A wind power tower drum bolt robot zero-sample visual positioning method

Abstract

The present application belongs to the technical field of robot vision positioning, and particularly relates to a wind power tower drum bolt robot zero sample vision positioning method. The present application comprises: a laser displacement sensor acquires distance values in real time, calculates the estimated value of the bolt diameter and the position of the bolt top center point; if there is a bolt directly below the laser displacement sensor, the distance between the actuator and the bolt in the Z direction is calculated, and a light spot is formed at the top of the bolt; the actuator is moved close to the vicinity of the bolt top center point; the relative physical distance between the bolt top center point and the actuator in the X direction and the Y direction is calculated, and the actuator is fine-tuned; the actuator is moved by △Z in the Z direction, and the execution action is completed. The present application can complete the vision positioning task under the condition of zero sample, greatly avoids the cost of data collection and labeling, and completes the high-precision bolt vision positioning task with very low hardware cost and less hardware quantity, and the positioning accuracy reaches 0.1 mm.

Description

Technical Field

[0001] This invention belongs to the field of robot vision positioning technology, and in particular relates to a zero-sample vision positioning method for wind turbine tower bolt robots. Background Technology

[0002] Wind turbine towers are a crucial component of wind turbines. The various tower sections are connected by flanges and secured with high-strength bolts. Bolts are extremely important in wind turbine units; a significant portion of serious wind turbine collapses are caused by the failure or breakage of these fastening bolts. Currently, the inspection and tightening of tower bolts are primarily performed by qualified inspectors in high-altitude, confined spaces. This process is highly repetitive, time-consuming, and labor-intensive. Therefore, robots have emerged for application in this field, aiming to reduce the human burden through automation and intelligence. These robots are equipped with vision and perception systems, enabling them to perform tasks such as positioning, inspection, and tightening in complex industrial environments.

[0003] Visual localization is a technique that determines the position of a device, robot, or camera relative to its environmental targets by processing image information; zero-shot visual localization is a method for achieving visual localization tasks without direct samples; large-scale visual models refer to deep learning models pre-trained on large-scale data, which have stronger feature representations and better generalization capabilities. With the emergence of large-scale visual models such as SAM and FastSAM, the performance of zero-shot visual localization tasks has the potential to be significantly improved.

[0004] Patents with publication numbers CN116292141A, CN115615592A, and CN116372555A all mention using visual positioning technology to obtain the bolt position, but do not explain how to achieve visual positioning.

[0005] Patent CN115239669A describes a machine vision perception method for detecting loose bolts on wind turbine towers. This method first uses the CANNY operator to perform edge detection on the entire image, and then performs post-processing calculations using specific parameters to obtain the bolt center point (x, y). The performance of this method largely depends on the choice of parameters, thus it is limited when faced with significant noise (such as oil stains on the bolt surface and changes in lighting) or large variations (such as the bolt diameter and height). Furthermore, this method does not calculate the Z-coordinate of the bolt center point.

[0006] Therefore, the existing patent's technical solution does not involve a visual positioning method for wind turbine tower bolt robots.

[0007] Patent CN115401247A details a visual positioning method for automated drilling in robots. By setting positioning marks at designated drilling locations, it utilizes 3D vision cameras, 2D vision cameras, and distance sensors to achieve high-precision positioning of the drilling plane and position. This method works in conjunction with robots and automated guided vehicles (AGVs) to achieve visual positioning-guided drilling. However, this method requires numerous and expensive laser displacement sensors, especially high-cost 3D vision cameras, and also relies on the assistance of positioning marks.

[0008] Patent CN113232015A describes a robot spatial localization and grasping control method based on template matching, and patent CN111951302A describes a robot visual localization method based on feature matching. Both methods rely on a limited amount of sample data and match predefined templates or features, thus potentially performing poorly in new and complex environments. The introduction of deep learning technology provides an effective way to overcome these shortcomings, improving the robustness and generalization ability of robot visual localization. Patent CN115375759A describes a robot visual localization method and apparatus based on object detection. First, an IR image object detection model is used to detect the support legs of an industrial pallet, enabling the forklift robot to locate the pallet's position. Then, the distance between the pallet and the forklift robot is evaluated by combining a depth image with the pallet's position. The performance and robustness of this method mainly depend on the scale and quality of the pallet sample data.

