Surface damage positioning method suitable for ultrahigh-strength concrete wind power mixing tower

Abstract

The invention discloses a surface damage positioning method suitable for an ultrahigh-strength concrete wind power mixing tower. The surface damage positioning method comprises the steps that benchmark identification pieces containing radio frequency chips and visual recognition patterns are annularly arranged at the preset height of a tower at intervals; the inspection equipment carries a millimeter wave radar, a binocular vision camera and a radio frequency identification module to spirally cruise along the tower drum, and synchronously obtains initial coordinates, real-time image streams and radar reflection signals; correcting the image flow and extracting features, establishing pixel and physical space mapping in combination with the size of the reference identifier, and generating an initial three-dimensional grid model; radar signals are analyzed, mutation feature points are extracted, and a suspected damage area is preliminarily positioned in combination with a three-dimensional grid; controlling equipment to perform fixed-point shooting, extracting damage contour features and calculating actual geometric parameters; the parameters and the physical coordinates are imported into a cylindrical coordinate system conversion model, the damage height, the circumferential angle and the distance between the damage height and the nearest reference identification piece are output, and accurate positioning is achieved; and reliable data support is provided for operation and maintenance of the wind power mixed tower.

Description

Technical Field

[0001] This method relates to the field of monitoring technology for concrete wind turbine hybrid towers, specifically to a method for locating surface damage in ultra-high strength concrete wind turbine hybrid towers. Background Technology

[0002] With the rapid development of the wind power industry, ultra-high strength concrete wind turbine towers, as key load-bearing structures of wind turbine generators, typically exceed 50 meters in height, with some reaching over 100 meters. Long-term exposure to complex natural environments (such as strong winds, temperature differences, rain, snow, and salt spray) makes their surfaces prone to cracking, spalling, and carbonization. If these damages are not detected and repaired in time, they may gradually extend into the structure, leading to decreased load-bearing capacity, deterioration of durability, and even tower collapse. Traditional manual inspection methods are not only inefficient and costly but also pose significant safety risks during high-altitude operations, making them unsuitable for the operation and maintenance needs of large-scale wind farms. Therefore, developing an efficient, accurate, and automated surface damage location technology for wind turbine towers is of significant practical importance for ensuring the safe operation of wind power equipment, reducing operation and maintenance costs, and extending the service life of the tower. This technology enables rapid identification and precise location of damage, providing reliable data support for subsequent assessment and repair, and is a key link in the wind power industry's transformation towards intelligent and unmanned operation and maintenance.

[0003] Traditional wind turbine tower surface damage detection and location technology relies primarily on manual inspection, supplemented by simple tools. Specifically, inspectors visually inspect the tower surface by climbing the tower, riding in an aerial work platform, or using binoculars, marking the location of damage and recording its approximate shape. For minor damage, tools such as tape measures and thickness gauges may be used to measure dimensions, and the height and circumferential location of the damage may be described in conjunction with tower segment markings (such as prefabricated joints).

[0004] Its advantages lie in its simple operation and low initial investment, making it suitable for spot checks in small wind farms or low-risk areas. However, its disadvantages are more prominent: First, it has low accuracy, relying on manual experience for judgment, and the description of damage location is mostly qualitative (e.g., "there is a crack about 30 meters from the bottom"), making it difficult to quantify coordinates and easily overlooking minor damage; second, it is inefficient, requiring several hours or even days to inspect a single tower, making it unsuitable for the inspection needs of large-scale wind farms; third, it has poor safety, with risks of falls and tool drops during high-altitude operations, especially difficult to carry out in severe weather; fourth, it has weak data traceability, as manually recorded damage information is difficult to compare with historical data, making it impossible to assess the damage development trend. In addition, traditional methods have poor adaptability to the cylindrical curved surfaces of towers, with positioning errors of several meters in circumferential angles and heights, seriously affecting the targeted nature of subsequent repairs.

[0005] With advancements in automation technology, existing technologies are increasingly incorporating equipment such as drones and wall-climbing robots to assist in inspection. Drones, equipped with high-definition cameras or infrared sensors, cruise along the tower to capture images, and damage is identified manually or through simple algorithms. GPS positioning is then used to roughly determine the height of the damage. Wall-climbing robots, on the other hand, attach to the tower surface magnetically or through negative pressure, carrying sensors to move and detect damage, recording its location. These technologies, to some extent, improve inspection efficiency and reduce human risk.

[0006] However, existing technologies still have significant limitations: drones are greatly affected by wind, resulting in blurred images, and GPS is easily blocked near the tower, with positioning accuracy mostly between 0.5 and 1 meter, which is insufficient for precise repair needs; the motion trajectory planning of wall-climbing robots is complex, requiring high surface flatness, and they are prone to getting stuck in areas with peeling or cracks; in addition, existing technologies mostly rely on single sensors (such as vision or infrared), which are sensitive to changes in lighting and surface stains, with damage identification accuracy of less than 60%. More importantly, existing technologies have not established a unified coordinate transformation model, making it difficult to correlate damage locations with the tower structure model, failing to provide data support for stress analysis and durability assessment, resulting in "location as the endpoint" and a lack of subsequent application value.

[0007] Chinese patent (application number 202411915381.6) discloses a "surface wave sensor and detection method for wind turbine tower defects." Its core principle is to excite surface waves using a magnetically attracted sensor, determine the presence of defects by utilizing the reflected echoes of these surface waves at the defect locations, and achieve full-circle scanning through sensor movement. This technology relies on the magnetostrictive mechanism, identifying defects by analyzing the time and frequency domain characteristics of the echo signals, and is suitable for detecting near-surface corrosion and weld defects in steel towers.

[0008] However, this technology has serious shortcomings: First, its applicability is limited, only applicable to steel towers, and cannot be adapted to ultra-high strength concrete materials (surface waves attenuate rapidly in concrete, resulting in extremely low signal-to-noise ratio); second, its positioning accuracy is insufficient, only able to estimate the approximate location of defects through the sensor's movement path, and unable to output specific quantitative coordinates such as height and circumferential angle; third, its function is limited, only able to determine the existence of defects, and unable to identify damage types (such as cracks and spalling) and geometric parameters (length and area); fourth, its operation is severely restricted, the sensor needs to be closely attached to the tower surface (lift-off value <1mm), making it difficult to work stably on concrete surfaces with spalling or protrusions, and the magnetic design has insufficient adhesion to concrete, making it prone to falling off. Summary of the Invention

[0009] Based on the above-mentioned technical problems, this application discloses a method for locating surface damage in ultra-high strength concrete wind turbine towers, specifically including:

[0010] S1. Reference markers are set at predetermined heights and circumferential intervals on the ultra-high strength concrete wind power tower. The reference markers have built-in radio frequency chips with unique codes and have visual recognition patterns containing location information on their outer surfaces.

[0011] S2. The inspection equipment equipped with millimeter-wave radar, binocular vision camera and radio frequency identification module performs a spiral cruise motion along the tower axis, and simultaneously executes: reads the encoding information of the reference marker through the radio frequency identification module to obtain the initial position coordinates, collects real-time image streams including the reference marker and the tower surface through the binocular vision camera, and transmits frequency-modulated continuous waves and receives reflected signals through the millimeter-wave radar.

