A wind power tower cylinder wall-climbing robot weld seam tracking control system based on visual intelligent identification

Abstract

The application discloses a wind power tower cylinder wall-climbing robot welding seam tracking control system based on visual intelligent identification, relates to the technical field of wall-climbing robot control, and comprises the following steps: collecting tower cylinder outer wall images, driving wheel encoder rotation increments and TOFD scanning data, setting a welding seam identification area and four brightness reference areas, identifying blade shadow events and their moving directions according to grayscale changes, and calculating the time when the blade shadow reaches the first side edge and the second side edge of the welding seam; generating a double-edge welding seam center line or a single-edge compensation center line according to the longitudinal projection length of the welding seam edge, and calculating the target rotation amount of the four driving wheels according to the center line; and verifying the tracking result by combining the welding seam position and the TOFD scanning data after the event ends, so that the influence of the blade shadow on welding seam visual identification and robot tracking control is reduced.

Description

Technical Field

[0001] This invention relates to the field of wall-climbing robot control technology, specifically a weld seam tracking control system for a wind turbine tower wall-climbing robot based on visual intelligent recognition. Background Technology

[0002] When wall-climbing robots perform visual tracking, they typically extract the weld location based on brightness differences, texture features, or edge features in the image. However, during the operation of wind turbines, the rotating blades create constantly moving shadows on the tower surface, causing changes in the local brightness and contrast of the weld image. When the shadow passes over the weld, some weld edges may become blurred, broken, or temporarily unrecognizable, causing the weld position obtained by the robot to deviate. Existing visual tracking methods usually adjust the robot's movement based on the recognition results of the current image, making it difficult to adjust the tracking method before the shadow reaches the weld. Furthermore, tracking interruptions or incorrect turns are prone to occur when the weld edge is partially missing.

[0003] Chinese invention patent application CN120839769A, published on October 28, 2025, discloses a method, device, and robot for tracking weld seams in a wall-climbing robot. The method includes: acquiring weld seam images via a camera; inputting the weld seam images into a rotating target detection model to obtain the corner coordinates of the weld seam's rotating bounding box; calculating the robot's offset and yaw angle relative to the weld seam based on the corner coordinates; and then outputting the differential speed adjustment amount through a differential speed adjustment prediction model to control the speed of the left and right wheels and enable the robot to track the weld seam. However, during wind turbine operation, the rotating blades create continuously moving local shadows on the outer wall of the tower, making it difficult to maintain complete shielding from external light in the actual working environment. When the moving shadow passes through the weld seam area, the local brightness and edge integrity of the weld seam image change. Calculating the robot's offset and yaw angle solely based on the current weld seam image may still result in weld seam positioning fluctuations or tracking interruptions.

[0004] Therefore, the present invention provides a weld seam tracking control system for a wind turbine tower climbing robot based on visual intelligent recognition. Summary of the Invention

[0005] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a wind turbine tower climbing robot weld seam tracking control system based on visual intelligent recognition. By setting a weld seam recognition area and four brightness reference areas, it identifies blade shadow events and their movement directions based on grayscale changes, calculates the time when the blade shadow reaches the first and second side edges of the weld seam, and switches between the double-edge weld seam centerline and the single-edge compensation centerline based on the longitudinal projection length of the weld seam edge. Combined with the encoder rotation increment, it generates the target rotation amount of the four drive wheels. Finally, it verifies the tracking results using the weld seam position after the event ends and TOFD scan data, thus achieving continuous positioning and stable tracking of wind turbine tower weld seams under blade shadow interference conditions.

[0006] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition, comprising: S1. Acquire images of the outer wall of the tower, encoder rotation increments, and TOFD scanning data to obtain the grayscale median range, reference weld edge, reference weld centerline, and reference weld width. S2. Identify leaf glimpse events based on grayscale changes in four brightness reference areas; S3. Calculate the speed of the blade's sweeping leading edge, the arrival time of the first side edge, and the arrival time of the second side edge; S4. Obtain the weld positioning result based on the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the lengths of the paired edges on both sides; S5. Generate the target rotation amount of the four drive wheels based on the arrival time of the first side edge, the arrival time of the second side edge, and the weld positioning result; S6. Verify the weld tracking results based on the weld location and TOFD scan data after the event ends.

[0007] Preferably, images of the outer wall of the tower are continuously acquired and the acquisition time is recorded; the cumulative rotation angles recorded by the encoders of the four drive wheels at the current acquisition time and the previous acquisition time are obtained, and the cumulative rotation angle of each drive wheel at the current acquisition time is subtracted from the cumulative rotation angle of the previous acquisition time to obtain the encoder rotation increments of the four drive wheels between the two acquisition times; the TOFD scan data formed between the previous acquisition time and the current acquisition time are acquired, and the current image, acquisition time, encoder rotation increments and TOFD scan data are assigned the same acquisition number; The robot's forward direction is consistent with the longitudinal direction of the tower's outer wall image; taking the upper left corner of the current tower's outer wall image as the origin, the direction from left to right is recorded as the horizontal direction of the image, and the direction from top to bottom is recorded as the vertical direction of the image; taking the robot's forward direction as the reference, the two drive wheels located on the left side of the robot are recorded as the left drive wheels, and the two drive wheels located on the right side of the robot are recorded as the right drive wheels; the distance between the line connecting the centers of the two left drive wheels and the line connecting the centers of the two right drive wheels is recorded as the center-to-center distance between the left and right drive wheels; The edges of the weld seam are extracted using OpenCV image processing tools. The weld seam edge on the left side of the current tower outer wall image is designated as the first side edge, and the weld seam edge on the right side of the current tower outer wall image is designated as the second side edge. The midpoints of the two edges at the same vertical coordinate are connected to obtain the weld seam centerline, and the median of the horizontal distance between the two edges is designated as the weld seam width. The first side edge of the weld seam is extended to the left of the current tower outer wall image by one weld seam width, and the second side edge of the weld seam is extended to the right of the current tower outer wall image by one weld seam width to obtain the weld seam recognition area. The first brightness reference area, the second brightness reference area, the weld seam recognition area, the third brightness reference area, and the fourth brightness reference area are set sequentially from left to right of the current tower outer wall image. In N consecutive images of the tower outer wall, the region that moves laterally and has a gray level lower than the surrounding tower surface is denoted as the blade sweeping region. When no blade sweeping region appears in the image, N consecutive images of the tower outer wall are selected as reference images, where N is a positive integer not less than 2. The reference images are converted into grayscale images using OpenCV, and the grayscale median of each brightness reference area in each reference image is calculated. The minimum and maximum values ​​among the N grayscale medians are denoted as the lower limit and upper limit of the grayscale median, respectively, and the range between them is denoted as the grayscale median range. The N reference images are registered to the same coordinate system, and the median of the lateral coordinates of the first and second side edges of the weld at each longitudinal position is taken to obtain the first and second side edges of the reference weld. The midpoint between the two reference weld edges is used to obtain the centerline of the reference weld, and the median of the lateral distance between the two reference weld edges is denoted as the width of the reference weld.

[0008] Preferably, the current tower outer wall image is converted into a grayscale image using OpenCV, and the current grayscale median of the four brightness reference areas is calculated. The acquisition time corresponding to the first time the grayscale median of the first brightness reference area, the second brightness reference area, the third brightness reference area, and the fourth brightness reference area is recorded respectively as the first darkening time, the second darkening time, the third darkening time, and the fourth darkening time. As the first darkening moment, the second darkening moment, the third darkening moment, and the fourth darkening moment increase sequentially, the current grayscale change is recorded as a blade sweeping event moving from the left to the right of the current tower outer wall image; as the fourth darkening moment, the third darkening moment, the second darkening moment, and the first darkening moment increase sequentially, the current grayscale change is recorded as a blade sweeping event moving from the right to the left of the current tower outer wall image. When all four brightness reference areas are below their respective grayscale median lower limit for the first time in the tower outer wall image corresponding to the same acquisition time, this brightness change is recorded as the overall illumination change; the earliest darkening moment in the blade shadow event is recorded as the event start time; when the four current grayscale medians are all within their respective grayscale median range in the images corresponding to N consecutive acquisition numbers, the last acquisition time is recorded as the event end time.