[0009] When selecting a laser displacement sensor, the accuracy of obtaining Z-coordinate information using only a 2D vision camera and algorithms such as stereo matching or monocular depth estimation is affected by various factors, including camera parameters, scene characteristics, and distance range. Reasonable accuracy is typically achieved within a medium to close distance range. For example, a wind turbine tower bolt robot requires sub-millimeter-level Z-coordinate positioning accuracy for the center point of the bolt top, making the method using only a 2D vision camera infeasible.

[0010] To obtain high-precision three-dimensional spatial relative positions of environmental targets, various laser displacement sensors are often used, especially high-cost 3D vision cameras. However, this choice not only increases costs but also adds to the difficulty of robot structural design. For example, a wind turbine tower bolt robot needs to overcome obstacles and crawl on the narrow flange wall, which places high demands on the robot's miniaturization and thus imposes greater constraints on the selection of laser displacement sensors.

[0011] In visual positioning methods, manually set positioning markers can help robots quickly and accurately locate target positions. However, this not only increases deployment and maintenance costs but also places a high degree of dependence on the visibility and integrity of the markers. For example, marking the center point of the top of a wind turbine tower bolt can be obscured by dust or oil, damaged, or blurred or invisible due to other changes, thus affecting the robot's positioning accuracy.

[0012] Visual localization methods based on template matching or feature matching rely on a limited number of sample data, which limits their adaptability; visual localization methods based on deep learning rely on large-scale, high-quality sample data, which is difficult and costly to collect. Summary of the Invention

[0013] The purpose of this invention is to overcome the shortcomings and deficiencies of the prior art and to propose a zero-sample visual positioning method for wind turbine tower bolt robots.

[0014] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0015] A zero-sample visual localization method for wind turbine tower bolt robots includes:

[0016] The wind turbine tower bolt robot walks along the positive X direction. The laser displacement sensor acquires the distance value in real time. The distance between the laser displacement sensor and the bolt in the Y direction is adjusted. Based on the distance value, the distance Z1 from the laser displacement sensor to the flange surface and the distance Z2 from the laser displacement sensor to the top of the bolt are obtained. The X direction is either clockwise or counterclockwise, and the Y direction is perpendicular to the X direction on the horizontal plane.

[0017] Based on the current distance value Z* and the distance Z2 from the laser displacement sensor to the top of the bolt, determine whether there is a bolt directly below the laser displacement sensor. If not, the wind turbine tower bolt robot continues to move and determine. If it exists, calculate the relative physical distance ΔZ between the actuator and the bolt in the Z direction, and form a light spot on the top of the bolt. The Z direction is perpendicular to the X and Y directions.

[0018] The wind turbine tower bolt robot moves along the direction that makes the actuator move toward the center point of the bolt top, based on the bolt radius R1 and the distance from the laser displacement sensor to the actuator.

[0019] Calculate the relative physical distance (△X, △Y) between the center point of the bolt top and the actuator in the X and Y directions, and fine-tune the actuator according to the relative physical distance (△X, △Y) until the relative physical distance (△X, △Y) is less than the set first threshold.

[0020] The actuator moves a relative physical distance △Z in the Z direction to complete the execution action.

[0021] Further, it also includes performing mean filtering on the distance value.

[0022] Further, adjusting the distance between the laser displacement sensor and the bolt in the Y direction includes:

[0023] Obtaining the distance R from the laser displacement sensor to the central axis of the tower barrel according to the tower barrel radius and the known distance from the laser displacement sensor to the tower barrel wall TD ;

[0024] Calculating the arc length S swept by the laser displacement sensor on the top surface of the bolt according to the encoder value of the wheel motor of the wind power tower barrel bolt robot;

[0025] Calculating the chord length l corresponding to the arc length S, which is expressed by the formula as follows:

[0026] l = 2 * R TD * sin(S / (2 * R TD ))

[0027] Obtaining the distance Y from the chord length to the center point of the bolt top through the Pythagorean theorem D , which is expressed by the formula as follows:

[0028]

[0029] Moving the laser displacement sensor a distance Y in the Y direction towards the bolt D .