[0012] S3. Perform distortion correction and feature extraction on the real-time image stream, identify the pixel coordinates of the visual recognition pattern in the image, establish the mapping relationship between image pixels and physical space in combination with the actual size of the reference marker, and generate an initial three-dimensional mesh model of the tower surface.

[0013] S4. Perform time-frequency analysis on the reflected signal of the millimeter-wave radar, extract signal change feature points, and combine them with the spatial coordinates of the initial three-dimensional mesh model to preliminarily locate the physical coordinates of the suspected damage area.

[0014] S5. Based on the physical coordinates of the suspected damaged area, control the inspection equipment to take fixed-point multi-angle pictures of the current area to obtain damage detail images, extract damage contour features through image segmentation algorithm, and calculate the actual geometric parameters of the damage, including length, width and area, in combination with the mapping relationship described in S3.

[0015] S6. Import the actual geometric parameters and physical coordinates of the damage into the preset tower cylindrical coordinate system transformation model, and output the height value, circumferential angle value and distance value of the damage on the tower surface, and complete the damage location.

[0016] Preferably, in step S1, reference markers are set at preset height intervals around the ultra-high strength concrete wind turbine tower. Specifically, this means that reference markers are set at preset height intervals along the tower height direction. Select There are several preset height planes, and the vertical distance of each preset height plane from the bottom of the tower is [missing information]. ,in ,and Not exceeding the total height of the tower; within each preset height plane, at angular intervals along the circumferential direction of the tower. Evenly distributed A reference marker, the arc length distance between two adjacent reference markers in the circumferential direction. The formula is: ,in , The outer radius of the tower at the current preset height plane; the visual identification pattern of each reference marker includes the height value of its preset height plane. and the initial angle value in the circumferential direction This forms an orderly division and identification of the different height planes and circumferential positions of the tower.

[0017] Preferably, in step S2, a real-time image stream including the reference marker and the tower surface is acquired using a binocular vision camera. Specifically, the inspection equipment operates at a constant linear velocity. The system performs a spiral cruise along the tower axis, with the binocular vision camera operating at a fixed frame rate. Acquire images and obtain the axial displacement difference between adjacent frames. The formula is: The camera's field of view is along the axis. Circumferential direction Determine the range of tower axial height covered by a single frame image. and circumferential angle range The formula is: Circumferential angle range ,in The vertical distance between the camera and the surface of the tower. To collect the tower radius at the specified height, continuous data collection is required. Frame images, constructing a real-time image stream , For the first The frame contains images of the reference markers and the surface of the tower.

[0018] Preferably, in step S3, distortion correction and feature extraction of the real-time image stream are performed to identify the pixel coordinates of the visual recognition pattern in the image. Specifically, this involves pre-correcting the distortion parameters of the binocular camera using the Zhang calibration method and performing single-frame image... Perform distortion correction formula ,in As the distortion offset, image feature points are extracted using the SIFT algorithm, and a feature response threshold is constructed for the visually recognized pattern. The formula is: ,in , The width and height of the pattern in pixels. Filter response values ​​based on pixel grayscale values. The pattern area, Using an empirical threshold, the boundary ellipse of the identified pattern region is fitted using the least squares method to obtain the coordinates of the ellipse's center pixel. The formula is: ,in For boundary pixel coordinates, To fit the ellipse radius, output the pixel coordinates of the pattern in the image coordinate system. .

[0019] Preferably, in step S3, the mapping relationship between image pixels and physical space is established based on the actual size of the reference marker to generate an initial three-dimensional mesh model of the tower surface. Specifically, this involves obtaining the actual dimensions of the reference marker: length and width... In the image, the pixel size is Calculate pixel-physical mapping coefficients The formula is: For all reference markers in the image stream, the formula is used. Establish a mapping, where The center pixel coordinates of the image. Using physical plane coordinates and combining the real-time pose of the inspection equipment, the image pixels are converted into coordinates in the cylindrical coordinate system of the tower. ,in The axial height, For circumferential angle, The height is the tower radius; by traversing the feature points of the image stream and using the coordinates of the reference marker as constraints, an initial three-dimensional mesh is constructed using the Delaunay triangulation algorithm to generate the initial three-dimensional mesh model of the tower surface.

[0020] Preferably, in step S4, time-frequency analysis is performed on the reflected signal of the millimeter-wave radar to extract signal abrupt change feature points. Specifically, this involves acquiring the reflected signal received by the millimeter-wave radar. Preprocessing is performed to remove the DC component, and then the data is processed according to the time window length. Perform frame segmentation to obtain frame signals The formula is: ,in For the Hanning window function, Using the frame number as the frame number, perform a short-time Fourier transform on each frame signal to obtain the time spectrum. The formula is: ,in For frequency, For time, calculate the power spectral density of the time spectrum. Through power spectral density Calculate the signal abrupt change characteristic quantity The formula is: Set mutation threshold , For empirical coefficients, when When a signal abrupt change is detected at the corresponding position in the current frame, the center time of the abruptly changed frame is extracted. and the corresponding propagation distance of the radar transmitted wave The formula is: ,in For the speed of light, As a characteristic point of signal abrupt change.

[0021] Preferably, in step S4, the physical coordinates of the suspected damaged area are initially located by combining the spatial coordinates of the initial three-dimensional mesh model. Specifically, this involves identifying the signal abrupt change feature points of the millimeter-wave radar. Convert to three-dimensional coordinates in radar coordinate system ,in , The radar elevation angle, This represents the radar azimuth angle; obtained through a coordinate transformation matrix. Convert the radar coordinate system coordinates to the tower's global coordinate system coordinates. Traverse the node set of the initial 3D mesh model Calculate the Euclidean distance between the global coordinates and the mesh nodes. ,filter The nodes constitute the candidate region, among which The distance threshold is the geometric center of the candidate region. As the physical coordinates of the suspected damaged area, This represents the number of candidate nodes.

[0022] Preferably, in step S5, based on the physical coordinates of the suspected damaged area, the inspection equipment performs fixed-point multi-angle photography of the current area to obtain detailed images of the damage, and extracts the damage contour features using an image segmentation algorithm. Specifically, the inspection equipment moves to the physical coordinates of the suspected damaged area. For the corresponding orbital trajectory, damage detail images are captured at azimuth and elevation angle intervals, centered on the coordinates. An improved U-Net segmentation algorithm is applied to the acquired detail images, incorporating an attention-based loss function. Training the model, where For cross-entropy loss, To compare the losses, , For the corresponding weight coefficients, output the binarized mask of the damaged region. ,in =1 indicates a damaged pixel. =0 represents background pixels; damage contours are extracted based on masks, and the contour pixel sequence is obtained through a chain code algorithm. Calculate the minimum bounding rectangle of the contour, with pixel width and height respectively. Through formula Calculate the circularity of the contour and obtain the feature parameters of the damaged contour, where The perimeter of the outline in pixels. The area of ​​the damaged pixel.