[0009] Preferably, in the current tower outer wall image, the average of the lateral coordinates of the left and right boundaries of each brightness reference area is taken as the lateral coordinate of the center of the brightness reference area; the average of the longitudinal coordinates of the upper and lower boundaries of the weld identification area is taken as the longitudinal coordinate of the detection line, and a straight line parallel to the lateral direction of the current tower outer wall image is drawn at this longitudinal coordinate, and this straight line is recorded as the detection line; the lateral coordinates of the intersection points of the detection line with the first side edge and the second side edge of the reference weld are read respectively. For a blade sweeping event moving from left to right in the current tower outer wall image, the lateral distance between the centers of the first and second brightness reference areas is divided by the difference between the second darkening moment and the first darkening moment to obtain the blade sweeping edge movement speed. At the second darkening moment, the blade sweeping edge is located at the center lateral coordinate of the second brightness reference area. The first lateral distance is obtained by subtracting the center lateral coordinate of the second brightness reference area from the lateral coordinate of the intersection of the detection line and the first side edge of the reference weld. The second lateral distance is obtained by subtracting the center lateral coordinate of the second brightness reference area from the lateral coordinate of the intersection of the detection line and the second side edge of the reference weld. The first and second lateral distances are divided by the blade sweeping edge movement speed and added to the second darkening moment to obtain the arrival time of the first and second side edges. For a blade sweeping shadow event moving from right to left in the current tower outer wall image, the lateral distance between the centers of the fourth and third brightness reference areas is divided by the difference between the third and fourth darkening times to obtain the blade sweeping shadow leading edge speed. At the third darkening time, the blade sweeping shadow leading edge is located at the center lateral coordinate of the third brightness reference area. The second side lateral distance is obtained by subtracting the lateral coordinate of the intersection of the detection line and the second side edge of the reference weld from the center lateral coordinate of the third brightness reference area. The first side lateral distance is obtained by subtracting the lateral coordinate of the intersection of the detection line and the first side edge of the reference weld from the center lateral coordinate of the third brightness reference area. The second side lateral distance and the first side lateral distance are divided by the blade sweeping shadow leading edge speed, and then added to the third darkening time to obtain the arrival time of the second side edge and the arrival time of the first side edge. When the darkening times of the two brightness reference areas used to calculate the blade sweeping shadow leading edge speed are the same, the blade sweeping shadow leading edge speed, the arrival time of the first side edge, and the arrival time of the second side edge are not calculated.

[0010] Preferably, the edge map of the weld recognition area is generated by the Canny function of OpenCV, and the curve composed of interconnected edge pixels is extracted by the findContours function, and each curve is recorded as a continuous edge contour; on both sides of the reference weld center line, the continuous edge contour located on the left side of the reference weld center line is recorded as the first side candidate contour, and the continuous edge contour located on the right side of the reference weld center line is recorded as the second side candidate contour. For each candidate profile of the first side, at each longitudinal coordinate covered by the candidate profile of the first side and the first edge of the reference weld, calculate the absolute value of the difference between the transverse coordinate of the candidate profile of the first side and the transverse coordinate of the first edge of the reference weld, and record the average value of each absolute value as the average transverse distance of the first side. Select the candidate profile of the first side with the smallest average transverse distance of the first side as the first edge of the current weld. For each candidate profile of the second side, at each longitudinal coordinate covered by the candidate profile of the second side and the second side edge of the reference weld, calculate the absolute value of the difference between the transverse coordinate of the candidate profile of the second side and the transverse coordinate of the second side edge of the reference weld, and record the average value of each absolute value as the average transverse distance of the second side. Select the candidate profile of the second side with the smallest average transverse distance of the second side as the current second side edge of the weld. The `boundingRect` function is called to obtain the bounding rectangles of the first and second side edges of the current weld. The lengths of the two bounding rectangles along the longitudinal direction of the image are recorded as the longitudinal projection lengths of the first and second side edges, respectively. The longitudinal coordinate ranges of the first and second side edges of the current weld are obtained. The length of the overlapping part of the two longitudinal coordinate ranges is recorded as the length of the paired edge. Half of the longitudinal length of the weld recognition area is recorded as the edge usage length. When the longitudinal projection length of the first side edge is not less than the edge usage length, the lateral coordinates of each point on the first side edge of the current weld are increased by half the width of the reference weld to obtain the first side single edge compensation center line; when the longitudinal projection length of the second side edge is not less than the edge usage length, the lateral coordinates of each point on the second side edge of the current weld are decreased by half the width of the reference weld to obtain the second side single edge compensation center line; when the longitudinal projection lengths of the first side edge, the second side edge, and the paired edge lengths are all not less than the edge usage length, the midpoints of the first side edge and the second side edge of the current weld at the same longitudinal coordinate are connected to obtain the double edge weld center line; when the longitudinal projection lengths of the first side edge and the second side edge are both less than the edge usage length, it is determined that the current tower outer wall image cannot provide weld positioning results, and the robot forward movement is stopped. The first side single edge compensation center line, the second side single edge compensation center line, and the double edge weld center line are collectively recorded as the weld positioning result.

[0011] Preferably, for a blade sweeping event moving from the left to the right of the current tower outer wall image, when the difference between the arrival time of the first side edge and the current acquisition time is less than or equal to the difference between the current acquisition time and the previous acquisition time, the second side single edge compensation center line is used as the weld seam tracking target; when the second side single edge compensation center line is not obtained, the robot's forward movement is stopped. For a blade sweeping event moving from the right to the left in the current tower outer wall image, when the difference between the arrival time of the second side edge and the current acquisition time is less than or equal to the difference between the current acquisition time and the previous acquisition time, the first side single edge compensation center line is used as the weld seam tracking target; if the first side single edge compensation center line is not obtained, the robot's forward movement is stopped. If neither the arrival time of the first side edge nor the arrival time of the second side edge triggers the selection of the single-edge compensation centerline, and the centerline of the double-edge weld has already been obtained, the double-edge weld centerline is used; if only the centerline of the first side single-edge compensation is obtained, the first side single-edge compensation centerline is used; if only the centerline of the second side single-edge compensation is obtained, the second side single-edge compensation centerline is used; if both the first and second side single-edge compensation centerlines are obtained, and the longitudinal projection length of the first side edge is greater than or equal to the longitudinal projection length of the second side edge, the first side single-edge compensation centerline is used; if the longitudinal projection length of the first side edge is less than the longitudinal projection length of the second side edge, the second side single-edge compensation centerline is used; if neither the centerline of the double-edge weld nor any single-edge compensation centerline is obtained, no weld tracking target is generated, the target rotation of all four drive wheels is set to zero, and the robot's forward movement is stopped; when a weld tracking target is generated, subsequent calculations of lateral tracking deviation, directional tracking deviation, and target rotation are performed. The OpenCV fitLine function is called to fit the center line point of the weld tracking target located in the lower third of the weld recognition area. The horizontal coordinate of the intersection point of the fitted line and the lower boundary of the weld recognition area is subtracted from the horizontal center coordinate of the image to obtain the horizontal tracking deviation. The directional tracking deviation is obtained according to the angle between the fitted line and the vertical direction of the image. To obtain the robot's forward velocity, divide the lateral tracking deviation by the lateral width of the weld recognition area to obtain the lateral deviation ratio; divide the directional tracking deviation by 90 degrees to obtain the directional deviation ratio; multiply the sum of the lateral deviation ratio and the directional deviation ratio by the ratio of the forward velocity to the center distance between the left and right drive wheels to obtain the target steering angular velocity; the positive and negative signs of the target steering angular velocity indicate that the robot is turning to the right and left, respectively, and the absolute value of the target steering angular velocity indicates the magnitude of the robot's steering speed. To obtain the radius of the drive wheels, multiply the rotation increments of the four encoders corresponding to the most recent acquisition number before the start of the event by the radius of the drive wheels to get the distance traveled by the four drive wheels. The forward velocity of the event is obtained by dividing the average distance traveled by the four drive wheels by the difference between the acquisition time corresponding to the current acquisition number and the acquisition time corresponding to the previous acquisition number. When using a double-edge weld centerline, the forward velocity of the event is taken as the current forward velocity; when using a first-side single-edge compensation centerline, the ratio of the longitudinal projection length of the first-side edge to the longitudinal length of the weld identification area is multiplied by the forward velocity of the event to obtain the current forward velocity; when using a second-side single-edge compensation centerline, the ratio of the longitudinal projection length of the second-side edge to the longitudinal length of the weld identification area is multiplied by the forward velocity of the event to obtain the current forward velocity. Let the linear velocities of the left and right targets be denoted as follows: and Calculate the linear velocities of the target on the left and the target on the right; respectively and Divide by the radius of the drive wheel, then multiply by the difference between the current acquisition time and the previous acquisition time to obtain the target rotation amount of the left drive wheel and the target rotation amount of the right drive wheel, and send them to the corresponding drive wheels.