[0030] Further, obtaining the distance Z1 from the laser displacement sensor to the flange surface and the distance Z2 from the laser displacement sensor to the bolt top according to the distance value includes:

[0031] Taking any one of the distance values obtained by the laser displacement sensor in real time as the initial distance value denoted as H1. When the difference between the current distance value and H1 is greater than the second threshold, the current distance value is denoted as H2. When the difference between the current distance value and H2 is greater than the second threshold, the current distance value is denoted as H3. If H3 < H1 and H3 < H2, then H1 is the distance Z1 from the laser displacement sensor to the flange surface, and H3 is the distance Z2 from the laser displacement sensor to the bolt top. If the condition H3 < H1 and H3 < H2 is not satisfied, the initial distance value H1 is re-recorded until the distance Z1 from the laser displacement sensor to the flange surface and the distance Z2 from the laser displacement sensor to the bolt top are obtained.

[0032] Further, judging whether there is a bolt directly below the laser displacement sensor according to the current distance value Z* and the distance Z2 from the laser displacement sensor to the bolt top includes:

[0033] When Z*≤Z2, it is determined that there is a bolt directly below the laser displacement sensor; when Z*>Z2, it is determined that there is no bolt directly below the laser displacement sensor.

[0034] Furthermore, the wind turbine tower bolt robot moves along a direction that causes the actuator to move towards the center point of the bolt top, based on the bolt radius R1 and the distance from the laser displacement sensor to the actuator, including:

[0035] The actuator and laser displacement sensor are installed sequentially along the moving direction of the wind turbine tower bolting robot. The wind turbine tower bolting robot moves a distance equal to the bolt radius R1 plus the distance from the laser displacement sensor to the actuator in a direction tending towards the center point of the bolt top.

[0036] Furthermore, the calculation of the relative physical distance (ΔX, ΔY) between the center point of the bolt top and the actuator in the X and Y directions includes:

[0037] Images are acquired using a 2D vision camera and preprocessed to obtain an image of the top surface of the bolt;

[0038] The pixel coordinates (X0, Y0) of the light spot on the top surface of the bolt are calculated using a feature point detection algorithm and used as interactive point cues for the large visual model.

[0039] Mark the actuator as a positioning point on the top surface image of the bolt;

[0040] The segmentation result of the bolt top surface image is obtained by using a large visual model, wherein the segmentation result is the pixel data of the bolt top surface in the bolt top surface image;

[0041] Based on the segmentation results, a post-processing algorithm is used to obtain the relative physical distance (△X, △Y) between the center point of the bolt top and the positioning point.

[0042] Furthermore, the preprocessing includes cropping the image based on the bolt radius R1 and the distance from the 2D vision camera to the top of the bolt, so that the cropped bolt top image contains only one bolt.

[0043] Compared with existing technologies, the significant advantages of this invention are as follows: 1. By introducing a large-scale visual model, it successfully overcomes the limitations of existing visual positioning methods in terms of adaptability and dependence on sample data and manual labeling. It cleverly uses a laser displacement sensor to create light spot coordinates on the top of the bolt, serving as a cue input for the large-scale visual model. This method enables visual positioning tasks to be completed with zero samples, greatly avoiding the costs of data acquisition and labeling. 2. Using a high-precision laser displacement sensor and an inexpensive 2D vision camera, it achieves high-precision bolt visual positioning with very low hardware cost and a small number of hardware components, achieving a positioning accuracy of 0.1mm. Attached Figure Description

[0044] Figure 1 This is a schematic diagram illustrating the application environment of the present invention;

