absorb vertical vibrations, enabling the video capture module to obtain clearer images, thereby improving the accuracy of visual inspection of tower bolt loosening. 3. A mass pendulum is formed by using a mass block and a sling, and then the mass block is supported on the lower mounting seat through a damping component. This can effectively absorb radial kinetic energy from the horizontal direction, reduce and control the vibration and shaking between the camera base and the camera body, thereby enhancing the stability of the structure, further enabling the video shooting module to obtain clearer images, and improving the accuracy of visual detection of loose tower bolts.

[0010] A second aspect of the present application relates to a method for visually detecting loose tower bolts of a wind turbine, which is performed according to the following steps: S1. Use a dataset of known bolt images to train a YOLOv11-OBB model using a machine learning algorithm until the YOLOv11-OBB model achieves a set accuracy for bolt annotation using feature-oriented bounding boxes. Then use the YOLOv11-OBB model as the object detection model. S2. Using the above-mentioned visual detection device for loose tower bolts, collect a panoramic image of the interior of the wind turbine tower, where the panoramic image is a 360° panoramic monitoring image including all tower bolts; S3, performing cropping, scaling, denoising, normalization, and binarization processing on the panoramic image to obtain a preprocessed image; S4. Using the feature oriented bounding box of the target detection model to mark each bolt in the preprocessed image, and extracting the feature oriented bounding box parameters of each bolt, thereby calculating the feature oriented bounding box information of each bolt based on the feature oriented bounding box parameters of each bolt; S5. Determine whether the bolt is loose based on the characteristic oriented bounding box information of the bolt, wherein the characteristic oriented bounding box information of the bolt includes the displacement of the center point of the bolt. and the bolt rotation angle ; To determine whether the bolts are loose, follow these steps: Displacement judgment: Set the displacement threshold of the center point of the bolt ,when When , it is marked as normal displacement; when When , it is marked as fan shaking; Rotation angle judgment: set the rotation angle threshold of the bolt ,when When , it is marked as normal rotation angle; when When , it is marked as potentially loose; Joint decision-making: and , it is judged that the bolt is loose.

[0011] The above-mentioned visual detection method for loose tower bolts not only has all the advantages of the above-mentioned visual detection device for loose tower bolts, but also, through the feature-oriented bounding box detection technology of the YOLOv11-OBB model, based on the training of a large number of samples, can not only accurately locate the bolts, but also significantly improve the detection accuracy and intelligence level of loose bolts. It can accurately identify whether the bolts are loose in the complex and changeable wind turbine operating environment, providing solid and powerful support for timely carrying out loose bolt repair work and ensuring the continuous and stable operation of the wind turbine. At the same time, compared with the sensor detection method, only one visual detection device for loose tower bolts is needed to detect all bolts, which greatly reduces the investment cost; and, through the above-mentioned dual-threshold collaborative judgment strategy, the problem of false alarm of a single feature is solved, the detection accuracy is greatly improved, and unnecessary manual inspections can be avoided, reducing labor costs and improving the efficiency of manual inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1This is a schematic diagram of the structure of a visual detection device for loose tower bolts; Figure 2 This is a cross-sectional view of a visual inspection device for loose tower bolts; Figure 3 This is a structural diagram of the vibration reduction mechanism from one perspective; Figure 4 This is a structural diagram of the vibration reduction mechanism from another perspective; Figure 5 is a cross-sectional view of the vibration reduction mechanism; Figure 6 Schematic diagram of the installation of the tower bolt loosening visual detection device inside the tower; Figure 7 is the current angle of the bolt Less than Schematic diagram of the rotation of the feature orientation bounding box when ; Figure 8 is the current angle of the bolt Greater than , less than Schematic diagram of the rotation of the feature orientation bounding box when ; Figure 9 This is the average precision mean detection effect diagram. DETAILED DESCRIPTION

[0013] The present invention will be further described below with reference to the embodiments and accompanying drawings.

[0014] Example 1: like Figure 1-Figure 5 As shown, a visual detection device for loose tower bolts of a wind turbine is provided, which mainly comprises a camera base 1 and a camera body 2 arranged at the bottom of the camera base 1. The camera base 1 is installed inside the tower in the form of a bracket, making the visual detection device for loose tower bolts more stable and reliable.