[0023] Preferably, in step S5, the actual geometric parameters of the damage are calculated in conjunction with the mapping relationship described in step S3, specifically by using the pixel-physical space mapping coefficient. The pixel parameters of the damaged contour are converted into actual physical parameters, where the actual width of the damage is... Actual height The formula for the actual area of ​​damage is: For linear damage, extract the pixel coordinates of the two points furthest apart in the contour sequence. and The actual length is Combined with the normal vector of the damaged region in the initial 3D mesh model Through formula Calculate the angle between the damage length direction and the normal vector of the tower surface, where Given the damage length direction vector, the output includes the actual geometric parameters of the damage, including length, width, area, and angle.

[0024] Preferably, in step S6, the actual geometric parameters and physical coordinates of the damage are imported into a preset tower cylindrical coordinate system transformation model, and the height value, circumferential angle value, and distance value relative to the nearest reference marker of the damage on the tower surface are output. Specifically, the physical coordinates of the damage are... Import the tower cylindrical coordinate system transformation model and calculate the height of the damage on the tower surface. Calculate the circumferential angle value The formula is: From all benchmark markers, select the marker with the smallest difference in height from the damage, satisfying the requirement... A collection of identifiers, among which Using the height of the reference marker as an example, calculate the spatial straight-line distance between the damage and each marker from the marker set, using the following formula: , The minimum value is taken as the physical coordinate of the reference marker. The output includes the distance value relative to the nearest reference marker. , and Damage localization.

[0025] Compared with the prior art, the technical solution of this application has the following technical effects:

[0026] This application establishes a precise mapping relationship between image pixels and physical space by setting a reference marker with an RF chip and a visual recognition pattern, combined with multi-sensor data fusion of the inspection equipment. After transformation to a cylindrical coordinate system, it can clearly output the height of the damage on the tower surface, the circumferential angle, and the distance to the nearest reference marker, thus solving the problem of ambiguous damage location description and providing accurate location guidance for subsequent maintenance.

[0027] This application collects multiple types of data simultaneously during the spiral cruise of the inspection equipment along the tower. After the suspected damage area is initially located by millimeter-wave radar, the automatic control equipment performs fixed-point multi-angle shooting and extracts damage features by combining image segmentation algorithms. The entire process can complete damage identification and parameter calculation without human intervention, which effectively improves the automation level of the detection process and reduces the tediousness and uncertainty of manual operation.

[0028] This application employs millimeter-wave radar and a binocular vision camera working together. The millimeter-wave radar is unaffected by environmental factors such as light and surface stains, and can stably identify suspected damage. The binocular vision can capture damage details. The combination of the two enhances the ability to identify damage of different types and under different environments, ensuring that damage can still be accurately identified and its contour features extracted under complex conditions.

[0029] This application can clearly understand the specific shape and distribution of damage by calculating the actual geometric parameters of the damage (length, width, area, etc.) and associating them with its spatial coordinates. This information can be directly used to analyze the impact of the damage on the tower structure, making it easier to formulate targeted maintenance plans. At the same time, it provides a complete data foundation for long-term tracking of damage changes and assessment of structural status.

[0030] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0031] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0033] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:

[0034] Figure 1 Flowchart of a method for locating surface damage in ultra-high strength concrete wind turbine towers;

[0035] Figure 2 Design diagram for the distribution of benchmark markers for health monitoring of 60-meter-class tower structures;

[0036] Figure 3 A graph showing the height-radius relationship of a 60-meter-class tower structure;

[0037] Figure 4 Design diagram for the distribution of benchmark markers for health monitoring of 100-meter-class tower structures;

[0038] Figure 5 A graph showing the height-radius relationship of a 100-meter-class tower structure;

[0039] Figure 6 The thermal distribution map of sudden changes in radar signal for surface crack detection;

[0040] Figure 7 This is a stitched image of multi-view visual recognition of surface cracks.

[0041] Figure 8 A cloud distribution map of abrupt changes in radar signals for concrete spalling detection;

[0042] Figure 9 A graph showing the change in radar signal intensity for surface crack detection;

[0043] Figure 10 Thermodynamic enhancement diagram of sudden changes in radar signal for concrete spalling detection;

[0044] Figure 11 A comparison chart showing the environmental adaptability of surface crack detection and concrete spalling detection.

[0045] Figure 12 This is a schematic diagram of the motion strategy for detecting surface cracks and concrete spalling. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0047] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0048] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0049] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0050] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0051] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0052] Example 1

[0053] This embodiment mainly describes a method for locating surface damage in ultra-high strength concrete wind turbine towers, such as... Figure 1 As shown, it includes:

[0054] S1. Reference markers are set at predetermined heights and circumferential intervals on the ultra-high strength concrete wind power tower. The reference markers have built-in radio frequency chips with unique codes and have visual recognition patterns containing location information on their outer surfaces.

[0055] S2. The inspection equipment equipped with millimeter-wave radar, binocular vision camera and radio frequency identification module performs a spiral cruise motion along the tower axis, and simultaneously executes: reads the encoding information of the reference marker through the radio frequency identification module to obtain the initial position coordinates, collects real-time image streams including the reference marker and the tower surface through the binocular vision camera, and transmits frequency-modulated continuous waves and receives reflected signals through the millimeter-wave radar.

[0056] S3. Perform distortion correction and feature extraction on the real-time image stream, identify the pixel coordinates of the visual recognition pattern in the image, establish the mapping relationship between image pixels and physical space in combination with the actual size of the reference marker, and generate an initial three-dimensional mesh model of the tower surface.

[0057] S4. Perform time-frequency analysis on the reflected signal of the millimeter-wave radar, extract signal change feature points, and combine them with the spatial coordinates of the initial three-dimensional mesh model to preliminarily locate the physical coordinates of the suspected damage area.

[0058] S5. Based on the physical coordinates of the suspected damaged area, control the inspection equipment to take fixed-point multi-angle pictures of the current area to obtain damage detail images, extract damage contour features through image segmentation algorithm, and calculate the actual geometric parameters of the damage, including length, width and area, in combination with the mapping relationship described in S3.

[0059] S6. Import the actual geometric parameters and physical coordinates of the damage into the preset tower cylindrical coordinate system transformation model, and output the height value, circumferential angle value and distance value of the damage on the tower surface, and complete the damage location.

[0060] Furthermore, in S1, reference markers are set at preset height circumferential intervals on the ultra-high strength concrete wind turbine tower. Specifically, in the tower height direction, reference markers are set at preset height intervals. Select There are several preset height planes, and the vertical distance of each preset height plane from the bottom of the tower is [missing information]. ,in ,and Not exceeding the total height of the tower; within each preset height plane, at angular intervals along the circumferential direction of the tower. Evenly distributed A reference marker, the arc length distance between two adjacent reference markers in the circumferential direction. The formula is: ,in , The outer radius of the tower at the current preset height plane; the visual identification pattern of each reference marker includes the height value of its preset height plane. and the initial angle value in the circumferential direction This forms an orderly division and identification of the different height planes and circumferential positions of the tower.