[0045] Figure 2 This is a flowchart of a zero-sample visual localization method for wind turbine tower bolt robots according to the present invention;

[0046] Figure 3 This is a diagram showing the positional relationship between the laser displacement sensor of this invention and the wind turbine tower and bolts;

[0047] Figure 4 This is a schematic diagram illustrating the calculation of the distance the laser displacement sensor of the present invention moves in the Y direction. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0049] This invention provides a zero-sample visual localization method for wind turbine tower bolt robots, which can be applied to applications such as... Figure 1 In the application environment shown, where Figure 1 This is a schematic diagram of the internal environment of a wind turbine tower. An ultrasonic probe is installed on the wind turbine tower bolting robot to detect the preload of the wind turbine bolts. This method can also be applied to the high-precision positioning of blade root bolts, main bearing bolts, and other similar structural bolts in wind power plants. The actuator described in this invention can be an ultrasonic probe or other tools.

[0050] Considering the high-precision positioning requirements of the relative position of the environmental target in three-dimensional space, and the significant constraints on the miniaturization and low cost of laser displacement sensors for the wind turbine tower bolting robot, a high-precision laser displacement sensor and an inexpensive 2D vision camera were selected. The laser displacement sensor and 2D vision camera are located below the wind turbine tower bolting robot. The relative position of the actuator to the 2D vision camera and laser displacement sensor is fixed. The laser displacement sensor is used to obtain the changing distance values ​​between the robot and the plane below during its movement to determine the presence of bolts, calculate the relative distance in the Z direction from the top surface of the bolt to the camera and the actuator, and calculate the spacing between two adjacent bolts. The 2D vision camera is used to acquire images containing the bolts to calculate the relative distances between the center point of the bolt top and the actuator in the X and Y directions.

[0051] Considering the limitations in adaptability of visual localization methods based on edge detection, template matching, or feature matching, and the fact that deep learning-based visual localization methods rely on large-scale, high-quality sample data, leading to significant challenges and high costs in data collection, this paper proposes a solution. By introducing a large-scale visual model, a highly adaptable method can achieve excellent segmentation of the circular top of bolts without fine-tuning model weights using bolt sample data or manually setting localization markers.

[0052] Furthermore, considering that the basic visual model requires input prompts, and that the laser displacement sensor forms a light spot on the top of the bolt when it detects its presence, the pixel coordinates of this light spot can be quickly calculated using a feature point detection algorithm, serving as a point prompt for the basic visual model. It should be noted that the laser displacement sensor can also continuously emit a beam of light; when the laser displacement sensor moves above the bolt, a light spot is naturally formed.

[0053] like Figure 2 As shown, a zero-sample visual localization method for wind turbine tower bolt robots includes the following steps:

[0054] Step 1: Obtain and calculate the position parameters of the laser displacement sensor and the geometric parameters of the bolt.

[0055] 1.1 According to the initial parameters, the wind turbine tower bolt robot walks smoothly along the positive X direction of the tower flange, reads the distance value obtained by the laser displacement sensor in real time, and performs data preprocessing on the distance value, wherein the data preprocessing is mean filtering.

[0056] Then adjust the position of the laser displacement sensor relative to the bolt in the Y direction, specifically, as follows: Figure 3 As shown, the uppermost arc represents the tower wall, the circles below the arc represent bolts, and the bottommost point represents the tower's central axis. The distance R from the laser displacement sensor to the tower's central axis is obtained based on the tower radius and the known distance from the laser displacement sensor to the tower wall. TD The distance from the laser displacement sensor to the tower wall is a fixed value, initially determined by the specifications of the wind turbine tower bolts to ensure the laser sensor can scan the bolt surface. The arc length S swept by the laser displacement sensor on the top surface of the bolt is calculated based on the encoder value of the wheel motor of the wind turbine tower bolt robot. Arc length S = (actually measured encoder value / encoder value per motor revolution) * 2 * 3.14 * tower radius.