[0012] Preferably, after the blade sweeping event is determined in S2, the edges on both sides of the weld are re-extracted using OpenCV; when the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the length of the paired edges on both sides are not less than the edge usage length, the midpoints of the first side edge and the second side edge of the current weld at the same longitudinal coordinate are connected to obtain the center line of the double-edge weld after the event ends. The fitLine function is called to fit the centerline point of the double-edge weld centerline located in the lower third of the weld recognition area after the event ends. The absolute value of the difference between the horizontal coordinate of the intersection of the fitted line and the lower boundary of the weld recognition area and the horizontal center coordinate of the image is recorded as the horizontal deviation at the end of the event; the absolute value of the horizontal tracking deviation corresponding to the most recent acquisition number before the start of the event is recorded as the horizontal deviation at the start of the event. Using TomoView TOFD analysis software, an A-scan image is generated based on the TOFD scan data corresponding to each acquisition number. The A-scan image is a waveform diagram with ultrasonic wave propagation time as the horizontal axis and ultrasonic wave received signal amplitude as the vertical axis. It is determined whether a direct wave and a bottom echo exist simultaneously in each A-scan image, and the acquisition number with both direct wave and bottom echo is recorded as a valid TOFD acquisition number. The number of valid TOFD acquisition numbers in the N most recent acquisition numbers before the start of the event is divided by N to obtain the TOFD continuity ratio before the event. The number of valid TOFD acquisition numbers between the start and end of the event is divided by the total number of acquisition numbers in that period to obtain the TOFD continuity ratio of the event. When the lateral deviation at the end of the event is not greater than the lateral deviation at the beginning of the event, and the continuous TOFD ratio of the event is not lower than the continuous TOFD ratio before the event, the weld tracking result during this blade glide event is deemed valid, and weld tracking continues. When the lateral deviation at the end of the event is greater than the lateral deviation at the beginning of the event, or when the continuous TOFD ratio of the event is lower than the continuous TOFD ratio before the event, the weld tracking result during this blade sweep event is deemed invalid, and the robot's forward movement is stopped. After stopping, the edges on both sides of the weld are re-extracted and the TOFD scan data is checked. If the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the length of the paired edges on both sides corresponding to N consecutive acquisition numbers are not less than the edge usage length, and the N consecutive acquisition numbers are all valid TOFD acquisition numbers, weld tracking is resumed; otherwise, the robot remains in the adsorption state and a manual review prompt is output.

[0013] A wind turbine tower climbing robot weld seam tracking control system based on visual intelligent recognition includes a data acquisition and reference acquisition module, a blade shadow event recognition module, an arrival time calculation module, a weld seam positioning module, a drive control module, and a tracking result verification module. The data acquisition and benchmark acquisition module is used to acquire images of the outer wall of the tower, the acquisition time, the encoder rotation increment of the four drive wheels and TOFD scan data and assign them the same acquisition number. It sets four brightness reference areas and weld recognition areas. Based on the benchmark image where no blade grazing area appears, it acquires the grayscale median range of each brightness reference area, the first side edge of the benchmark weld, the second side edge of the benchmark weld, the center line of the benchmark weld and the width of the benchmark weld. The leaf shadow event recognition module is used to identify the movement direction of the leaf shadow event based on the order in which the median gray value of the four brightness reference areas first falls below the lower limit of their respective median gray values, and to record the start and end times of the event. The arrival time calculation module is used to calculate the moving speed of the blade's shadow leading edge based on the difference between the lateral distance between the centers of the brightness reference areas and the darkening time, and to calculate the arrival time of the first side edge and the arrival time of the second side edge. The weld positioning module is used to extract the first side candidate contour and the second side candidate contour, obtain the first side edge and the second side edge of the current weld based on the average lateral distance between the candidate contour and the reference weld edge, and generate the first side single edge compensation center line, the second side single edge compensation center line or the double edge weld center line based on the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge and the length of the paired edges on both sides. The drive control module is used to select the weld tracking target based on the moving direction of the blade sweeping event, the arrival time of the first side edge, the arrival time of the second side edge, and the weld positioning result; calculate the lateral tracking deviation and directional tracking deviation based on the weld tracking target; calculate the forward speed of the event and the current forward speed based on the encoder rotation increment; and generate the target rotation amount of the four drive wheels based on the forward speed and the target steering angular velocity. The tracking result verification module is used to calculate the lateral deviation at the end of the event, the lateral deviation at the beginning of the event, the continuous ratio of TOFD before the event, and the continuous ratio of TOFD during the event after the blade sweep event ends. Based on the deviation comparison results and the TOFD continuous ratio comparison results, it determines whether the weld tracking result is valid, and stops the robot's forward movement when the weld tracking result is invalid.

[0014] (III) Beneficial Effects This invention provides a weld seam tracking control system for a wind turbine tower climbing robot based on visual intelligent recognition, which has the following advantages: 1. By identifying the blade shadow event and its movement direction through the grayscale change sequence of four brightness reference areas, and further calculating the time when the blade shadow reaches the first and second side edges of the weld, the robot can switch the weld tracking target before the weld edge is about to be occluded, reducing the impact of blade shadow on the timeliness of visual recognition.

[0015] 2. Based on the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the length of the paired edges on both sides, generate the center line of the double-edge weld, the center line of the first side single edge compensation, or the center line of the second side single edge compensation. This enables the system to obtain usable weld positioning results even when the weld edge is partially missing or partially obscured, thus improving the continuity of weld identification under complex lighting conditions.

[0016] 3. Calculate the lateral tracking deviation and directional tracking deviation based on the weld seam tracking target, and generate the target rotation amount of the four drive wheels by combining the current forward speed and the target turning angular velocity. This enables the robot to adjust the rotation state of the left and right drive wheels in a timely manner, reduce the robot's lateral and directional offset relative to the weld seam, and improve the weld seam tracking stability during the wall climbing process.