[0015] The camera body 2 includes a main mounting base 2a disposed at the bottom of the camera base 1 and a pan / tilt head 2b mounted at the bottom of the main mounting base 2a. There is no direct connection between the camera base 1 and the main mounting base 2a. The pan / tilt head 2b is mounted on a video capture module. Specifically, the video capture module includes a main camera 11 mounted at the bottom of the pan / tilt head 2b and a plurality of auxiliary cameras 12 evenly distributed along the circumference of the main mounting base 2a. The main camera 11 faces downward, while the auxiliary cameras 12 each face downward and diagonally outward.

[0016] See Figure 6The main camera 11 features high resolution, high resolution, and multiple zoom, making captured images clearer. Each auxiliary camera 12 can capture images from different angles. Therefore, when the visual inspection device for loose tower bolts is installed inside a wind turbine tower, a 360° panoramic monitoring image of all tower bolts can be spliced ​​together, enabling comprehensive identification of loose bolts.

[0017] Furthermore, due to the dim environment inside the tower, please refer to Figure 1 In this embodiment, a plurality of fill lights 13 are provided at the bottom of the gimbal 2b and are evenly distributed around the main camera 11 along the circumferential direction, thereby effectively improving the ambient brightness, reducing noise and blur problems, and making the captured image clearer.

[0018] See Figure 1-Figure 5 A vibration reduction mechanism mounting cavity 3 is formed between the camera base 1 and the camera body 2. Specifically, the bottom of the camera base 1 is concave to form an upper cavity, and the top of the main body mounting seat 2a is concave to form a lower cavity. The upper cavity and the lower cavity together constitute the vibration reduction mechanism mounting cavity 3. The vibration reduction mechanism mounting cavity 3 is provided with a vibration reduction mechanism. The vibration reduction mechanism includes an upper mounting seat 4 fixedly mounted on the camera base 1, a lower mounting seat 5 fixedly mounted on the main body mounting seat 2a, and a middle mounting seat 6 arranged between the upper mounting seat 4 and the lower mounting seat 5. The middle mounting seat 6 is provided with a mounting seat through hole 6a. In this embodiment, the upper mounting seat 4, the lower mounting seat 5, and the middle mounting seat 6 are all disc structures and are coaxially arranged in sequence from top to bottom. The mounting seat through hole 6a is a circular structure and is located at the center of the middle mounting seat 6. The upper mounting seat 4 is fixedly mounted on the cavity wall of the upper cavity through the upper bracket 4a, and the lower mounting seat 5 is fixedly mounted on the cavity wall of the lower cavity through the lower bracket 5b, thereby ensuring the reliability of the installation.

[0019] At least three horizontal vibration-damping springs 7 evenly distributed along the central axis of the mounting seat through hole 6a are fixedly connected between the upper mounting seat 4 and the middle mounting seat 6. Each horizontal vibration-damping spring 7 extends in the horizontal direction, and the extension direction of each horizontal vibration-damping spring 7 is perpendicular to the radial direction of the mounting seat through hole 6a, that is: the top of the circumferential outer wall of each horizontal vibration-damping spring 7 is fixedly connected to the bottom surface of the upper mounting seat 4, and the bottom of the circumferential outer wall of each horizontal vibration-damping spring 7 is fixedly connected to the top surface of the middle mounting seat 6.

[0020] At least three vertical damping springs 8 are fixedly connected between the lower mounting seat 5 and the middle mounting seat 6. They are evenly distributed along the central axis of the mounting seat through-hole 6a. Each vertical damping spring 8 extends vertically. Specifically, the upper end of each vertical damping spring 8 is fixedly connected to the bottom surface of the middle mounting seat 6, and the lower end of each vertical damping spring 8 is fixedly connected to the top surface of the lower mounting seat 5.

[0021] Therefore, by setting up a vibration damping mechanism, each horizontal vibration damping spring 7 can effectively absorb vibrations from the horizontal direction, and each vertical vibration damping spring 8 can effectively absorb vibrations from the vertical direction, so that the video shooting module can obtain clearer images, thereby improving the accuracy of visual detection of loose tower bolts.