[0057] The chord length l corresponding to arc length S is calculated using the following formula:

[0058] l = 2 * R TD *sin(S / (2*R TD ))

[0059] As shown Figure 4 in the figure, the distance Y from the chord length l to the center point of the bolt top is obtained by the Pythagorean theorem D , which is expressed by the formula as follows:

[0060]

[0061] The laser displacement sensor moves a distance Y in the Y direction D . Determine the moving direction of the laser displacement sensor according to the bolt specification. If the bolt specification exceeds M64, that is, the bolt diameter is greater than or equal to 64 mm, the laser displacement sensor moves in the positive direction of the Y direction. If the bolt specification is less than M64, that is, the bolt diameter is less than 64 mm, the laser displacement sensor moves in the negative direction of the Y direction. The X direction is the clockwise direction or the counterclockwise direction, and the Y direction is perpendicular to the X direction on the horizontal plane

[0062] 1.2. Obtain the distance Z1 from the laser displacement sensor to the flange surface and the distance Z2 from the laser displacement sensor to the bolt top according to the distance value

[0063] Specifically, take any one of the distance values obtained by the laser displacement sensor in real time as the initial distance value and denote it as H1. When the difference between the current distance value and H1 is greater than the second threshold, denote the current distance value as H2. When the difference between the current distance value and H2 is greater than the second threshold, denote the current distance value as H3. If H3 < H1 and H3 < H2, then H1 is the distance Z1 from the laser displacement sensor to the flange surface, and H3 is the distance Z2 from the laser displacement sensor to the bolt top. If the condition H3 < H1 and H3 < H2 is not satisfied, record the initial distance value H1 again until the distance Z1 from the laser displacement sensor to the flange surface and the distance Z2 from the laser displacement sensor to the bolt top are obtained

[0064] Step 2: Calculate the high-precision relative physical distance between the center point of the target bolt top and the actuator in the Z direction, and the rough relative physical distances in the X and Y directions, so as to complete the preliminary alignment of the actuator and the target point

[0065] 2.1. During the robot walking process, compare the current distance value Z* obtained by the laser displacement sensor with the distance Z2 from the laser displacement sensor to the bolt top to judge whether there is a bolt directly below the laser displacement sensor. When Z* ≤ Z2, it is judged that there is a bolt directly below the laser displacement sensor. When Z* > Z2, it is judged that there is no bolt directly below the laser displacement sensor

[0066] 2.2. If there is no bolt, continue with step 2.1

[0067] 2.3 If a bolt exists, calculate the relative physical distance ΔZ between the actuator and the bolt in the Z direction, and form a light spot on the top of the bolt. The Z direction is perpendicular to the X and Y directions.

[0068] The relative positions of the actuator, the 2D vision camera, and the laser displacement sensor are fixed, so the relative distance is known. The relative physical distance ΔZ between the actuator and the bolt in the Z direction is calculated based on the relative distance between the actuator and the laser displacement sensor and the distance Z2 from the laser displacement sensor to the top of the bolt.

[0069] 2.4 Simultaneously, the wind turbine tower bolt robot moves according to the bolt radius R1 and the distance from the laser displacement sensor to the actuator, bringing the actuator closer to the center point of the bolt top. The bolt radius R1 is obtained by scanning the QR code on the bolt.

[0070] Specific movement method: Under normal circumstances, the actuator and the laser displacement sensor are installed sequentially along the movement direction of the wind turbine tower bolt robot. The wind turbine tower bolt robot moves a distance equal to the bolt radius R1 plus the distance from the laser displacement sensor to the actuator in the direction of the center point of the bolt top.

[0071] When the actuator and laser displacement sensor are installed sequentially in the opposite direction to the movement direction of the wind turbine tower bolting robot, the robot tends to move towards the center point of the bolt's top, which is the difference between the bolt radius R1 and the distance from the laser displacement sensor to the actuator. In this case, the laser displacement sensor needs to ignore bolts that have already been detected when detecting bolts.

[0072] Step 3: Calculate the high-precision relative physical distance between the center point of the top of the target bolt in the X and Y directions, and fine-tune the actuator accordingly. Repeat this step until the relative distance is less than the set first threshold.