[0061] Furthermore, in S2, a real-time image stream containing the reference marker and the tower surface is acquired using a binocular vision camera. Specifically, the inspection equipment operates at a constant linear velocity. The system performs a spiral cruise along the tower axis, with the binocular vision camera operating at a fixed frame rate. Acquire images and obtain the axial displacement difference between adjacent frames. The formula is: The camera's field of view is along the axis. Circumferential direction Determine the range of tower axial height covered by a single frame image. and circumferential angle range The formula is: Circumferential angle range ,in The vertical distance between the camera and the surface of the tower. To collect the tower radius at the specified height, continuous data collection is required. Frame images, constructing a real-time image stream , For the first The frame contains images of the reference markers and the surface of the tower.

[0062] Furthermore, in S3, distortion correction and feature extraction are performed on the real-time image stream to identify the pixel coordinates of the visual recognition pattern in the image. Specifically, the distortion parameters of the binocular camera are pre-corrected using the Zhang calibration method, and the single-frame image is processed. Perform distortion correction formula ,in As the distortion offset, image feature points are extracted using the SIFT algorithm, and a feature response threshold is constructed for the visually recognized pattern. The formula is: ,in , The width and height of the pattern in pixels. Filter response values ​​based on pixel grayscale values. The pattern area, Using an empirical threshold, the boundary ellipse of the identified pattern region is fitted using the least squares method to obtain the coordinates of the ellipse's center pixel. The formula is: ,in For boundary pixel coordinates, To fit the ellipse radius, output the pixel coordinates of the pattern in the image coordinate system. .

[0063] Furthermore, in S3, the mapping relationship between image pixels and physical space is established based on the actual size of the reference marker, generating an initial three-dimensional mesh model of the tower surface. Specifically, this involves obtaining the actual dimensions of the reference marker: length and width... In the image, the pixel size is Calculate pixel-physical mapping coefficients The formula is: For all reference markers in the image stream, the formula is used. Establish a mapping, where The center pixel coordinates of the image. Using physical plane coordinates and combining the real-time pose of the inspection equipment, the image pixels are converted into coordinates in the cylindrical coordinate system of the tower. ,in The axial height, For circumferential angle, The height is the tower radius; by traversing the feature points of the image stream and using the coordinates of the reference marker as constraints, an initial three-dimensional mesh is constructed using the Delaunay triangulation algorithm to generate the initial three-dimensional mesh model of the tower surface.

[0064] Furthermore, in S4, time-frequency analysis is performed on the reflected signal of the millimeter-wave radar to extract signal abrupt change feature points. Specifically, this involves acquiring the reflected signal received by the millimeter-wave radar. Preprocessing is performed to remove the DC component, and then the data is processed according to the time window length. Perform frame segmentation to obtain frame signals The formula is: ,in For the Hanning window function, Using the frame number as the frame number, perform a short-time Fourier transform on each frame signal to obtain the time spectrum. The formula is: ,in For frequency, For time, calculate the power spectral density of the time spectrum. Through power spectral density Calculate the signal abrupt change characteristic quantity The formula is: Set mutation threshold , For empirical coefficients, when When a signal abrupt change is detected at the corresponding position in the current frame, the center time of the abruptly changed frame is extracted. and the corresponding propagation distance of the radar transmitted wave The formula is: ,in For the speed of light, As a characteristic point of signal abrupt change.

[0065] Furthermore, in S4, the spatial coordinates of the initial 3D mesh model are combined to initially locate the physical coordinates of the suspected damaged area. Specifically, this involves identifying the signal abrupt change characteristic points of the millimeter-wave radar. Convert to three-dimensional coordinates in radar coordinate system ,in , The radar elevation angle, This represents the radar azimuth angle; obtained through a coordinate transformation matrix. Convert the radar coordinate system coordinates to the tower's global coordinate system coordinates. Traverse the node set of the initial 3D mesh model Calculate the Euclidean distance between the global coordinates and the mesh nodes. ,filter The nodes constitute the candidate region, among which The distance threshold is the geometric center of the candidate region. As the physical coordinates of the suspected damaged area, This represents the number of candidate nodes.

[0066] Furthermore, in S5, based on the physical coordinates of the suspected damaged area, the inspection equipment performs fixed-point multi-angle photography of the current area to obtain detailed damage images, and extracts damage contour features through an image segmentation algorithm. Specifically, the inspection equipment moves to the physical coordinates of the suspected damaged area. For the corresponding orbital trajectory, damage detail images are captured at azimuth and elevation angle intervals, centered on the coordinates. An improved U-Net segmentation algorithm is applied to the acquired detail images, incorporating an attention-based loss function. Training the model, where For cross-entropy loss, To compare the losses, , For the corresponding weight coefficients, output the binarized mask of the damaged region. ,in =1 indicates a damaged pixel. =0 represents background pixels; damage contours are extracted based on masks, and the contour pixel sequence is obtained through a chain code algorithm. Calculate the minimum bounding rectangle of the contour, with pixel width and height respectively. Through formula Calculate the circularity of the contour and obtain the feature parameters of the damaged contour, where The perimeter of the outline in pixels. The area of ​​the damaged pixel.

[0067] Furthermore, in S5, the actual geometric parameters of the damage are calculated in conjunction with the mapping relationship described in S3, specifically by using the pixel-physical space mapping coefficient. The pixel parameters of the damaged contour are converted into actual physical parameters, where the actual width of the damage is... Actual height The formula for the actual area of ​​damage is: For linear damage, extract the pixel coordinates of the two points furthest apart in the contour sequence. and The actual length is Combined with the normal vector of the damaged region in the initial 3D mesh model Through formula Calculate the angle between the damage length direction and the normal vector of the tower surface, where Given the damage length direction vector, the output includes the actual geometric parameters of the damage, including length, width, area, and angle.

[0068] Furthermore, in S6, the actual geometric parameters and physical coordinates of the damage are imported into a preset tower cylindrical coordinate system transformation model, and the height value, circumferential angle value, and distance value relative to the nearest reference marker of the damage on the tower surface are output. Specifically, the physical coordinates of the damage are... Import the tower cylindrical coordinate system transformation model and calculate the height of the damage on the tower surface. Calculate the circumferential angle value The formula is: From all benchmark markers, select the marker with the smallest difference in height from the damage, satisfying the requirement... A collection of identifiers, among which Using the height of the reference marker as an example, calculate the spatial straight-line distance between the damage and each marker from the marker set, using the following formula: , The minimum value is taken as the physical coordinate of the reference marker. The output includes the distance value relative to the nearest reference marker. , and Damage localization.

[0069] This implementation details how to achieve precise damage localization by using multi-sensor fusion technology (millimeter-wave radar + binocular vision + radio frequency identification) to establish absolute coordinates through reference markers, combined with 3D mesh modeling and cylindrical coordinate system transformation. This technology is applicable to concrete towers and can output damage type, geometric parameters, and spatial coordinates, filling the technological gap in precise localization of surface damage on ultra-high strength concrete wind turbine towers.