[0022] In this embodiment, a mass 15 is suspended from the bottom of the camera base 1 via a sling 14. The sling 14 extends downward along the central axis of the mounting base through-hole 6a. Specifically, the upper end of the sling 14 is connected to the bottom of the camera base 1, while the lower end of the sling 14 is connected to the top of the mass 15. The mass 15 and the sling 14 form a pendulum mass. Simultaneously, the mass 15 is supported on the upper surface of the lower mounting base 5 by at least three damping assemblies evenly distributed along its circumference. Typically, the damping assemblies connect to the mass 15 at locations below the mass 15. Therefore, the sling 14, mass 15, and damping assemblies form a tuned mass damper that effectively absorbs radial kinetic energy from the horizontal direction, reducing and controlling vibration and shake between the camera base 1 and the camera body 2. This enhances structural stability, further enabling the video capture module to capture clearer images and improving the accuracy of visual inspection for loose tower bolts.

[0023] Furthermore, each damping assembly consists of a damper 16 and an air spring 17. The upper and lower ends of each damper 16 and air spring 17 are hinged to the mass 15 and the lower mounting base 5, respectively. The hinged mounting arrangement of the dampers 16 and air springs 17 ensures the stability of the tuned mass damper system. The dampers 16 damp the oscillation of the mass 15, allowing the frequency of the pendulum mass to be adjusted by using dampers 16 with different damping forces. The air springs 17 adjust the frequency of the pendulum mass, allowing the frequency of the pendulum mass to be adjusted by using air springs 17 with different spring forces. Therefore, by adding air springs 17, the length of the sling 14 can be freely adjusted according to the specific usage scenario, increasing the flexibility of the tuned mass damper arrangement. Furthermore, because each set of dampers 16 and air springs 17 is positioned adjacent to each other, they provide better coordination, further reducing and controlling vibration and shaking between the camera base 1 and the camera body 2.

[0024] Furthermore, the dampers 16 and air springs 17 of each damping assembly are arranged in an "eight" shape, thereby improving the stability of the structure.

[0025] See Figure 4 and Figure 5At least three guide safety posts 9 extending vertically downward are fixedly connected to the bottom of the middle mounting seat 6. Safety post holes 5a are defined in the lower mounting seat 5, corresponding to each guide safety post 9. The lower end of each guide safety post 9 passes through the corresponding safety post hole 5a and is then expanded to form a support head 9a having a larger diameter than the safety post hole 5a. Under normal conditions, the lower end of each vertical damping spring 8 is fixedly connected to the upper surface of the lower mounting seat 5, positioning the lower mounting seat 5 above each support head 9a. A gap remains between the lower surface of the lower mounting seat 5 and the upper surface of each support head 9a. If a vertical damping spring 8 breaks, the lower mounting seat 5 is supported on the corresponding support head 9a at the location of the broken vertical damping spring 8. In extreme cases, if all vertical damping springs 8 break, the lower mounting seat 5 is simultaneously supported on all support heads 9a. Through such a design, not only the dynamic stability of the lower mounting seat 5 during the vibration reduction process is guaranteed, but also even if the vertical vibration reduction spring 8 breaks after long-term use, each guide safety column 9 can support the lower mounting seat 5 through the support head 9a, thereby preventing the camera body 2 from falling. In this case, although the vibration reduction effect is deteriorated, the video shooting module can still work normally, which plays a safety role.

[0026] Furthermore, the number of guide safety columns 9 is preferably the same as the number of vertical vibration-damping springs 8, and each vertical vibration-damping spring 8 is mounted on each guide safety column 9 in a one-to-one correspondence, thereby further improving the dynamic stability of the lower mounting seat 5 during the vibration reduction process.

[0027] See Figure 3 The top and / or bottom of the horizontal vibration damping spring 7 are fixedly connected with auxiliary mounting ribs 10 extending along the length direction thereof. Specifically, when only the top of the horizontal vibration damping spring 7 is installed with the auxiliary mounting ribs 10, the auxiliary mounting ribs 10 are first welded to the top of the horizontal vibration damping spring 7, and then welded and fixed to the upper mounting seat 4 through the auxiliary mounting ribs 10; when only the bottom of the horizontal vibration damping spring 7 is installed with the auxiliary mounting ribs 10, the auxiliary mounting ribs 10 are first welded to the bottom of the horizontal vibration damping spring 7, and then welded and fixed to the middle mounting seat 6 through the auxiliary mounting ribs 10; when both the top and bottom of the horizontal vibration damping spring 7 are installed with auxiliary mounting ribs 10, two auxiliary mounting ribs 10 are first welded to the top and bottom of the horizontal vibration damping spring 7, respectively, and then welded and fixed to the upper mounting seat 4 and the middle mounting seat 6 through the auxiliary mounting ribs 10; through the above design, the reliability and stability of the installation of the horizontal vibration damping spring 7 are guaranteed, and the service life is extended.