[0017] 4. Verify the weld seam tracking results based on the lateral deviation at the end of the event, the lateral deviation at the beginning of the event, the continuous TOFD ratio before the event, and the continuous TOFD ratio during the event. When the tracking results are invalid, stop the robot's forward movement and output a manual review prompt. This can promptly detect abnormal tracking states and reduce the impact of continued operation after deviating from the weld seam on the reliability of the detection results. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the process of a wind turbine tower climbing robot weld seam tracking control method based on visual intelligent recognition according to the present invention. Figure 2 This is a schematic diagram of the weld seam tracking control system for a wind turbine tower climbing robot based on visual intelligent recognition, according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 This invention provides a method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition, comprising the following steps: S1. Acquire images of the outer wall of the tower, encoder rotation increments, and TOFD scan data to obtain the grayscale median range, reference weld edge, reference weld centerline, and reference weld width.

[0021] Continuously acquire images of the tower's outer wall and record the acquisition time; obtain the cumulative rotation angles recorded by the encoders of the four drive wheels at the current acquisition time and the previous acquisition time; subtract the cumulative rotation angle of the drive wheel at the previous acquisition time from the cumulative rotation angle of the drive wheel at the current acquisition time to obtain the encoder rotation increments of the four drive wheels between the two acquisition times, with both the cumulative rotation angle and the encoder rotation increments expressed in radians; acquire the TOFD scan data formed between the previous acquisition time and the current acquisition time, and assign the same acquisition number to the current image, acquisition time, encoder rotation increment, and TOFD scan data; The camera is fixedly mounted at the front of the robot, so that the longitudinal direction of the image of the outer wall of the tower is consistent with the robot's forward direction. Taking the upper left corner of the current image of the outer wall of the tower as the origin, the direction from left to right is recorded as the horizontal direction of the image, and the direction from top to bottom is recorded as the vertical direction of the image. Taking the robot's forward direction as the reference, the two drive wheels located on the left side of the robot are recorded as the left drive wheels, and the two drive wheels located on the right side of the robot are recorded as the right drive wheels. The distance between the center line connecting the two left drive wheels and the center line connecting the two right drive wheels is recorded as the center distance between the left and right drive wheels. The edges of the weld seam are extracted using OpenCV image processing tools. The weld seam edge on the left side of the current tower outer wall image is designated as the first side edge, and the weld seam edge on the right side of the current tower outer wall image is designated as the second side edge. The midpoints of the two edges at the same vertical coordinate are connected to obtain the weld seam centerline, and the median of the horizontal distance between the two edges is designated as the weld seam width. The first side edge of the weld seam is extended to the left of the current tower outer wall image by one weld seam width, and the second side edge of the weld seam is extended to the right of the current tower outer wall image by one weld seam width to obtain the weld seam recognition area. From left to right of the current tower outer wall image, a first brightness reference area, a second brightness reference area, a weld seam recognition area, a third brightness reference area, and a fourth brightness reference area are set sequentially, so that the five areas are arranged in ascending order of horizontal coordinate, and the four brightness reference areas are of the same size and do not overlap with the weld seam recognition area. In N consecutive images of the tower outer wall, the region that moves laterally and has a gray level lower than the surrounding tower surface is denoted as the blade sweeping region. When no blade sweeping region appears in the image, N consecutive images of the tower outer wall are selected as reference images, where N is a positive integer not less than 2. The reference images are converted into grayscale images using OpenCV, and the grayscale median of each brightness reference area in each reference image is calculated. The minimum and maximum values ​​among the N grayscale medians are denoted as the lower limit and upper limit of the grayscale median, respectively, and the range between them is denoted as the grayscale median range. The N reference images are registered to the same coordinate system, and the median of the lateral coordinates of the first and second side edges of the weld at each longitudinal position is taken to obtain the first and second side edges of the reference weld. The midpoint between the two reference weld edges is used to obtain the centerline of the reference weld, and the median of the lateral distance between the two reference weld edges is denoted as the width of the reference weld.

[0022] S2. Identify leaf shadow events based on grayscale changes in four brightness reference areas.

[0023] The current tower outer wall image is converted into a grayscale image using OpenCV, and the current grayscale median of the four brightness reference areas is calculated. The acquisition time corresponding to the first time the grayscale median of the first brightness reference area, the second brightness reference area, the third brightness reference area, and the fourth brightness reference area is recorded as the first darkening time, the second darkening time, the third darkening time, and the fourth darkening time, respectively. As the first darkening moment, the second darkening moment, the third darkening moment, and the fourth darkening moment increase sequentially, the current grayscale change is recorded as a blade sweeping event moving from the left to the right of the current tower outer wall image; as the fourth darkening moment, the third darkening moment, the second darkening moment, and the first darkening moment increase sequentially, the current grayscale change is recorded as a blade sweeping event moving from the right to the left of the current tower outer wall image. When all four brightness reference areas fall below their respective grayscale median lower limit for the first time in the tower outer wall image corresponding to the same acquisition time, this brightness change is recorded as an overall illumination change, and subsequent blade vignetting processing is not performed; the earliest darkening moment in the blade vignetting event is recorded as the event start time; when the four current grayscale medians are all within their respective grayscale median range in the images corresponding to N consecutive acquisition numbers, the last acquisition time is recorded as the event end time.

[0024] S3. Calculate the speed of the blade's sweeping leading edge, the arrival time of the first side edge, and the arrival time of the second side edge.

[0025] In the current image of the outer wall of the tower, the average horizontal coordinates of the left and right boundaries of each brightness reference area are taken as the horizontal coordinates of the center of the brightness reference area; the average vertical coordinates of the upper and lower boundaries of the weld identification area are taken as the vertical coordinates of the detection line, and a straight line parallel to the horizontal direction of the current image of the outer wall of the tower is drawn at this vertical coordinate, and this straight line is recorded as the detection line; the horizontal coordinates of the intersections of the detection line with the first side edge and the second side edge of the reference weld are read respectively. For a blade sweeping event moving from left to right in the current tower outer wall image, the lateral distance between the centers of the first and second brightness reference areas is divided by the difference between the second darkening moment and the first darkening moment to obtain the blade sweeping edge movement speed. At the second darkening moment, the blade sweeping edge is located at the center lateral coordinate of the second brightness reference area. The first lateral distance is obtained by subtracting the center lateral coordinate of the second brightness reference area from the lateral coordinate of the intersection of the detection line and the first side edge of the reference weld. The second lateral distance is obtained by subtracting the center lateral coordinate of the second brightness reference area from the lateral coordinate of the intersection of the detection line and the second side edge of the reference weld. The first and second lateral distances are divided by the blade sweeping edge movement speed and added to the second darkening moment to obtain the arrival time of the first and second side edges. For a blade sweeping shadow event moving from right to left in the current tower outer wall image, the lateral distance between the centers of the fourth and third brightness reference areas is divided by the difference between the third and fourth darkening times to obtain the blade sweeping shadow leading edge speed. At the third darkening time, the blade sweeping shadow leading edge is located at the center lateral coordinate of the third brightness reference area. The second side lateral distance is obtained by subtracting the lateral coordinate of the intersection of the detection line and the second side edge of the reference weld from the center lateral coordinate of the third brightness reference area. The first side lateral distance is obtained by subtracting the lateral coordinate of the intersection of the detection line and the first side edge of the reference weld from the center lateral coordinate of the third brightness reference area. The second side lateral distance and the first side lateral distance are divided by the blade sweeping shadow leading edge speed, and then added to the third darkening time to obtain the arrival time of the second side edge and the arrival time of the first side edge. When the darkening times of the two brightness reference areas used to calculate the blade sweeping shadow leading edge speed are the same, the blade sweeping shadow leading edge speed, the arrival time of the first side edge, and the arrival time of the second side edge are not calculated.

[0026] S4. Obtain the weld positioning result based on the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the lengths of the paired edges on both sides.