[0070] Based on Example 1, this example describes in detail the distribution scheme and compatibility verification of the reference markers under different tower heights, specifically:

[0071] The design benchmark component distribution schemes for ultra-high strength concrete wind turbine hybrid towers of two typical heights, 60 meters and 100 meters, were selected. The 60-meter tower has a bottom diameter of 8.5 meters and a top diameter of 4.2 meters, with the wall thickness gradually decreasing from 0.6 meters at the bottom to 0.3 meters at the top. It is conical in shape and divided into six prefabricated sections, each 10 meters high. The 100-meter tower has a bottom diameter of 10.2 meters and a top diameter of 5.0 meters, with the wall thickness gradually decreasing from 0.8 meters at the bottom to 0.4 meters at the top. It is divided into 13 prefabricated sections, each approximately 7.7 meters high (the top section is 6.9 meters high).

[0072] The reference marker is a 150mm diameter, 20mm thick circular 304 stainless steel plate with a white reflective coating. The center contains a black QR code displaying the height (accurate to 0.1 meters), circumferential angle (accurate to 0.1°), installation date, and tower number. An embedded passive RFID chip is used, with the encoding format "ID-height-angle" (e.g., "ID-10-60" corresponds to a height of 10 meters and an azimuth of 60°). The marker is fixed to the tower surface with expansion bolts, ensuring a gap of ≤1mm between the marker and the tower wall for a level installation.

[0073] For the distribution of reference markers for a 60-meter tower, six height planes are selected at preset height intervals H=10 meters, namely 10 meters, 20 meters, 30 meters, 40 meters, 50 meters, and 60 meters from the bottom (corresponding to the top of each precast section). The outer radii of each plane are 7.8 meters, 7.1 meters, 6.4 meters, 5.7 meters, 5.0 meters, and 4.2 meters, respectively. Within each height plane, six reference markers are evenly placed along the circumferential direction at angular intervals θ=60° (360° / 60°=6, satisfying full circumferential coverage). The arc length distances between adjacent markers are 8.17 meters, 7.43 meters, 6.70 meters, 5.96 meters, 5.23 meters, and 4.39 meters, respectively, with circumferential starting angles of 0°, 60°, 120°, 180°, 240°, and 300° (with the eastward direction from the bottom of the tower as the 0° reference). Figures 2-3 As shown, the six red dots marking the height planes (10 meters, 20 meters...60 meters) correspond to the top positions of the precast tower sections. Each plane has six black dots (reference markers) evenly distributed circumferentially at 60° intervals. Combined with the radius data in the side notes (7.8 meters for 10-meter height, 4.2 meters for 60-meter height), the distribution of the markers follows the tapered structure of the tower. From bottom to top, the arc length of adjacent markers decreases from 8.17 meters to 4.39 meters. Figure 3The smooth downward trend of the height-radius relationship curve on the right side of the figure is completely consistent. The distribution of the reference markers takes into account the characteristics of the tower structure (dividing the height plane according to the prefabricated section) and ensures full spatial coverage through circumferential angle intervals, providing a uniform and stable coordinate reference for subsequent damage location.

[0074] For the distribution of reference markers for 100-meter towers, 13 height planes are selected at preset height intervals H=8 meters for the 100-meter towers, including 8 meters from the bottom, 16 meters...96 meters (a total of 12) and the top 100 meters. The outer radii of each plane are 9.5 meters, 8.9 meters, 8.3 meters, 7.7 meters, 7.1 meters, 6.5 meters, 5.9 meters, 5.6 meters, 5.3 meters, 5.0 meters, 4.7 meters, 4.4 meters and 5.0 meters respectively. Eight reference markers (360° / 45°=8) are evenly spaced along the circumferential direction at angular intervals of θ=45° on each height plane. The arc lengths of adjacent markers are 7.46m, 6.99m, 6.51m, 6.04m, 5.57m, 5.10m, 4.63m, 4.39m, 4.16m, 3.92m, 3.69m, 3.45m, and 3.92m, respectively. The starting angles along the circumference are 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°. As shown in Figure 4-5, the 13 height planes marked with red dots in the side view (8m, 16m...100m) are distributed at 8-meter intervals, which is denser than that of the 60-meter tower. The eight black dots (reference markers) on each plane are arranged at 45° angular intervals. The radius data (8m height 9.5m) is noted in the side annotation. The arc length of the marker (5.0 meters at 100 meters and 5.0 meters at 100 meters) decreases from 7.46 meters to 3.45 meters (96 meters high), with only the top 100 meters slightly increasing to 3.92 meters due to structural changes. Figure 5 The trend of the right-hand height-radius curve is consistent. The distribution of markers on the 100-meter tower improves the spatial positioning accuracy of the tall structure by increasing the number of height planes (13 vs. 6) and the number of circumferential markers (8 vs. 6). At the same time, the special adjustment from 96 meters to 100 meters reflects the adaptability to the irregular structure at the top of the tower.

[0075] Through actual installation testing, the binocular visual recognition rate and radio frequency reading success rate of the 36 markers on the 60-meter tower were both 100%. For the 104 markers on the 100-meter tower, only two markers at the top 100 meters required secondary recognition due to wind vibration. The overall recognition rate reached 98.1%, and the radio frequency reading success rate was 100%. Taking the 30-meter (60-meter class) and 40-meter (100-meter class) height planes as examples, the actual coordinates of the markers deviated from the design coordinates by ≤±5mm and ≤±8mm respectively, meeting the accuracy requirements of 3D modeling. When the inspection equipment cruised along the spiral trajectory, the millimeter-wave radar scanning range, binocular visual field of view, and marker distribution area were completely covered. There were no collisions or signal loss during equipment movement. The real-time display of the marker recognition frame and the QR code matching degree reached 100%, verifying the adaptability of the distribution density to equipment operation.

[0076] This embodiment uses a differentiated distribution scheme, with 60-meter towers using a lower density (36 units) to balance cost and accuracy, and 100-meter towers using a higher density (104 units) to ensure the stability of the tall structure. The unified identification parameters facilitate standardized production and installation. The interspersed distribution details, size correlations, and equipment coordination status visually demonstrate the adaptability of the scheme to towers of different heights, providing a reliable reference coordinate system for subsequent damage location.

[0077] Based on Example 1, this example describes in detail the detection and localization process and parameter verification for different damage types, specifically as follows:

[0078] Two typical damage types commonly seen in ultra-high strength concrete wind turbine towers (surface cracks and concrete spalling) were selected for detection and positioning experiments. A 60-meter-class ultra-high strength concrete wind turbine tower (parameters same as in Example 1) was used, and two types of artificially simulated damage were pre-set on its surface:

[0079] Type A (Surface Crack): Select a 120° position on a plane at a height of 30 meters (outer circle radius of 6.4 meters) and set a longitudinal crack with a length of 2.5 meters, a width of 0.3 mm, and a depth of 5 mm. The crack direction is parallel to the tower axis (deviation ≤ 3°).