[0028] Example 2: A method for visually detecting loose tower bolts of a wind turbine is performed in the following steps: S1. Use a dataset of images of known bolts and train the YOLOv11-OBB model using a machine learning algorithm until the accuracy of the YOLOv11-OBB model in labeling bolts using feature-oriented bounding boxes reaches the set standard. Then use the YOLOv11-OBB model as the object detection model.

[0029] Specifically, step S1 is performed according to the following steps: S11. Data collection.

[0030] The tower bolt loosening visual inspection device of Example 1 is used to collect multiple sets of images of intact and loosened bolts at different angles, distances, tightness states, numbers, and interference environments. Typically, at least 10,000 images of intact bolts and at least 10,000 images of loosened bolts are collected.

[0031] S12. Data labeling.

[0032] Use a data annotation tool (such as label-studio) to annotate the feature orientation bounding box information of the bolts in all the images of non-loose bolts and loose bolts in step S11 to obtain an image dataset of known bolts.

[0033] The marking information includes: the center point coordinates, width, height, angle, and looseness label of the bolt.

[0034] S13. Data division.

[0035] The image dataset of known bolts is divided into a training set and a validation set. The ratio of the training set to the validation set is preferably 8:2 to ensure that the trained model has good generalization ability and does not suffer from overfitting problems.

[0036] S14. Model training.

[0037] First, the training set is input into the YOLOv11-OBB model for training, and the labeled target image is output. Then, the labeled target image is evaluated using the validation set. If it fails, the next round of training is carried out. If it passes, the training is terminated and the YOLOv11-OBB model is used as the target detection model.

[0038] The parameters used to evaluate the labeled target images using the validation set include precision, recall, and average precision.

[0039] S2. Using the visual detection device for loose tower bolts of Example 1, the video capture module captures a panoramic image of the interior of the wind turbine tower. This panoramic image is a 360° panoramic monitoring image that includes all tower bolts. Specifically, a main camera 11 with high pixel count, high resolution, and multiple zoom capabilities is used, and preferably eight auxiliary cameras 12 are used to capture images from various angles, thereby stitching together a 360° panoramic monitoring image that includes all tower bolts.

[0040] S3. The panoramic image is cropped, scaled, denoised, normalized, and binarized to obtain a preprocessed image. This reduces image noise and improves the contrast and clarity of the bolts, improving data quality for subsequent model adaptation.

[0041] Specifically, step S3 is performed according to the following steps: S31. Crop the 360° panoramic monitoring image.

[0042] Remove irrelevant areas from the 360° panoramic monitoring image and focus on the area containing the target bolt to reduce computational redundancy. Typically, a 50-100 pixel boundary area is expanded around the bolt to ensure complete coverage of the target.

[0043] S32: scaling the cropped image.

[0044] Unify the image size to the model input size to meet the network structure requirements. Typically, bilinear interpolation is used to resize the image, with the input size set to 640×640 to maintain the target ratio and avoid distortion.

[0045] S33: De-noise the scaled image.

[0046] Use Gaussian filtering algorithm to eliminate random noise in the image caused by imaging system or environmental factors, and improve image clarity and feature recognition.

[0047] S34, performing normalization processing on the denoised image.

[0048] Linear normalization is typically used to normalize denoised images, mapping each pixel value in the image from its original range of [0, 255] to the range of [0, 1]. Image normalization is used to standardize the pixel values ​​of the input image to a uniform range, improving model training stability and inference consistency.

[0049] S35. Perform binarization processing on the normalized image.

[0050] By setting a fixed threshold, the grayscale image is converted into a black and white binary image for subsequent edge extraction and angle estimation. This binarization process enhances the edge features of the target bolt in the image. By extracting the edge contour of the target bolt, the displacement and rotation angle of the target bolt can be accurately estimated.

[0051] S4. Use the target detection model to extract the characteristic orientation bounding box parameters of the bolt in the preprocessed image, and use this to calculate the characteristic orientation bounding box information of the bolt.