[0027] The edge map of the weld recognition area is generated using OpenCV's Canny function. The curve composed of interconnected edge pixels is extracted using the findContours function, and each curve is recorded as a continuous edge contour. On both sides of the reference weld center line, the continuous edge contour located to the left of the reference weld center line is recorded as the first side candidate contour, and the continuous edge contour located to the right of the reference weld center line is recorded as the second side candidate contour. For each candidate profile of the first side, at each longitudinal coordinate covered by the candidate profile of the first side and the first edge of the reference weld, calculate the absolute value of the difference between the transverse coordinate of the candidate profile of the first side and the transverse coordinate of the first edge of the reference weld, and record the average value of each absolute value as the average transverse distance of the first side. Select the candidate profile of the first side with the smallest average transverse distance of the first side as the first edge of the current weld. For each candidate profile of the second side, at each longitudinal coordinate covered by the candidate profile of the second side and the second side edge of the reference weld, calculate the absolute value of the difference between the transverse coordinate of the candidate profile of the second side and the transverse coordinate of the second side edge of the reference weld, and record the average value of each absolute value as the average transverse distance of the second side. Select the candidate profile of the second side with the smallest average transverse distance of the second side as the current second side edge of the weld. The `boundingRect` function is called to obtain the bounding rectangles of the first and second side edges of the current weld. The lengths of the two bounding rectangles along the longitudinal direction of the image are recorded as the longitudinal projection lengths of the first and second side edges, respectively. The longitudinal coordinate ranges of the first and second side edges of the current weld are obtained. The length of the overlapping part of the two longitudinal coordinate ranges is recorded as the length of the paired edge. Half of the longitudinal length of the weld recognition area is recorded as the edge usage length. When the longitudinal projection length of the first side edge is not less than the edge usage length, the lateral coordinates of each point on the first side edge of the current weld are increased by half the width of the reference weld to obtain the first side single edge compensation center line; when the longitudinal projection length of the second side edge is not less than the edge usage length, the lateral coordinates of each point on the second side edge of the current weld are decreased by half the width of the reference weld to obtain the second side single edge compensation center line; when the longitudinal projection lengths of the first side edge, the second side edge, and the paired edge lengths are all not less than the edge usage length, the midpoints of the first side edge and the second side edge of the current weld at the same longitudinal coordinate are connected to obtain the double edge weld center line; when the longitudinal projection lengths of the first side edge and the second side edge are both less than the edge usage length, it is determined that the current tower outer wall image cannot provide weld positioning results, and the robot forward movement is stopped. The first side single edge compensation center line, the second side single edge compensation center line, and the double edge weld center line are collectively recorded as the weld positioning result.

[0028] S5. Generate the target rotation amount of the four drive wheels based on the arrival time of the first side edge, the arrival time of the second side edge, and the weld positioning result.

[0029] For blade sweeping events moving from the left to the right of the current tower outer wall image, when the difference between the arrival time of the first side edge and the current acquisition time is less than or equal to the difference between the current acquisition time and the previous acquisition time, the second side single edge compensation center line is used as the weld seam tracking target; when the second side single edge compensation center line is not obtained, the robot's forward movement is stopped. For a blade sweeping event moving from the right to the left in the current tower outer wall image, when the difference between the arrival time of the second side edge and the current acquisition time is less than or equal to the difference between the current acquisition time and the previous acquisition time, the first side single edge compensation center line is used as the weld seam tracking target; if the first side single edge compensation center line is not obtained, the robot's forward movement is stopped. If neither the arrival time of the first side edge nor the arrival time of the second side edge triggers the selection of the single-edge compensation centerline, and the centerline of the double-edge weld has already been obtained, the double-edge weld centerline is used; if only the centerline of the first side single-edge compensation is obtained, the first side single-edge compensation centerline is used; if only the centerline of the second side single-edge compensation is obtained, the second side single-edge compensation centerline is used; if both the first and second side single-edge compensation centerlines are obtained, and the longitudinal projection length of the first side edge is greater than or equal to the longitudinal projection length of the second side edge, the first side single-edge compensation centerline is used; if the longitudinal projection length of the first side edge is less than the longitudinal projection length of the second side edge, the second side single-edge compensation centerline is used; if neither the centerline of the double-edge weld nor any single-edge compensation centerline is obtained, no weld tracking target is generated, the target rotation of all four drive wheels is set to zero, and the robot's forward movement is stopped; when a weld tracking target is generated, subsequent calculations of lateral tracking deviation, directional tracking deviation, and target rotation are performed. The OpenCV fitLine function is called to fit the center line point of the weld tracking target located in the lower third of the weld recognition area. The horizontal coordinate of the intersection point of the fitted line and the lower boundary of the weld recognition area is subtracted from the horizontal center coordinate of the image to obtain the horizontal tracking deviation. The directional tracking deviation is obtained according to the angle between the fitted line and the vertical direction of the image. To obtain the robot's forward velocity, divide the lateral tracking deviation by the lateral width of the weld recognition area to obtain the lateral deviation ratio; divide the directional tracking deviation by 90 degrees to obtain the directional deviation ratio; multiply the sum of the lateral deviation ratio and the directional deviation ratio by the ratio of the forward velocity to the center distance between the left and right drive wheels to obtain the target steering angular velocity; the positive and negative signs of the target steering angular velocity indicate that the robot is turning to the right and left, respectively, and the absolute value of the target steering angular velocity indicates the magnitude of the robot's steering speed. To obtain the radius of the drive wheels, multiply the rotation increments of the four encoders corresponding to the most recent acquisition number before the start of the event by the radius of the drive wheels to get the distance traveled by the four drive wheels. The forward velocity of the event is obtained by dividing the average distance traveled by the four drive wheels by the difference between the acquisition time corresponding to the current acquisition number and the acquisition time corresponding to the previous acquisition number. When using a double-edge weld centerline, the forward velocity of the event is taken as the current forward velocity; when using a first-side single-edge compensation centerline, the ratio of the longitudinal projection length of the first-side edge to the longitudinal length of the weld identification area is multiplied by the forward velocity of the event to obtain the current forward velocity; when using a second-side single-edge compensation centerline, the ratio of the longitudinal projection length of the second-side edge to the longitudinal length of the weld identification area is multiplied by the forward velocity of the event to obtain the current forward velocity. Let the linear velocities of the left and right targets be denoted as follows: and Calculate the linear velocities of the target on the left and the target on the right: ; ; Where v is the current forward velocity, b is the center distance between the left and right drive wheels, and ω is the target steering angular velocity; respectively and Divide by the radius of the drive wheel, then multiply by the difference between the current acquisition time and the previous acquisition time to obtain the target rotation amount of the left drive wheel and the target rotation amount of the right drive wheel, and send them to the corresponding drive wheels.

[0030] S6. Verify the weld tracking results based on the weld location and TOFD scan data after the event ends.