[0073] 3.1. Open the 2D vision camera, acquire an image and preprocess the image to obtain the top surface image of the bolt.

[0074] The preprocessing involves cropping the complete image based on the bolt radius R1 and the distance Z3 from the 2D vision camera to the top of the bolt, ensuring that the cropped image of the bolt top surface contains only one bolt. The distance Z3 from the 2D vision camera to the top of the bolt is calculated based on the distance Z2 from the laser displacement sensor to the top of the bolt and the relative distance between the 2D vision camera and the laser displacement sensor.

[0075] 3.2. The pixel coordinates (X0, Y0) of the light spot on the top surface of the bolt are calculated using a feature point detection algorithm and used as an interactive point cue for the large-scale visual model (e.g., SAM, EfficientSAM, MobileSAM, FastSAM). The actuator is then marked as a positioning point on the top surface of the bolt.

[0076] 3.3. The segmentation result of the bolt top surface image is obtained by calculating the large-scale visual model, where the segmentation result is the pixel data of the bolt top surface in the bolt top surface image. The relative physical distance (△X, △Y) between the center point of the bolt top and the positioning point of the actuator is obtained by using a post-processing algorithm.

[0077] The post-processing algorithm can be implemented as follows: First, a contour lookup algorithm is used to obtain the contour information of the top of the bolt, and a contour image is obtained based on the contour information. Then, the central moment of the contour image is calculated, and the centroid of the contour image, i.e., the pixel coordinates of the center point of the top of the bolt, is obtained through the central moment. Finally, the pixel distance between this point and the corresponding marker point of the actuator is calculated according to Euler's formula, and the relative physical distance (△X, △Y) between the center point of the top of the bolt and the positioning point of the actuator is obtained according to the formula: pixel distance * distance in the Z direction from the camera lens to the top of the bolt * pixel size / lens focal length.

[0078] 3.4 If (△X, △Y) is greater than or equal to the set first threshold, the actuator fine-tunes △X and △Y, and then executes step 3.3 again.

[0079] 3.5 If (△X, △Y) is less than the set first threshold, it means that the alignment between the actuator and the target point is completed. The actuator moves △Z in the Z direction to complete the execution action; the robot continues to move to work on the next bolt.

[0080] The embodiments described above are merely illustrative of one or more implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the appended claims.

Claims

1. A zero-sample visual localization method for wind turbine tower bolt robots, characterized in that, The zero-sample visual localization method for wind turbine tower bolt robots includes: The wind turbine tower bolt robot walks along the positive X direction. The laser displacement sensor acquires the distance value in real time. The distance between the laser displacement sensor and the bolt in the Y direction is adjusted. Based on the distance value, the distance Z1 from the laser displacement sensor to the flange surface and the distance Z2 from the laser displacement sensor to the top of the bolt are obtained. The X direction is either clockwise or counterclockwise, and the Y direction is perpendicular to the X direction on the horizontal plane. Based on the current distance value Z* obtained by the laser displacement sensor and the distance Z2 from the laser displacement sensor to the top of the bolt, it is determined whether there is a bolt directly below the laser displacement sensor. If there is no bolt, the wind turbine tower bolt robot continues to walk and make judgments. If there is a bolt, the relative physical distance ΔZ between the actuator and the bolt in the Z direction is calculated, and a light spot is formed on the top of the bolt. The Z direction is perpendicular to the X and Y directions. The wind turbine tower bolt robot moves along the direction that makes the actuator move toward the center point of the bolt top, based on the bolt radius R1 and the distance from the laser displacement sensor to the actuator. Calculate the relative physical distance (△X, △Y) between the center point of the bolt top and the actuator in the X and Y directions, and fine-tune the actuator according to the relative physical distance (△X, △Y) until the relative physical distance (△X, △Y) is less than the set first threshold. The actuator moves a relative physical distance △Z in the Z direction to complete the execution action; in, The calculation of the relative physical distance (ΔX, ΔY) between the center point of the bolt top and the actuator in the X and Y directions includes: Images are acquired using a 2D vision camera and preprocessed to obtain an image of the top surface of the bolt; The pixel coordinates (X0, Y0) of the light spot on the top surface of the bolt are calculated using a feature point detection algorithm and used as interactive point cues for the basic visual model. Mark the actuator as a positioning point on the top surface image of the bolt; The segmentation result of the bolt top surface image is obtained by using a large visual model, wherein the segmentation result is the pixel data of the bolt top surface in the bolt top surface image; Based on the segmentation results, a post-processing algorithm is used to obtain the relative physical distance (△X, △Y) between the center point of the bolt top and the positioning point.