[0052] Specifically, in step S4, see Figure 7 and Figure 8 , use the target detection model to detect the preprocessed image, make the four sides of the rectangular bounding box tangent to the edge contour of the target bolt after binary processing, obtain the rectangular bounding box information around the bolt, and obtain the four vertex coordinates of the rectangular bounding box in the order of top, right, bottom, and left image orientation ( , )、( , )、( , )、( , ). Specifically, the coordinates are ( , ) is located at the top of the image, with coordinates ( , ) is located at the rightmost position of the image, and its coordinates are ( , ) is located at the bottom of the image, with coordinates ( , ) is located at the leftmost position of the image. That is: 、 、 and middle, is the maximum value, is the minimum value; 、 、 and middle, is the maximum value, is the minimum value.

[0053] The current center point coordinates of the bolt are calculated by the coordinates of the four vertices of the rectangular bounding box ( , ), the formula is as follows: ; ; Set the initial center point coordinates when the bolt is installed to ( , ), the center point displacement of the bolt for: ; The calculated coordinates are ( , )’s vertices and coordinates are ( , )'s vertices : ; The calculated coordinates are ( , )’s vertices and coordinates are ( , )'s vertices : ; Set the initial angle when the bolt is installed to , which is a known parameter. Set the current angle of the bolt to Specifically, the initial angle of the bolt after installation is The lower half axis of the major axis of the edge contour after binarization processing when the target bolt is installed is xoy Coordinate system x Angle of the positive axis, see Figure 7 and Figure 8 , the current angle of the bolt is The lower half axis of the long axis of the edge contour after binary processing of the target bolt in the current state is xoy Coordinate system x Angle from the positive axis.

[0054] Therefore, the current angle of the bolt is calculated as follows: : See Figure 7 ,when ≥ The current angle of the bolt Less than , the current angle of the bolt is calculated by the following formula : ; See Figure 8 ,when The current angle of the bolt Greater than , less than , the current angle of the bolt is calculated by the following formula : ; Therefore, the rotation angle of the bolt It can be directly calculated by the following formula: .

[0055] S5. Determine whether the bolt is loose based on the characteristic oriented bounding box information of the bolt.

[0056] Due to the significant vibrations present during wind turbine operation, the bolts are not stationary. Consequently, there is a significant deviation between the center point and angle of the initial position and the vibration position, making the data unreliable. In this embodiment, images of the bolts are captured when they have returned to their initial position. At this point, the center point displacement value obtained through machine vision is typically not equal to 0. Through training, reliable thresholds for center displacement and angular rotation are obtained to prepare for policy decisions.

[0057] Center point displacement threshold: By training the model on multiple bolts, a center point displacement threshold is determined. When the center point displacement exceeds this threshold, it indicates that the fan is vibrating and deviating significantly, and the bolt is no longer considered loose. If the center point displacement does not exceed the threshold, the bolt is further determined to be loose based on the angle deviation.

[0058] Angle threshold determination: By training the model on multiple bolts, an angle threshold is obtained; when the rotation angle of the bolt exceeds the threshold, the bolt is determined to be loose.

[0059] Finally, through calculation and comparison with historical data, a comprehensive judgment is made based on the center point and angle displacement thresholds. When the center point displacement does not exceed the set threshold and the angle deviation of the bolt exceeds the respective set thresholds, the bolt is judged to be in a loose state.

[0060] Specifically, step S5 includes: Displacement judgment: Set the displacement threshold of the center point of the bolt ,when When , it is marked as normal displacement; when When , it is marked as fan shaking; Rotation angle judgment: set the rotation angle threshold of the bolt ,when When , it is marked as normal rotation angle; when When , it is marked as potentially loose; Joint decision-making: and , it is judged that the bolt is loose.

[0061] Therefore, the above-mentioned dual-threshold collaborative judgment strategy solves the problem of single-feature false alarms and greatly improves the detection accuracy.

[0062] The inspection results are then visualized and a report generated. When loose bolts are detected, an alarm is immediately issued, along with corresponding loosening images, real-time video, and information such as the unit and tower location. Alarms are triggered on the backend management server and relevant personnel are notified.

