A blade damage inspection map construction method and a blade damage inspection method

Abstract

The present invention relates to the technical field of wind turbine blade damage detection, and provides a method for constructing a blade damage inspection map and a blade damage inspection method. The method for constructing a blade damage inspection map includes the following steps: obtaining a three-dimensional image of the depth damage of the blade to be inspected; obtaining a three-dimensional image of the surface damage of the blade to be inspected; first fusing the three-dimensional image of the depth damage and the three-dimensional image of the surface damage to obtain a damage fusion image, and then extracting all the damage areas in the damage fusion image; or first extracting the deep damage area in the three-dimensional image of the depth damage, extracting the surface damage area in the three-dimensional image of the surface damage, and then fusing the three-dimensional image of the depth damage and the three-dimensional image of the surface damage to obtain a damage fusion image with all the damage areas. The present invention can realize full inspection of depth and surface damage, and improves inspection efficiency, reliability, accuracy and safety.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine blade damage detection, and in particular to a blade damage inspection map construction method and a blade damage inspection method. Background Art

[0002] Due to the limitation of wind resources and environment, wind farms are mainly located in suburbs, deserts, Gobi Desert, grasslands and offshore areas away from cities. Although the wind energy reserve has obvious advantages, it is also faced with adverse effects of harsh environment. The blades of wind turbines are subjected to erosion by strong winds, strong sand, salt spray, frost, rain and snow for a long time, which can easily cause damage to the surface of the blades. At the same time, the blades are subjected to unstable loads for a long time, which is prone to deep structural fatigue damage. For example, lightning strikes are one of the main causes of damage to wind turbine blades. It will produce destructive internal shock waves inside the blades, causing blade delamination, degumming and blade tip shedding. For another example, the icing of blades will cause its rotation to be unbalanced, which will impose greater harmful loads on the hub. Blades with surface damage are prone to cracks, cavities and other phenomena on the surface, and blades with deep damage are prone to crack extension inside, which seriously affects the safe use of wind turbine blades. Blade damage will also cause wind turbines to operate at reduced power, resulting in reduced power generation and shortened effective use time. Therefore, if blade damage is not discovered in time, it will greatly cause blade damage to intensify, tower collapse and even other dangerous accidents.

[0003] Regular inspections of blades for damage have gradually become the main means of timely detecting blade damage. Traditional blade maintenance mainly relies on manpower. Specifically, during regular inspections, inspectors use high-power telescope observations and visual inspections in high-altitude hanging baskets to check whether the blades are damaged. These methods have many disadvantages: they need to rely on experience and judgment, which is highly subjective; they need to stop the machine for inspection, resulting in large power generation losses; and they work at high altitudes, which affects personal safety. For this reason, the relevant technology has improved the traditional manual inspection method and used drones to inspect blades for damage.

[0004] However, when drones are used for blade damage inspection in related technologies, operators need to rely on their experience to control the inspection route of the drones each time, which can easily lead to missed inspections or repeated inspections, resulting in low inspection efficiency. In addition, when drones are used for damage inspection in related technologies, they usually only focus on the surface damage locations of the blades, ignoring the focus on the deep damage locations, resulting in low detection reliability, low detection accuracy, and poor inspection results. Therefore, how to construct a damage inspection map that includes deep damage areas and surface damage areas to improve the efficiency, accuracy, and reliability of damage inspections has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to provide a blade damage inspection map construction method and a blade damage inspection method to solve the problems of low efficiency, poor accuracy and poor reliability of the blade damage inspection method in the prior art.

[0006] In order to achieve the above object, according to a first aspect of the present invention, there is provided a

[0007] The method for constructing a blade damage inspection map comprises the following steps: obtaining a deep damage three-dimensional image of a blade to be inspected; obtaining a surface damage three-dimensional image of the blade to be inspected; firstly fusing the deep damage three-dimensional image and the surface damage three-dimensional image to obtain a damage fusion image, and then extracting all damage areas in the damage fusion image; or firstly extracting the deep damage area in the deep damage three-dimensional image, extracting the surface damage area in the surface damage three-dimensional image, and then fusing the deep damage three-dimensional image and the surface damage three-dimensional image to obtain a damage fusion image with all damage areas.

[0008] Furthermore, obtaining the deep damage image of the blade to be inspected specifically includes the following steps: establishing a three-dimensional simulation analysis model of the wind turbine blade to be inspected; performing stress analysis on the three-dimensional simulation model to obtain a three-dimensional stress cloud map; using the three-dimensional stress cloud map as a three-dimensional image of deep damage; obtaining the surface damage image of the blade to be inspected specifically includes the following steps: using a drone to photograph the blade to be inspected to obtain a three-dimensional image of the surface damage of the blade to be inspected.