2. The zero-sample visual positioning method for wind turbine tower bolt robots according to claim 1, characterized in that, It also includes mean filtering of the distance values.

3. The zero-sample visual localization method for wind turbine tower bolt robots according to claim 1, characterized in that, The adjustment of the distance between the laser displacement sensor and the bolt in the Y direction includes: The distance R from the laser displacement sensor to the central axis of the tower is obtained based on the tower radius and the known distance from the laser displacement sensor to the tower wall. TD ; The arc length S swept by the laser displacement sensor on the top surface of the bolt is calculated based on the encoder value of the wheel motor of the wind turbine tower bolt robot. Calculate the chord length corresponding to arc length S. l The formula is as follows: l =2* R TD * sin( S / (2*R TD )); The distance Y from the chord length to the center point of the bolt top can be obtained using the Pythagorean theorem. D The formula is as follows: ； The laser displacement sensor moves a distance Y in the Y direction toward the bolt. D .

4. The zero-sample visual localization method for wind turbine tower bolt robots according to claim 1, characterized in that, The method of obtaining the distance Z1 from the laser displacement sensor to the flange surface and the distance Z2 from the laser displacement sensor to the top of the bolt based on the distance value includes: Take any one of the distance values obtained by the laser displacement sensor in real time as the initial distance value, denoted as H1. When the difference between the current distance value obtained by the laser displacement sensor and H1 is greater than the second threshold, record the current distance value obtained by the laser displacement sensor as H2. When the difference between the current distance value obtained by the laser displacement sensor and H2 is greater than the second threshold, record the current distance value obtained by the laser displacement sensor as H3. If H3 < H1 and H3 < H2, then H1 is the distance Z1 from the laser displacement sensor to the flange surface, and H3 is the distance Z2 from the laser displacement sensor to the top of the bolt. If the condition H3 < H1 and H3 < H2 is not satisfied, re-record the initial distance value H1 until the distance Z1 from the laser displacement sensor to the flange surface and the distance Z2 from the laser displacement sensor to the top of the bolt are obtained.

5. The zero-sample visual localization method for wind turbine tower bolt robots according to claim 1, characterized in that, Judging whether there is a bolt directly below the laser displacement sensor according to the current distance value Z* obtained by the laser displacement sensor and the distance Z2 from the laser displacement sensor to the top of the bolt includes: When Z* ≤ Z2, it is judged that there is a bolt directly below the laser displacement sensor. When Z* > Z2, it is judged that there is no bolt directly below the laser displacement sensor.

6. The zero-sample visual localization method for wind turbine tower bolt robots according to claim 1, characterized in that, The wind power tower bolt robot moves along the direction that makes the actuator tend to the center point of the top of the bolt according to the bolt radius R1 and the distance from the laser displacement sensor to the actuator, including: The actuator and the laser displacement sensor are installed in sequence along the moving direction of the wind power tower bolt robot. The wind power tower bolt robot moves a distance equal to the bolt radius R1 plus the distance from the laser displacement sensor to the actuator along the direction tending to the center point of the top of the bolt.

7. The zero-sample visual localization method for wind turbine tower bolt robots according to claim 1, characterized in that, The preprocessing includes cropping the image according to the bolt radius R1 and the distance from the 2D vision camera to the top of the bolt, so that the cropped bolt top surface image only contains one bolt.

Images