[0031] After the blade sweeping event is determined in S2, the edges of both sides of the weld are re-extracted using OpenCV. When the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the length of the paired edges on both sides are all not less than the edge usage length, the midpoints of the first side edge and the second side edge of the current weld at the same longitudinal coordinate are connected to obtain the center line of the double-edge weld after the event ends. The fitLine function is called to fit the centerline point of the double-edge weld centerline located in the lower third of the weld recognition area after the event ends. The absolute value of the difference between the horizontal coordinate of the intersection of the fitted line and the lower boundary of the weld recognition area and the horizontal center coordinate of the image is recorded as the horizontal deviation at the end of the event; the absolute value of the horizontal tracking deviation corresponding to the most recent acquisition number before the start of the event is recorded as the horizontal deviation at the start of the event. Using TomoView TOFD analysis software, an A-scan image is generated based on the TOFD scan data corresponding to each acquisition number. The A-scan image is a waveform diagram with ultrasonic wave propagation time as the horizontal axis and ultrasonic wave received signal amplitude as the vertical axis. It is determined whether a direct wave and a bottom echo exist simultaneously in each A-scan image, and the acquisition number with both direct wave and bottom echo is recorded as a valid TOFD acquisition number. The number of valid TOFD acquisition numbers in the N most recent acquisition numbers before the start of the event is divided by N to obtain the TOFD continuity ratio before the event. The number of valid TOFD acquisition numbers between the start and end of the event is divided by the total number of acquisition numbers in that period to obtain the TOFD continuity ratio of the event. When the lateral deviation at the end of the event is not greater than the lateral deviation at the beginning of the event, and the continuous TOFD ratio of the event is not lower than the continuous TOFD ratio before the event, the weld tracking result during this blade glide event is deemed valid, and weld tracking continues. When the lateral deviation at the end of the event is greater than the lateral deviation at the beginning of the event, or when the continuous TOFD ratio of the event is lower than the continuous TOFD ratio before the event, the weld tracking result during this blade sweep event is deemed invalid, and the robot's forward movement is stopped. After stopping, the edges on both sides of the weld are re-extracted and the TOFD scan data is checked. If the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the length of the paired edges on both sides corresponding to N consecutive acquisition numbers are not less than the edge usage length, and the N consecutive acquisition numbers are all valid TOFD acquisition numbers, weld tracking is resumed; otherwise, the robot remains in the adsorption state and a manual review prompt is output.

[0032] Please see Figure 2 This invention provides a wind turbine tower climbing robot weld seam tracking control system based on visual intelligent recognition, including a data acquisition and reference acquisition module, a blade shadow event recognition module, an arrival time calculation module, a weld seam positioning module, a drive control module, and a tracking result verification module. The data acquisition and benchmark acquisition module is used to acquire images of the outer wall of the tower, the acquisition time, the encoder rotation increment of the four drive wheels and TOFD scan data and assign them the same acquisition number. It sets four brightness reference areas and weld recognition areas. Based on the benchmark image where no blade grazing area appears, it acquires the grayscale median range of each brightness reference area, the first side edge of the benchmark weld, the second side edge of the benchmark weld, the center line of the benchmark weld and the width of the benchmark weld. The leaf shadow event recognition module is used to identify the movement direction of the leaf shadow event based on the order in which the median gray value of the four brightness reference areas first falls below the lower limit of their respective median gray values, and to record the start and end times of the event. The arrival time calculation module is used to calculate the moving speed of the blade's shadow leading edge based on the difference between the lateral distance between the centers of the brightness reference areas and the darkening time, and to calculate the arrival time of the first side edge and the arrival time of the second side edge. The weld positioning module is used to extract the first side candidate contour and the second side candidate contour, obtain the first side edge and the second side edge of the current weld based on the average lateral distance between the candidate contour and the reference weld edge, and generate the first side single edge compensation center line, the second side single edge compensation center line or the double edge weld center line based on the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge and the length of the paired edges on both sides. The drive control module is used to select the weld tracking target based on the moving direction of the blade sweeping event, the arrival time of the first side edge, the arrival time of the second side edge, and the weld positioning result; calculate the lateral tracking deviation and directional tracking deviation based on the weld tracking target; calculate the forward speed of the event and the current forward speed based on the encoder rotation increment; and generate the target rotation amount of the four drive wheels based on the forward speed and the target steering angular velocity. The tracking result verification module is used to calculate the lateral deviation at the end of the event, the lateral deviation at the beginning of the event, the continuous ratio of TOFD before the event, and the continuous ratio of TOFD during the event after the blade sweep event ends. Based on the deviation comparison results and the TOFD continuous ratio comparison results, it determines whether the weld tracking result is valid, and stops the robot's forward movement when the weld tracking result is invalid.

[0033] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0034] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0035] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition, characterized in that, include: S1. Acquire images of the outer wall of the tower, encoder rotation increments, and TOFD scanning data to obtain the grayscale median range, reference weld edge, reference weld centerline, and reference weld width. S2. Identify leaf glimpse events based on grayscale changes in four brightness reference areas; S3. Calculate the speed of the blade's sweeping leading edge, the arrival time of the first side edge, and the arrival time of the second side edge; S4. Obtain the weld positioning result based on the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the lengths of the paired edges on both sides; S5. Generate the target rotation amount of the four drive wheels based on the arrival time of the first side edge, the arrival time of the second side edge, and the weld positioning result; S6. Verify the weld tracking results based on the weld location and TOFD scan data after the event ends.

2. The method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition as described in claim 1, characterized in that: In S1, images of the tower outer wall, encoder rotation increments, and TOFD scan data are acquired to obtain the grayscale median range, reference weld edge, reference weld centerline, and reference weld width, as follows: Continuously acquire images of the tower's outer wall and record the acquisition time; obtain the cumulative rotation angles recorded by the encoders of the four drive wheels at the current acquisition time and the previous acquisition time; subtract the cumulative rotation angle of the drive wheel at the previous acquisition time from the cumulative rotation angle of the drive wheel at the current acquisition time to obtain the encoder rotation increments of the four drive wheels between the two acquisition times; acquire the TOFD scan data formed between the previous acquisition time and the current acquisition time, and assign the same acquisition number to the current image, acquisition time, encoder rotation increments, and TOFD scan data; The robot's forward direction is consistent with the longitudinal direction of the tower's outer wall image; taking the upper left corner of the current tower's outer wall image as the origin, the direction from left to right is recorded as the horizontal direction of the image, and the direction from top to bottom is recorded as the vertical direction of the image; taking the robot's forward direction as the reference, the two drive wheels located on the left side of the robot are recorded as the left drive wheels, and the two drive wheels located on the right side of the robot are recorded as the right drive wheels; the distance between the line connecting the centers of the two left drive wheels and the line connecting the centers of the two right drive wheels is recorded as the center-to-center distance between the left and right drive wheels; The edges of the weld seam are extracted using OpenCV image processing tools. The weld seam edge on the left side of the current tower outer wall image is designated as the first side edge, and the weld seam edge on the right side of the current tower outer wall image is designated as the second side edge. The midpoints of the two edges at the same vertical coordinate are connected to obtain the weld seam centerline, and the median of the horizontal distance between the two edges is designated as the weld seam width. The first side edge of the weld seam is extended to the left of the current tower outer wall image by one weld seam width, and the second side edge of the weld seam is extended to the right of the current tower outer wall image by one weld seam width to obtain the weld seam recognition area. The first brightness reference area, the second brightness reference area, the weld seam recognition area, the third brightness reference area, and the fourth brightness reference area are set sequentially from left to right of the current tower outer wall image. The region in N consecutive images of the outer wall of the tower that moves laterally and has a gray level lower than the surrounding tower surface is denoted as the blade sweeping region. When no blade shadow area appears in the image, select N consecutive images of the outer wall of the tower as the reference image, where N is a positive integer not less than 2; The reference images are converted to grayscale using OpenCV. The median grayscale value of each brightness reference area in each reference image is calculated. The minimum and maximum values ​​among the N grayscale medians are recorded as the lower limit and upper limit of the grayscale median, respectively, and the range between them is recorded as the grayscale median range. The N reference images are registered to the same coordinate system. The median of the lateral coordinates of the first and second side edges of the weld at each vertical position is taken to obtain the first and second side edges of the reference weld. The midpoint between the two reference weld edges is used to obtain the centerline of the reference weld, and the median of the lateral distance between the two reference weld edges is recorded as the width of the reference weld.