[0063] Comparative Example: The bolt images were labeled and trained. The overall training environment was Pycharm. During the training process, 16 images were used for each training, with a total of 300 training times, a training image size of 640, 8 data loading threads, a weight decay coefficient of 0.0005, and a momentum parameter of 0.9. Indicators such as precision, recall, and average precision were used to evaluate the bolt recognition and detection effect. Figure 9 As can be seen, the YOLOv11-OBB object detection model is able to generate rectangular boxes around all the bolts in the image. Table 1 below shows that the YOLOv11-OBB object detection model achieves a precision of 95.4%, a recall of 88.7%, and an average precision of 93.9%, demonstrating high bolt recognition accuracy.

[0064] Table 1 Detection effect table of YOLOv11-OBB target detection model

[0065] The video capture module captures images of a bolt in five states: initial position, vibration position, vibration recovery position, rotation vibration position, and rotation vibration recovery position. The YOLOv11-OBB target detection model detects the bolt data in each state, as shown in Table 2. After multiple trainings, the center point displacement is set. Pixel, angular displacement When the bolt is in a vibrating state, the center point displacement is 46.22 pixels, which is greater than the threshold, so the bolt is no longer judged to be loose. When the bolt returns to a stationary state after vibration, the center point displacement is 1.28 pixels, which is less than the threshold, and the angular displacement is 0.2°, which is less than the threshold, indicating that the bolt is not loose. When the bolt is rotating and in a vibrating state, the center point displacement is 29.12 pixels, which is greater than the threshold, so the bolt is no longer judged to be loose. When the rotating bolt returns to a stationary state after vibration, the center point displacement is 3.87 pixels, which is less than the threshold, and the angular displacement is 9.2°, which is greater than the threshold, indicating that the bolt is loose.

[0066] Table 2 Results of bolt loosening detection by YOLOv11-OBB target detection model

[0067] It can be seen from Table 2 above that the tower bolt loosening visual detection method of this embodiment can better detect the loose state of the bolts, which helps to improve the detection accuracy and intelligence level of the bolt loosening.

[0068] Finally, it should be noted that the above description is only a preferred embodiment of the present invention. Under the guidance of the present invention, ordinary technicians in this field can make various similar expressions without violating the purpose and claims of the present invention. Such changes fall within the scope of protection of the present invention.

Claims

1. A visual detection device for loose tower bolts of a wind turbine, comprising a camera base (1) and a camera body (2) arranged at the bottom of the camera base (1), wherein a video shooting module is installed on the camera body (2), and the device is characterized in that: An upper mounting seat (4) is fixedly mounted on the bottom of the camera base (1), and a lower mounting seat (5) is fixedly mounted on the top of the camera body (2). A middle mounting seat (6) is provided between the upper mounting seat (4) and the lower mounting seat (5). A mounting seat through hole (6a) is provided on the middle mounting seat (6). The upper mounting seat (4) and the middle mounting seat (6) are connected by at least three horizontal vibration-damping springs (7) uniformly distributed along the central axis of the mounting seat through hole (6a). Each horizontal vibration-damping spring (7) is perpendicular to the radial direction of the mounting seat through hole (6a). The middle mounting seat (6) and the lower mounting seat (5) are connected by at least three vertical vibration-damping springs (8) uniformly distributed along the central axis of the mounting seat through hole (6a), and each vertical vibration-damping spring (8) extends in the vertical direction. A mass block (15) is suspended at the bottom of the camera base (1) through a sling (14), and the sling (14) extends downward along the central axis of the mounting seat through hole (6a). The mass block (15) is supported on the upper surface of the lower mounting seat (5) by at least three groups of damping components uniformly distributed along its circumference.

2. The visual detection device for loose tower bolts according to claim 1, characterized in that: The damping components are composed of a damper (16) and an air spring (17), and the upper and lower ends of each damper (16) and each air spring (17) are respectively hinged on the mass block (15) and the lower mounting seat (5).

3. The visual detection device for loose tower bolts according to claim 2, characterized in that: The damper (16) and the air spring (17) of the damping assembly are both arranged in an "eight" shape.

4. The visual detection device for loose tower bolts according to claim 1, characterized in that: The bottom of the middle mounting seat (6) is fixedly connected to at least three guide safety columns (9) extending downward in the vertical direction. The lower mounting seat (5) is provided with safety column through holes (5a) corresponding to each guide safety column (9). The lower end of each guide safety column (9) passes through the corresponding safety column through hole (5a) and is then expanded to form a support head (9a) having a diameter larger than the safety column through hole (5a). A gap is left between the lower surface of the lower mounting seat (5) and the upper surface of each support head (9a).