[0080] Type B (Concrete spalling): Select a spalling area at a 270° position on a plane at a height of 40 meters (outer circle radius of 5.7 meters). The spalling area is approximately an irregular polygon with the longest side being 1.8 meters, the shortest side being 0.9 meters, an area of ​​approximately 1.2 square meters, and a spalling depth of 15 mm (exposing the internal steel reinforcement protective layer).

[0081] The inspection equipment used in the experiment is equipped with millimeter-wave radar (working frequency 77GHz), binocular vision camera (resolution 2048×1536, frame rate 30fps) and radio frequency identification module (identification distance 0.5-3 meters), and cruises along the tower in a spiral manner (linear speed 0.6m / s, spiral angle 15°).

[0082] Detect and locate surface cracks (Type A), in the initial location stage:

[0083] When the inspection equipment cruised to a height range of 28-32 meters, the millimeter-wave radar began to detect abnormal reflection signals. Within the 118°-122° azimuth range of the 30-meter height plane, the power spectral density of the radar reflection signal exhibited continuous abrupt changes, with a total of 12 signal abrupt change characteristic points detected, distributed within an area of ​​29.8-30.2 meters axially and 119°-121° circumferentially. Based on the initial 3D mesh model (node ​​spacing 0.1 meters) of this height plane, the physical coordinates of the suspected damage area were calculated as (X=-4.2 meters, Y=5.1 meters, Z=30.0 meters), corresponding to a height of 30.0 meters and a circumferential angle of 120° in the tower cylindrical coordinate system, with a deviation from the preset crack location ≤0.2 meters.

[0084] The inspection equipment adjusted its pose based on the coordinates of the suspected area, capturing nine detailed images centered on the crack at azimuth intervals of 15° (0°, 15°, 345°) and elevation intervals of 10° (-5°, 0°, 5°). In the images acquired by the binocular vision camera, the crack area exhibited dark linear features. After segmentation using an improved U-Net algorithm, the resulting binarized mask had pixel dimensions of 2500 pixels in length and 3 pixels in width. The extracted crack contour pixel sequence contained 256 points, with a minimum bounding rectangle aspect ratio of 833:1 and a circularity C=0.012 (approaching a straight line feature).

[0085] Based on the pixel-physical mapping coefficient k = 0.001m / pixel (calibrated from a reference marker at a height of 30 meters), the actual parameters of the crack are calculated as follows: length = 2500 × 0.001 = 2.5 meters, width = 3 × 0.001 = 0.003 meters (i.e., 0.3 mm), which is completely consistent with the preset values. Through a cylindrical coordinate system transformation model, the crack's location is output as follows: height 30.02 meters (from the bottom), circumferential angle 120.1°, and straight-line distance of 0.15 meters from the nearest reference marker (120° orientation at a height of 30 meters).

[0086] like Figure 6 As shown, the millimeter-wave radar detected dense signal abrupt changes within a circumferential range of 118°-122° and a height range of 29.8-30.2 meters, confirming the strong radar wave reflection characteristics of the crack region and initially identifying the suspected damage area; Figure 7As shown, after stitching together nine detailed images from different angles, the crack edge fully reveals its linear characteristics. The location results clearly indicate that the crack is located at a height of 30.02 meters and a circumferential position of 120.1° on the tower, with a length of 2.5 meters and a width of 0.3 millimeters. It is only 0.15 meters away from the nearest reference marker (ID-30-120). The cross-sectional micrograph further confirms that the crack depth is 5 millimeters, accurately identifying the geometric parameters of the linear crack and precisely locating it.

[0087] Detection and location of concrete spalling (Type B), preliminary location stage:

[0088] When the inspection equipment cruised to a height range of 38-42 meters, the millimeter-wave radar detected a sudden change in strong reflection signal within a 265°-275° azimuth range at a height of 40 meters. A total of 28 signal change feature points were identified, distributed in an irregular polygonal area. Based on the initial 3D mesh model (node ​​spacing 0.1 meters) at the 40-meter height plane, the physical coordinates of the suspected damage area were calculated as (X=0.8 meters, Y=-5.6 meters, Z=40.0 meters), corresponding to a height of 40.0 meters and a circumference of 270° in the cylindrical coordinate system of the tower, with a deviation from the preset spalling location of ≤0.3 meters.

[0089] The inspection equipment captured nine detailed images around the suspected area at azimuth intervals of 20° (250°, 270°, 290°) and elevation intervals of 15° (-10°, 0°, 10°), covering the entire boundary of the peeling area. In the images acquired by the binocular vision camera, the peeling area appeared grayish-white (contrasting with the dark gray of the surrounding concrete). After segmentation using the improved U-Net algorithm, the pixel area of ​​the binarized mask was 1,200,000 pixels, the contour pixel sequence contained 1024 points, the pixel size of the minimum bounding rectangle was 2000×1000 (length×width), and the roundness C=0.68 (approaching polygonal features).

[0090] Based on the pixel-physical mapping coefficient k=0.001m / pixel of the 40-meter height plane, the actual parameters of the peeling area are calculated as follows: Area = 1,200,000 × (0.001)² = 1.2 square meters, longest side = 2000 × 0.001 = 2.0 meters (the deviation from the preset 1.8 meters is due to boundary irregularities), shortest side = 1000 × 0.001 = 1.0 meters (the deviation from the preset 0.9 meters is ≤0.1 meters). Through the cylindrical coordinate system transformation model, the positioning result of the peeling area is output as follows: height 40.05 meters (from the bottom), circumference 269.8°, and straight-line distance of 0.08 meters from the nearest reference marker (270° orientation of the 40-meter height plane).

[0091] As shown in Figure 8, the millimeter-wave radar detected 28 signal abrupt change points within a circumferential range of 265°-275° and a height range of 38-42 meters, forming a dense red area. The strong reflection characteristics of radar waves in the spalled area quickly delineated the suspected damage range and effectively identified the spalled boundary and internal state. It has the ability to detect damage in large areas, accurately locate the spalled position, and quantify geometric parameters.

[0092] Verification of multi-sensor collaborative detection effectiveness: Comparison of sensor data for Type A (surface cracks):

[0093] Millimeter-wave radar: The reflected signal in the crack area changed abruptly 12 times, and the signal strength was 40% higher than that of the surrounding area on average. The distance at which the crack was first detected was 3.5 meters (the straight-line distance between the equipment and the crack).

[0094] The pixel recognition rate of the crack obtained by the binocular vision camera was 98.7% (only some pixels were missed due to lighting within 0.1 meters at both ends), and the integrity of the contour extraction reached 96.3%.

[0095] Positioning accuracy: The final output of crack height and circumferential angle deviates from the preset position by ±0.02 meters and ±0.2°, respectively, and the distance deviation from the nearest reference marker is ±0.01 meters.