3. The method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition as described in claim 1, characterized in that: In S2, the leaf glimpse event is identified based on the grayscale changes in the four brightness reference areas, as follows: The current tower outer wall image is converted into a grayscale image using OpenCV, and the current grayscale median of the four brightness reference areas is calculated. The acquisition time corresponding to the first time the grayscale median of the first brightness reference area, the second brightness reference area, the third brightness reference area, and the fourth brightness reference area is recorded as the first darkening time, the second darkening time, the third darkening time, and the fourth darkening time, respectively. As the first darkening moment, the second darkening moment, the third darkening moment, and the fourth darkening moment increase sequentially, the current grayscale change is recorded as a blade sweeping event moving from the left to the right of the current tower outer wall image; as the fourth darkening moment, the third darkening moment, the second darkening moment, and the first darkening moment increase sequentially, the current grayscale change is recorded as a blade sweeping event moving from the right to the left of the current tower outer wall image. When all four brightness reference areas are below their respective grayscale median lower limit for the first time in the tower outer wall image corresponding to the same acquisition time, this brightness change is recorded as the overall illumination change; the earliest darkening moment in the blade shadow event is recorded as the event start time; when the four current grayscale medians are all within their respective grayscale median range in the images corresponding to N consecutive acquisition numbers, the last acquisition time is recorded as the event end time.

4. The method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition as described in claim 1, characterized in that: In S3, the moving speed of the blade's sweeping leading edge, the arrival time of the first side edge, and the arrival time of the second side edge are calculated as follows: In the current image of the outer wall of the tower, the average horizontal coordinates of the left and right boundaries of each brightness reference area are taken as the horizontal coordinates of the center of the brightness reference area; the average vertical coordinates of the upper and lower boundaries of the weld identification area are taken as the vertical coordinates of the detection line, and a straight line parallel to the horizontal direction of the current image of the outer wall of the tower is drawn at this vertical coordinate, and this straight line is recorded as the detection line; the horizontal coordinates of the intersections of the detection line with the first side edge and the second side edge of the reference weld are read respectively. For a blade sweeping event moving from left to right in the current tower outer wall image, the lateral distance between the centers of the first and second brightness reference areas is divided by the difference between the second darkening moment and the first darkening moment to obtain the blade sweeping edge movement speed. At the second darkening moment, the blade sweeping edge is located at the center lateral coordinate of the second brightness reference area. The first lateral distance is obtained by subtracting the center lateral coordinate of the second brightness reference area from the lateral coordinate of the intersection of the detection line and the first side edge of the reference weld. The second lateral distance is obtained by subtracting the center lateral coordinate of the second brightness reference area from the lateral coordinate of the intersection of the detection line and the second side edge of the reference weld. The first and second lateral distances are divided by the blade sweeping edge movement speed and added to the second darkening moment to obtain the arrival time of the first and second side edges. For a blade sweeping shadow event moving from right to left in the current tower outer wall image, the lateral distance between the centers of the fourth and third brightness reference areas is divided by the difference between the third and fourth darkening times to obtain the blade sweeping shadow leading edge speed. At the third darkening time, the blade sweeping shadow leading edge is located at the center lateral coordinate of the third brightness reference area. The second side lateral distance is obtained by subtracting the lateral coordinate of the intersection of the detection line and the second side edge of the reference weld from the center lateral coordinate of the third brightness reference area. The first side lateral distance is obtained by subtracting the lateral coordinate of the intersection of the detection line and the first side edge of the reference weld from the center lateral coordinate of the third brightness reference area. The second side lateral distance and the first side lateral distance are divided by the blade sweeping shadow leading edge speed, and then added to the third darkening time to obtain the arrival time of the second side edge and the arrival time of the first side edge. When the darkening times of the two brightness reference areas used to calculate the blade sweeping shadow leading edge speed are the same, the blade sweeping shadow leading edge speed, the arrival time of the first side edge, and the arrival time of the second side edge are not calculated.

5. The method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition according to claim 1, characterized in that: In S4, the weld positioning result is obtained based on the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the lengths of the paired edges on both sides, as follows: The edge map of the weld recognition area is generated using OpenCV's Canny function. The curve composed of interconnected edge pixels is extracted using the findContours function, and each curve is recorded as a continuous edge contour. On both sides of the reference weld center line, the continuous edge contour located to the left of the reference weld center line is recorded as the first side candidate contour, and the continuous edge contour located to the right of the reference weld center line is recorded as the second side candidate contour. For each candidate profile of the first side, at each longitudinal coordinate covered by the candidate profile of the first side and the first edge of the reference weld, calculate the absolute value of the difference between the transverse coordinate of the candidate profile of the first side and the transverse coordinate of the first edge of the reference weld, and record the average value of each absolute value as the average transverse distance of the first side. Select the candidate profile of the first side with the smallest average transverse distance of the first side as the first edge of the current weld. For each candidate profile of the second side, at each longitudinal coordinate covered by the candidate profile of the second side and the second side edge of the reference weld, calculate the absolute value of the difference between the transverse coordinate of the candidate profile of the second side and the transverse coordinate of the second side edge of the reference weld, and record the average value of each absolute value as the average transverse distance of the second side. Select the candidate profile of the second side with the smallest average transverse distance of the second side as the current second side edge of the weld. The `boundingRect` function is called to obtain the bounding rectangles of the first and second side edges of the current weld. The lengths of the two bounding rectangles along the longitudinal direction of the image are recorded as the longitudinal projection lengths of the first and second side edges, respectively. The longitudinal coordinate ranges of the first and second side edges of the current weld are obtained. The length of the overlapping part of the two longitudinal coordinate ranges is recorded as the length of the paired edge. Half of the longitudinal length of the weld recognition area is recorded as the edge usage length. When the longitudinal projection length of the first side edge is not less than the edge usage length, the lateral coordinates of each point on the first side edge of the current weld are increased by half the width of the reference weld to obtain the first side single edge compensation center line; when the longitudinal projection length of the second side edge is not less than the edge usage length, the lateral coordinates of each point on the second side edge of the current weld are decreased by half the width of the reference weld to obtain the second side single edge compensation center line; when the longitudinal projection lengths of the first side edge, the second side edge, and the paired edge lengths are all not less than the edge usage length, the midpoints of the first side edge and the second side edge of the current weld at the same longitudinal coordinate are connected to obtain the double edge weld center line; when the longitudinal projection lengths of the first side edge and the second side edge are both less than the edge usage length, it is determined that the current tower outer wall image cannot provide weld positioning results, and the robot forward movement is stopped. The first side single edge compensation center line, the second side single edge compensation center line, and the double edge weld center line are collectively recorded as the weld positioning result.