[0009] Furthermore, using a drone to photograph the blade to be inspected to obtain a three-dimensional image of the surface damage of the blade to be inspected specifically includes the following steps: using a drone to photograph the leading edge, pressure side, trailing edge, and suction side of the blade at four angles to obtain four images to form an image sequence; or using a drone to photograph the blade along a preset route that passes through the leading edge, pressure side, trailing edge, and suction side in sequence to obtain multiple images to form an image sequence; using a multi-view Figure 3 The three-dimensional reconstruction algorithm is used for image stitching, SFIT feature extraction and matching are performed on the image sequence, and feature extraction and matching are performed on the points, edges and contours in the image sequence; the three-dimensional space point coordinate cloud is calculated using linear triangulation, and a sparse point cloud is reconstructed; the bundle method is used for optimization; the dense space directed point cloud or patch is obtained through the classic dense algorithm PMVS; the erroneous point cloud or patch data is filtered out to obtain a complete macroscopic three-dimensional image of the blade as a three-dimensional image of surface damage.

[0010] Furthermore, the depth damage three-dimensional image and the surface damage three-dimensional image are fused to obtain a damage fusion image, which specifically includes the following steps: binarizing the depth damage three-dimensional image to obtain a depth damage binary image; binarizing the surface damage three-dimensional image to obtain a surface damage binary image; using the depth damage binary image as a matrix, mapping the surface damage binary image to the depth damage binary image in proportion to form a damage fusion image.

[0011] Furthermore, binarization of the three-dimensional image of depth damage is performed to obtain a binary image of depth damage, which specifically includes the following steps: extracting each pixel point in the three-dimensional image of depth damage; obtaining the color component of each pixel point; comparing the threshold range of the color component of each pixel point with the first set threshold range, and modifying one of the pixel points within the first set threshold range and the pixel points outside the first set threshold range to white and the other to black; binarization of the three-dimensional image of surface damage is performed to obtain a binary image of surface damage, which specifically includes the following steps: obtaining a grayscale image of the three-dimensional image of surface damage; extracting each pixel point in the grayscale image; obtaining the grayscale value of each pixel point; and modifying the pixel points whose grayscale values ​​are greater than the second set threshold range to white. Pixel points and pixel points whose grayscale values ​​are less than the second set threshold range are divided into two groups, one group is modified to black, and the other group is modified to white; wherein the second set threshold is determined by combining the custom threshold method and the iterative threshold method, including the following steps: S11, select an initial threshold G, G = (Gmin + Gmax) / 2, where Gmin is the minimum grayscale value of the grayscale image, and Gmax is the maximum grayscale value of the grayscale image; S12, divide the grayscale image into two groups according to the initial threshold G, denoted as P1 and P2; P1 is composed of all pixels with grayscale values ​​greater than G, and P2 is composed of all pixels with grayscale values ​​less than or equal to G; S13, calculate the mean values ​​μ1 and μ2 of P1 and P2; calculate a new threshold G1, and G1 = (μ1 + μ2) / 2; S14, if G1 no longer changes, G1 is the second set threshold; otherwise, repeat S2 and S3, recalculate the new threshold G1, until the difference of the new threshold G1 in the continuous iteration is less than the predetermined value, and use the new threshold G1 obtained in the latest iteration as the second set threshold; using the depth damage binary image as the matrix, mapping the surface damage binary image to the depth damage binary image in proportion to form a damage fusion image, which specifically includes the following steps: before applying stress to the three-dimensional simulation analysis model for stress analysis, obtain the original three-dimensional stress cloud map of the three-dimensional simulation analysis model, and record the original coordinates of each node in the original three-dimensional stress cloud map; obtain the leading edge in the depth damage binary image through the original coordinates of each node in the original three-dimensional stress cloud map The coordinates of the leading edge endpoint and the trailing edge endpoint in the depth damage binary image are used to calculate the first chord length of the blade in the depth damage binary image; the coordinates of the leading edge endpoint and the trailing edge endpoint in the surface damage binary image are obtained; the second chord length of the blade in the surface damage binary image is calculated using the coordinates of the leading edge endpoint and the trailing edge endpoint in the surface damage binary image; the depth damage binary image is divided into N first small segments according to the first chord length to form 2N first mapping points; the surface damage binary image is divided into N second small segments according to the second chord length to form 2N second mapping points; the 2N second mapping points are respectively matched one by one with the 2N first mapping points, and the surface damage binary image is fused with the depth damage binary image to obtain a damage fused image.