[0096] like Figure 9 As shown, the radar signal strength curve shows a significant peak at 3.5 meters, with the signal strength increasing by 40% compared to the background. This indicates that the radar can accurately capture the reflection characteristics of cracks at a relatively long distance, providing a basis for early detection. Meanwhile, the crack outline identified by vision has a 96% overlap with the actual crack, with only slight deviations at both ends due to illumination, verifying the accuracy of the image segmentation algorithm in extracting linear features. The bottom data frame shows that the multi-sensor fusion positioning error is ≤0.03 meters, with radar contributing 60% of the positioning information and vision contributing 40%. Radar provides initial position guidance, while vision optimizes detailed parameters. By fusing radar and vision, the limitations of a single sensor are overcome.

[0097] Comparison of sensor data for Type B (concrete spalling):

[0098] Millimeter-wave radar: The reflected signal in the spalled area changed abruptly 28 times, and the signal strength was 65% higher than that of the surrounding area on average. The distance at which the spalled area was first detected was 5.2 meters (the straight-line distance between the equipment and the spalled area).

[0099] Binocular vision camera: In 9 detailed images, the pixel recognition rate of the peeled area is 99.2%, and the integrity of the contour extraction is 98.5% (only the 0.05 square meter area in the upper right corner has a slight deviation due to shadow).

[0100] Positioning accuracy: The deviations of the final output peeling area height and circumferential angle from the preset position are ±0.05 meters and ±0.3°, respectively, and the distance deviation from the nearest reference marker is ±0.005 meters.

[0101] like Figure 10 As shown in the radar signal heatmap, the signal intensity of the dense red core area (28 abrupt change points) at a height of 40 meters and a 270° circumference is 65% higher than the background, indicating that the radar is sensitive to the strong reflection characteristics of large-area peeling and can quickly locate the damaged area.

[0102] Analysis of differences in detection procedures for different damage types

[0103] Detection time: Type A (cracks) took 45 seconds from initial detection to complete localization, including 20 seconds for fixed-point imaging and 15 seconds for parameter calculation; Type B (peeling) took 38 seconds, including 18 seconds for fixed-point imaging and 12 seconds for parameter calculation. The difference stems from the more regular contours of the peeling area, allowing for faster algorithm processing.

[0104] Sensor dependence: In crack detection, the signal abrupt change characteristics of millimeter-wave radar are more critical (contributing 60% of the positioning information), while vision is mainly used to confirm length and width; in peeling detection, the contour recognition of binocular vision plays a dominant role (contributing 70% of the positioning information), while radar is mainly used to quickly lock the area.

[0105] Environmental adaptability: Under simulated light rain, the visual recognition rate of crack detection dropped to 85.2% (rainwater blurred the crack edges), but the radar signal was unaffected, and the final positioning accuracy remained at ±0.05 meters; the visual recognition rate of spalling detection dropped to 90.1%, the radar signal strength fluctuation was ≤5%, and the positioning accuracy deviation was ±0.06 meters.

[0106] like Figure 11 As shown, the recognition rate for both types of damage exceeds 98% on sunny days, while the recognition rate for cracks decreases by 13.5% (to 85.2%) and for spalling by 9.1% (to 90.1%) on rainy days. This clearly demonstrates the advantage of the radar anti-interference capability of this application in regional damage detection; Figure 12 As shown, cracks are detected using a "straight-line approximation" path (efficient coverage along the direction), while spalling is detected using a "circular encirclement" path (comprehensive boundary capture). This information demonstrates that the technical solution can dynamically adjust the detection strategy based on the damage morphology (linear / regional), optimizing efficiency while ensuring accuracy. It meets the precise identification requirements for minute damages like cracks and adapts to the comprehensive detection requirements for large-area damages like spalling, showcasing the flexibility and versatility of the technology.

[0107] This embodiment verifies the adaptability of the technical solution to different damage types through detection and location experiments on two types of damage: surface cracks and concrete spalling. Data shows that the identification errors for crack detection (length and width) are ≤0.05 meters and 0.02 mm, respectively; the identification errors for spalling detection (area and longest side) are ≤0.05 square meters and 0.2 meters, respectively; and the location deviations for both types of damage are ≤0.05 meters (height) and ≤0.3° (circumferential), meeting the accuracy requirements for the operation and maintenance of wind power hybrid towers. (Appendix) Figures 4-8 The interspersed displays of damage characteristics, sensor data, positioning results, and process comparisons visually demonstrate the differentiated processing logic for linear damage (cracks) and regional damage (stripping). Cracks rely on the continuity of radar signal abrupt changes and the linear characteristics of the visual contour, while stripping relies on the high-intensity reflection of radar signals and the regional characteristics of the visual contour. Both achieve high-precision positioning through multi-sensor fusion. Experimental results prove that this technical solution can effectively cover typical damage types of wind turbine hybrid towers, providing reliable parameters and coordinate data for subsequent repairs.

[0108] The above are merely preferred embodiments of this method and do not limit the scope of protection of this method. For those skilled in the art, this method can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of this method, without departing from the principles and spirit of this method, through conventional substitutions or to achieve the same function, fall within the scope of protection of this method.

Claims

1. A method for locating surface damage in ultra-high strength concrete wind turbine towers, characterized in that, include: S1. Reference markers are set at predetermined heights and circumferential intervals on the ultra-high strength concrete wind power tower. The reference markers have built-in radio frequency chips with unique codes and have visual recognition patterns containing location information on their outer surfaces. S2. The inspection equipment equipped with millimeter-wave radar, binocular vision camera and radio frequency identification module performs a spiral cruise motion along the tower axis, and simultaneously executes: reads the encoding information of the reference marker through the radio frequency identification module to obtain the initial position coordinates, collects real-time image streams including the reference marker and the tower surface through the binocular vision camera, and transmits frequency-modulated continuous waves and receives reflected signals through the millimeter-wave radar. S3. Perform distortion correction and feature extraction on the real-time image stream, identify the pixel coordinates of the visual recognition pattern in the image, establish the mapping relationship between image pixels and physical space in combination with the actual size of the reference marker, and generate an initial three-dimensional mesh model of the tower surface. S4. Perform time-frequency analysis on the reflected signal of the millimeter-wave radar, extract signal change feature points, and combine them with the spatial coordinates of the initial three-dimensional mesh model to preliminarily locate the physical coordinates of the suspected damage area. S5. Based on the physical coordinates of the suspected damaged area, control the inspection equipment to take fixed-point multi-angle pictures of the current area to obtain damage detail images, extract damage contour features through image segmentation algorithm, and calculate the actual geometric parameters of the damage, including length, width and area, in combination with the mapping relationship described in S3. S6. Import the actual geometric parameters and physical coordinates of the damage into the preset tower cylindrical coordinate system transformation model, and output the height value, circumferential angle value and distance value of the damage on the tower surface, and complete the damage location.

2. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1, characterized in that, In step S1, reference markers are set at preset height circumferential intervals on the ultra-high strength concrete wind turbine tower. Specifically, this means that reference markers are set at preset height intervals along the tower height direction. Select There are several preset height planes, and the vertical height of each preset height plane from the bottom of the tower is [missing information]. ,in ,and Not exceeding the total height of the tower; within each preset height plane, at angular intervals along the circumferential direction of the tower. Evenly distributed A reference marker, the arc length distance between two adjacent reference markers in the circumferential direction. The formula is: ,in , The outer radius of the tower at the current preset height plane; the visual identification pattern of each reference marker includes the height value of its preset height plane. and the initial angle value in the circumferential direction This forms an orderly division and identification of the different height planes and circumferential positions of the tower.

3. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1, characterized in that, In step S2, a real-time image stream containing the reference marker and the tower surface is acquired using a binocular vision camera. Specifically, the inspection equipment operates at a constant linear velocity. The system performs a spiral cruise along the tower axis, with the binocular vision camera operating at a fixed frame rate. Acquire images and obtain the axial displacement difference between adjacent frames. The formula is: The camera's field of view is along the axis. Circumferential direction Determine the range of tower axial height covered by a single frame image. and circumferential angle range The formula is: Circumferential angle range ,in The vertical distance between the camera and the surface of the tower. To collect the tower radius at the specified height, continuous data collection is required. Frame images, constructing a real-time image stream , For the first The frame contains images of the reference markers and the surface of the tower.

4. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1 or 3, characterized in that, In step S3, distortion correction and feature extraction are performed on the real-time image stream to identify the pixel coordinates of the visual recognition pattern in the image. Specifically, this involves pre-correcting the distortion parameters of the binocular camera using the Zhang calibration method and performing single-frame image... Perform distortion correction formula ,in As the distortion offset, image feature points are extracted using the SIFT algorithm, and a feature response threshold is constructed for the visually recognized pattern. The formula is: ,in , The width and height of the pattern in pixels. Filter response values ​​based on pixel grayscale values. The pattern area, Using an empirical threshold, the boundary ellipse of the identified pattern region is fitted using the least squares method to obtain the coordinates of the ellipse's center pixel. The formula is: ,in For boundary pixel coordinates, To fit the ellipse radius, output the pixel coordinates of the pattern in the image coordinate system. .

5. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1, characterized in that, In step S3, the mapping relationship between image pixels and physical space is established based on the actual size of the reference marker, generating an initial three-dimensional mesh model of the tower surface. Specifically, this involves obtaining the actual dimensions of the reference marker: length and width... In the image, the pixel size is Calculate pixel-physical mapping coefficients The formula is: For all reference markers in the image stream, the formula is used. Establish a mapping, where The coordinates of the center pixel of the image. Using physical plane coordinates and combining the real-time pose of the inspection equipment, the image pixels are converted into coordinates in the cylindrical coordinate system of the tower. ,in The axial height, For circumferential angle, The height is the tower radius; by traversing the feature points of the image stream and using the coordinates of the reference marker as constraints, an initial three-dimensional mesh is constructed using the Delaunay triangulation algorithm to generate the initial three-dimensional mesh model of the tower surface.

6. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1, characterized in that, In step S4, time-frequency analysis is performed on the reflected signal of the millimeter-wave radar to extract signal abrupt change feature points. Specifically, this involves acquiring the reflected signal received by the millimeter-wave radar. Preprocessing is performed to remove the DC component, and then the data is processed according to the time window length. Perform frame segmentation to obtain frame signals The formula is: ,in For the Hanning window function, Using the frame number as the frame number, perform a short-time Fourier transform on each frame signal to obtain the time spectrum. The formula is: ,in For frequency, For time, calculate the power spectral density of the time spectrum. Through power spectral density Calculate the signal abrupt change characteristic quantity The formula is: Set mutation threshold , For empirical coefficients, when When a signal abrupt change is detected at the corresponding position in the current frame, the center time of the abruptly changed frame is extracted. and the corresponding propagation distance of the radar transmitted wave The formula is: ,in For the speed of light, As a characteristic point of signal abrupt change.

7. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1 or 6, characterized in that, In step S4, the spatial coordinates of the initial three-dimensional mesh model are combined to initially locate the physical coordinates of the suspected damaged area. Specifically, this involves identifying the signal abrupt change feature points of the millimeter-wave radar. Convert to three-dimensional coordinates in radar coordinate system ,in , The radar elevation angle, This represents the radar azimuth angle; obtained through a coordinate transformation matrix. Convert the radar coordinate system coordinates to the tower's global coordinate system coordinates. Traverse the node set of the initial 3D mesh model Calculate the Euclidean distance between the global coordinates and the mesh nodes. ,filter The nodes constitute the candidate region, among which The distance threshold is the geometric center of the candidate region. As the physical coordinates of the suspected damaged area, This represents the number of candidate nodes.

8. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1, characterized in that, In step S5, based on the physical coordinates of the suspected damage area, the inspection equipment performs fixed-point multi-angle photography of the current area to obtain detailed damage images, and extracts damage contour features using an image segmentation algorithm. Specifically, the inspection equipment moves to the physical coordinates of the suspected damage area. For the corresponding orbital trajectory, damage detail images are captured at azimuth and elevation angle intervals, centered on the coordinates. An improved U-Net segmentation algorithm is applied to the acquired detail images, incorporating an attention-based loss function. Training the model, where For cross-entropy loss, To compare the losses, , For the corresponding weight coefficients, output the binarized mask of the damaged region. ,in =1 indicates a damaged pixel. =0 indicates background pixels; Damage contours are extracted based on masking, and the contour pixel sequence is obtained through a chain code algorithm. Calculate the minimum bounding rectangle of the contour, with pixel width and height respectively. Through formula Calculate the circularity of the contour and obtain the feature parameters of the damaged contour, where The perimeter of the outline in pixels. The area of ​​the damaged pixel.

9. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1, 5, or 8, characterized in that, In step S5, the actual geometric parameters of the damage are calculated by combining the mapping relationship described in S3. Specifically, this is done through the pixel-physical space mapping coefficient. The pixel parameters of the damaged contour are converted into actual physical parameters, where the actual width of the damage is... Actual height The formula for the actual area of ​​damage is: For linear damage, extract the pixel coordinates of the two points furthest apart in the contour sequence. and The actual length is Combined with the normal vector of the damaged region in the initial 3D mesh model Through formula Calculate the angle between the damage length direction and the normal vector of the tower surface, where Given the damage length direction vector, the output includes the actual geometric parameters of the damage, including length, width, area, and angle.

10. The method for locating surface damage of ultra-high strength concrete wind turbine towers according to claim 1, characterized in that, In step S6, the actual geometric parameters and physical coordinates of the damage are imported into a preset tower cylindrical coordinate system transformation model, and the height value, circumferential angle value, and distance value relative to the nearest reference marker of the damage on the tower surface are output. Specifically, the physical coordinates of the damage are... Import the tower cylindrical coordinate system transformation model and calculate the height of the damage on the tower surface. Calculate the circumferential angle value The formula is: From all benchmark markers, select the marker with the smallest difference in height from the damage, satisfying the requirement... A collection of identification components, among which Using the height of the reference marker as an example, calculate the spatial straight-line distance between the damage and each marker from the marker set, using the following formula: , The minimum value is taken as the physical coordinate of the reference marker. The output includes the distance value relative to the nearest reference marker. , and Damage localization.

Images