6. The method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition according to claim 1, characterized in that: In S5, the target rotation amounts of the four drive wheels are generated based on the arrival times of the first and second side edges and the weld positioning results, as follows: For blade sweeping events moving from the left to the right of the current tower outer wall image, when the difference between the arrival time of the first side edge and the current acquisition time is less than or equal to the difference between the current acquisition time and the previous acquisition time, the second side single edge compensation center line is used as the weld seam tracking target; when the second side single edge compensation center line is not obtained, the robot's forward movement is stopped. For a blade sweeping event moving from the right to the left in the current tower outer wall image, when the difference between the arrival time of the second side edge and the current acquisition time is less than or equal to the difference between the current acquisition time and the previous acquisition time, the first side single edge compensation center line is used as the weld seam tracking target; if the first side single edge compensation center line is not obtained, the robot's forward movement is stopped. If neither the arrival time of the first side edge nor the arrival time of the second side edge triggers the selection of the single-edge compensation centerline, and the centerline of the double-edge weld has already been obtained, the double-edge weld centerline is used; if only the centerline of the first side single-edge compensation is obtained, the first side single-edge compensation centerline is used; if only the centerline of the second side single-edge compensation is obtained, the second side single-edge compensation centerline is used; if both the first and second side single-edge compensation centerlines are obtained, and the longitudinal projection length of the first side edge is greater than or equal to the longitudinal projection length of the second side edge, the first side single-edge compensation centerline is used; if the longitudinal projection length of the first side edge is less than the longitudinal projection length of the second side edge, the second side single-edge compensation centerline is used; if neither the centerline of the double-edge weld nor any single-edge compensation centerline is obtained, no weld tracking target is generated, the target rotation of all four drive wheels is set to zero, and the robot's forward movement is stopped; when a weld tracking target is generated, subsequent calculations of lateral tracking deviation, directional tracking deviation, and target rotation are performed. The OpenCV fitLine function is called to fit the center line point of the weld tracking target located in the lower third of the weld recognition area. The horizontal coordinate of the intersection point of the fitted line and the lower boundary of the weld recognition area is subtracted from the horizontal center coordinate of the image to obtain the horizontal tracking deviation. The directional tracking deviation is obtained according to the angle between the fitted line and the vertical direction of the image. To obtain the robot's forward velocity, divide the lateral tracking deviation by the lateral width of the weld recognition area to obtain the lateral deviation ratio; divide the directional tracking deviation by 90 degrees to obtain the directional deviation ratio; multiply the sum of the lateral deviation ratio and the directional deviation ratio by the ratio of the forward velocity to the center distance between the left and right drive wheels to obtain the target steering angular velocity; the positive and negative signs of the target steering angular velocity indicate that the robot is turning to the right and left, respectively, and the absolute value of the target steering angular velocity indicates the magnitude of the robot's steering speed. To obtain the radius of the drive wheels, multiply the rotation increments of the four encoders corresponding to the most recent acquisition number before the start of the event by the radius of the drive wheels to get the distance traveled by the four drive wheels. The forward velocity of the event is obtained by dividing the average distance traveled by the four drive wheels by the difference between the acquisition time corresponding to the current acquisition number and the acquisition time corresponding to the previous acquisition number. When using the centerline of a double-edge weld, the forward velocity of the event is taken as the current forward velocity; When using the first side single edge compensation centerline, the ratio of the longitudinal projection length of the first side edge to the longitudinal length of the weld identification area is multiplied by the event forward velocity to obtain the current forward velocity; when using the second side single edge compensation centerline, the ratio of the longitudinal projection length of the second side edge to the longitudinal length of the weld identification area is multiplied by the event forward velocity to obtain the current forward velocity. Let the linear velocities of the left and right targets be denoted as follows: and Calculate the linear velocities of the left and right targets respectively: and Divide by the radius of the drive wheel, then multiply by the difference between the current acquisition time and the previous acquisition time to obtain the target rotation amount of the left drive wheel and the target rotation amount of the right drive wheel, and send them to the corresponding drive wheels.

7. The method for tracking and controlling weld seams of a wind turbine tower climbing robot based on visual intelligent recognition according to claim 1, characterized in that: In S6, the weld tracking results are verified based on the weld location and TOFD scan data after the event ends, as follows: After the blade sweeping event is determined in S2, the edges of both sides of the weld are re-extracted using OpenCV. When the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the length of the paired edges on both sides are all not less than the edge usage length, the midpoints of the first side edge and the second side edge of the current weld at the same longitudinal coordinate are connected to obtain the center line of the double-edge weld after the event ends. The fitLine function is called to fit the centerline point of the double-edge weld centerline located in the lower third of the weld recognition area after the event ends. The absolute value of the difference between the horizontal coordinate of the intersection of the fitted line and the lower boundary of the weld recognition area and the horizontal center coordinate of the image is recorded as the horizontal deviation at the end of the event; the absolute value of the horizontal tracking deviation corresponding to the most recent acquisition number before the start of the event is recorded as the horizontal deviation at the start of the event. Using TomoView TOFD analysis software, an A-scan image is generated based on the TOFD scan data corresponding to each acquisition number. The A-scan image is a waveform diagram with ultrasonic wave propagation time as the horizontal axis and ultrasonic wave received signal amplitude as the vertical axis. It is determined whether a direct wave and a bottom echo exist simultaneously in each A-scan image, and the acquisition number with both direct wave and bottom echo is recorded as a valid TOFD acquisition number. The number of valid TOFD acquisition numbers in the N most recent acquisition numbers before the start of the event is divided by N to obtain the TOFD continuity ratio before the event. The number of valid TOFD acquisition numbers between the start and end of the event is divided by the total number of acquisition numbers in that period to obtain the TOFD continuity ratio of the event. When the lateral deviation at the end of the event is not greater than the lateral deviation at the beginning of the event, and the continuous TOFD ratio of the event is not lower than the continuous TOFD ratio before the event, the weld tracking result during this blade glide event is deemed valid, and weld tracking continues. When the lateral deviation at the end of the event is greater than the lateral deviation at the beginning of the event, or when the continuous TOFD ratio of the event is lower than the continuous TOFD ratio before the event, the weld tracking result during this blade sweep event is deemed invalid, and the robot's forward movement is stopped. After stopping, re-extract the edges on both sides of the weld and check the TOFD scanning data. If the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge, and the length of the paired edges on both sides corresponding to N consecutive acquisition numbers are not less than the edge usage length, and the N consecutive acquisition numbers are all valid TOFD acquisition numbers, then resume weld tracking; otherwise, keep the robot in the adsorption state and output a manual review prompt.

8. A wind turbine tower climbing robot weld seam tracking control system based on visual intelligent recognition, used in conjunction with the wind turbine tower climbing robot weld seam tracking control method based on visual intelligent recognition as described in any one of claims 1 to 7, characterized in that, It includes a data acquisition and benchmark acquisition module, a blade sweep event recognition module, an arrival time calculation module, a weld seam positioning module, a drive control module, and a tracking result verification module. The data acquisition and benchmark acquisition module is used to acquire images of the outer wall of the tower, the acquisition time, the encoder rotation increment of the four drive wheels and TOFD scan data and assign them the same acquisition number. It sets four brightness reference areas and weld recognition areas. Based on the benchmark image where no blade grazing area appears, it acquires the grayscale median range of each brightness reference area, the first side edge of the benchmark weld, the second side edge of the benchmark weld, the center line of the benchmark weld and the width of the benchmark weld. The leaf shadow event recognition module is used to identify the movement direction of the leaf shadow event based on the order in which the median gray value of the four brightness reference areas first falls below the lower limit of their respective median gray values, and to record the start and end times of the event. The arrival time calculation module is used to calculate the moving speed of the blade's shadow leading edge based on the difference between the lateral distance between the centers of the brightness reference areas and the darkening time, and to calculate the arrival time of the first side edge and the arrival time of the second side edge. The weld positioning module is used to extract the first side candidate contour and the second side candidate contour, obtain the first side edge and the second side edge of the current weld based on the average lateral distance between the candidate contour and the reference weld edge, and generate the first side single edge compensation center line, the second side single edge compensation center line or the double edge weld center line based on the longitudinal projection length of the first side edge, the longitudinal projection length of the second side edge and the length of the paired edges on both sides. The drive control module is used to select the weld tracking target based on the moving direction of the blade sweeping event, the arrival time of the first side edge, the arrival time of the second side edge, and the weld positioning result; calculate the lateral tracking deviation and directional tracking deviation based on the weld tracking target; calculate the forward speed of the event and the current forward speed based on the encoder rotation increment; and generate the target rotation amount of the four drive wheels based on the forward speed and the target steering angular velocity. The tracking result verification module is used to calculate the lateral deviation at the end of the event, the lateral deviation at the beginning of the event, the continuous ratio of TOFD before the event, and the continuous ratio of TOFD during the event after the blade sweep event ends. Based on the deviation comparison results and the TOFD continuous ratio comparison results, it determines whether the weld tracking result is valid, and stops the robot's forward movement when the weld tracking result is invalid.

Images