[0012] Furthermore, extracting all the damaged areas in the damaged fusion image specifically includes the following steps: extracting the boundary of the damaged area based on the alphashapes algorithm, in a spatial discrete point set A, there are m discrete points in total, and the m points can be connected into m*(m-1) / 2 line segments; for all discrete points, draw a circle with alpha as the radius, and the circle must pass through points Q1 and Q2. If there are no other discrete points in the circle, Q1 and Q2 must be on the boundary, and the line segment Q1-Q2 belongs to the boundary line segment; the specific calculation steps are as follows: S21, traverse all boundary line segments Q1-Q2; S22, if the length of Q1-Q2 is greater than alpha which is twice the diameter, skip it; S23, according to the geometric principle, obtain the centers R1 and R2 of the two circles; S24, if one of the two circles does not contain any other points in the point set A, then the line segment Q1-Q2 is the boundary; use multiple smooth straight lines to connect the scattered boundaries to form the contour line of the damaged area, that is, obtain multiple irregular polygons, each polygon corresponds to a damaged area.

[0013] Furthermore, extracting the deep damage area in the deep damage three-dimensional image specifically includes the following steps: screening and marking the node coordinates whose stress is greater than the allowable stress of the blade material from the three-dimensional stress cloud map, and the set of node coordinates forms the deep damage area; extracting the surface damage area in the surface damage three-dimensional image specifically includes the following steps: marking the surface damage area and its three-dimensional coordinates in the surface damage three-dimensional image through image recognition.

[0014] Furthermore, during image recognition, damage features in the image are extracted, including cracks, sand holes, pits, and peeling; the damage features are pre-input into the deep learning algorithm model for feature recognition training, and the trained deep learning algorithm model is used for image recognition; before image stitching and / or image recognition, the image information is also pre-processed, specifically including the following steps: histogram equalization processing and / or median filtering to eliminate part of the noise in the image; when using a drone to photograph the blades to be inspected, the flight speed of the drone is controlled to be less than or equal to 4m / s; the distance between the drone and the blade is controlled to be less than or equal to 0.5m; when controlling the drone to fly along a preset route, the position information of the drone is recorded, and according to the position information of the wind turbine blade, the blade area information where the drone is located when each image is taken is determined, and the stored image and position information are transmitted to the staff operating the drone for image processing and analysis in the background to obtain a three-dimensional image of surface damage.

[0015] Furthermore, the blade damage inspection map construction method also includes the following steps: defining the damage area as the area to be detected in the damage inspection map; defining the area outside the damage area as the walking area in the damage inspection map; the area to be detected and the walking area together form the damage inspection map.

[0016] According to another aspect of the present invention, a blade damage inspection method is provided, comprising the following steps: planning an inspection route of an unmanned aerial vehicle (UAV) according to the coordinates of an area to be inspected and a walking area in a blade damage inspection map, so that the UAV takes the shortest time to pass through all the areas to be inspected; wherein the blade damage inspection map is constructed according to the above-mentioned blade damage inspection map construction method; controlling the UAV to fly along the inspection route; and when the UAV passes through the area to be inspected, controlling the UAV to take pictures of the blades to be inspected to obtain inspection pictures.

[0017] By applying the blade damage inspection map construction method provided by the present invention, a damage fusion image containing deep damage areas and surface damage areas can be obtained. The blade damage inspection map formed by the damage fusion image is used to guide the flight of the drone, so that the drone can inspect all damage areas in the damage fusion image.

[0018] By applying the blade damage inspection method provided by the present invention, the inspection route of the UAV is rationally planned using the blade damage inspection map, so that the UAV takes the shortest time to pass through all the areas to be inspected, thereby being able to efficiently, accurately and reliably inspect both deep damage areas and surface damage areas.

[0019] Based on the implementations provided in the above aspects, this application can also be further combined to provide more implementations. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 A flowchart of three-dimensional modeling of a blade in an optional embodiment of the method for constructing a blade damage inspection map provided by the present invention is shown;

[0022] Figure 2 A flowchart of obtaining a three-dimensional image of surface damage of a blade in an optional embodiment of the method for constructing a blade damage inspection map provided by the present invention is shown;

[0023] Figure 3 A cross-sectional view of a blade is shown;

[0024] Figure 4 A flow chart showing the binarization processing of the three-dimensional image of surface damage in an optional embodiment of the method for constructing a blade damage inspection map provided by the present invention is shown;