5. The tower bolt loosening visual detection device according to claim 2, characterized in that: The number of the guide safety columns (9) is the same as the number of the vertical vibration damping springs (8), and each vertical vibration damping spring (8) is sleeved on each guide safety column (9) in a one-to-one correspondence.

6. The tower bolt loosening visual detection device according to claim 1, characterized in that: The top and / or bottom of the horizontal vibration damping spring (7) are fixedly connected to auxiliary mounting ribs (10) extending along the length direction thereof, and a side surface of each auxiliary mounting rib (10) away from the horizontal vibration damping spring (7) is fixedly connected to the corresponding upper mounting seat (4) or middle mounting seat (6).

7. The tower bolt loosening visual detection device according to claim 1, characterized in that: The camera body (2) includes a main mounting seat (2a) connected to a lower mounting seat (5) and a pan / tilt platform (2b) mounted at the bottom of the main mounting seat (2a). The video shooting module includes a main camera (11) mounted at the bottom of the pan / tilt platform (2b) and a plurality of auxiliary cameras (12) uniformly distributed along the circumference on the main mounting seat (2a). The main camera (11) faces downward, and each auxiliary camera (12) faces downward in an obliquely outward direction.

8. The tower bolt loosening visual detection device according to claim 7, characterized in that: The bottom of the pan-tilt platform (2b) is provided with a plurality of fill lights (13) uniformly distributed along the circumference around the main camera (11).

9. A method for visually detecting loose tower bolts of a wind turbine, characterized in that: Follow these steps: S1. Use a dataset of known bolt images to train a YOLOv11-OBB model using a machine learning algorithm until the YOLOv11-OBB model achieves a set accuracy for bolt annotation using feature-oriented bounding boxes. Then use the YOLOv11-OBB model as the object detection model. S2. Using the visual detection device for loose tower bolts according to any one of claims 1 to 8, a panoramic image of the interior of the wind turbine tower is collected, wherein the panoramic image is a 360° panoramic monitoring image including all tower bolts; S3, performing cropping, scaling, denoising, normalization, and binarization processing on the panoramic image to obtain a preprocessed image; S4. Using the feature oriented bounding box of the target detection model to mark each bolt in the preprocessed image, and extracting the feature oriented bounding box parameters of each bolt, thereby calculating the feature oriented bounding box information of each bolt based on the feature oriented bounding box parameters of each bolt; S5. Determine whether the bolt is loose based on the characteristic oriented bounding box information of the bolt, wherein the characteristic oriented bounding box information of the bolt includes the displacement of the center point of the bolt. and the bolt rotation angle ; To determine whether the bolts are loose, follow these steps: Displacement judgment: Set the displacement threshold of the center point of the bolt ,when When , it is marked as normal displacement; when When , it is marked as fan shaking; Rotation angle judgment: set the rotation angle threshold of the bolt ,when When , it is marked as normal rotation angle; when When , it is marked as potentially loose; Joint decision-making: and , it is judged that the bolt is loose.

10. The method for visually detecting loose tower bolts according to claim 9, wherein: In step S4, the pre-processed image is detected using the target detection model, and the four sides of the rectangular bounding box are made tangent to the edge contour of the target bolt after binary processing, so as to obtain the rectangular bounding box information around the bolt, and obtain the four vertex coordinates of the rectangular bounding box in the order of the image orientation of top, right, bottom, and left ( , )、( , )、( , )、( , ); The current center point coordinates of the bolt are calculated by the coordinates of the four vertices of the rectangular bounding box ( , ), the formula is as follows: ； ； Set the initial center point coordinates when the bolt is installed to ( , ), the center point displacement of the bolt for: ； The calculated coordinates are ( , )’s vertices and coordinates are ( , )'s vertices : ； The calculated coordinates are ( , )’s vertices and coordinates are ( , )'s vertices : ； Set the initial angle when the bolt is installed to , set the current angle of the bolt to , calculate the current angle of the bolt according to the following formula : when ≥ hour: ; when hour: ; The rotation angle of the bolt for: 。

Images