[0025] Figure 5 A mapping schematic diagram of fusing a three-dimensional image of deep damage with a three-dimensional image of surface damage in an optional embodiment of the method for constructing a blade damage inspection map provided by the present invention is shown. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0027] In an optional embodiment of the present invention, a method for constructing a blade damage inspection map is provided, comprising the following steps: obtaining a three-dimensional image of deep damage of the blade to be inspected; obtaining a three-dimensional image of surface damage of the blade to be inspected; firstly fusing the three-dimensional image of deep damage and the three-dimensional image of surface damage to obtain a fused damage image, and then extracting all the damaged areas in the fused damage image. In this way, the obtained fused damage image contains both the deep damage area and the surface damage area, and can reliably, quickly and efficiently guide the drone to inspect all the deep damage areas and surface damage areas, avoiding missed inspections and repeated inspections.

[0028] In the technical solution of first extracting the deep damage area in the deep damage three-dimensional image and then fusing the deep damage three-dimensional image with the surface damage three-dimensional image, two methods can be used to first extract the deep damage area in the deep damage three-dimensional image. The first method is to extract the deep damage area through the threshold range of the color component, and the second method is to extract the deep damage area through the threshold range of the stress value.

[0029] In another optional embodiment of the present invention, a method for constructing a blade damage inspection map is provided, comprising the following steps: obtaining a three-dimensional image of deep damage of the blade to be inspected; obtaining a three-dimensional image of surface damage of the blade to be inspected; first extracting the deep damage area in the three-dimensional image of deep damage, extracting the surface damage area in the three-dimensional image of surface damage, and then fusing the three-dimensional image of deep damage and the three-dimensional image of surface damage to obtain a damage fusion image with all damage areas. In this way, the obtained damage fusion image contains both the deep damage area and the surface damage area, and can reliably, quickly and efficiently guide the drone to inspect all the deep damage areas and surface damage areas, avoiding missed inspections and repeated inspections.

[0030] In the technical solution of first fusing the deep damage three-dimensional image and the surface damage three-dimensional image and then extracting all the damaged areas in the damaged fused image, the deep damage three-dimensional image and the surface damage three-dimensional image are first binarized; after the binarization, the deep damage area and the surface damage area are both black and the other undamaged areas are all white, or the deep damage area and the surface damage area are both white and the other undamaged areas are all black; then, boundary extraction is performed on the damaged area, that is, boundary extraction is performed on the black area or the white area, and the multiple polygons obtained represent all the damaged areas.

[0031] Optionally, obtaining the deep damage image of the blade to be inspected specifically includes the following steps: establishing a three-dimensional simulation analysis model of the wind turbine blade to be inspected; performing stress analysis on the three-dimensional simulation model to obtain a three-dimensional stress cloud map; and using the three-dimensional stress cloud map as a three-dimensional image of deep damage. In this way, the location of the deep damage area of ​​the blade can be obtained through stress analysis and reflected in the three-dimensional image of deep damage, which is conducive to subsequent processing of the deep damage area and guides the drone to inspect the deep damage area.

[0032] Optionally, obtaining the surface damage image of the blade to be inspected specifically includes the following steps: using a drone to photograph the blade to be inspected to obtain a three-dimensional image of the surface damage of the blade to be inspected. Since each wind turbine blade is in a different environment, the surface damage caused will also be different. The technical solution provided by the present invention can construct a more targeted blade damage inspection map for each blade to be inspected, thereby further improving the inspection efficiency and inspection effect.

[0033] After a new fan blade has been used for a period of time, some damage will appear on its surface, and the blade needs to be inspected. Before the inspection, a corresponding three-dimensional simulation model is established for the type of blade and a stress analysis is performed to obtain a three-dimensional image of the depth damage of the blade; a three-dimensional image of the surface damage of the blade is obtained by photographing the blade to be inspected using a drone; a blade damage inspection map for the blade can be obtained through the three-dimensional image of the depth damage and the three-dimensional image of the surface damage. After the blade damage inspection map is constructed, the drone can be guided to conduct regular inspections on the blade to improve the inspection efficiency and inspection results. After the fan blade continues to be used for a period of time, new surface damage may occur. It is only necessary to use the blade damage inspection map construction method provided by the present invention to regularly update the blade damage inspection map.

[0034] Optionally, using a drone to photograph the blade to be inspected to obtain a three-dimensional image of the surface damage of the blade to be inspected specifically includes the following steps: using a drone to photograph the leading edge, pressure side, trailing edge, and suction side of the blade at four angles to obtain four images to form an image sequence; or using a drone to photograph the blade along a preset route that passes through the leading edge, pressure side, trailing edge, and suction side in sequence to obtain multiple images to form an image sequence; using a multi-view Figure 3 3D reconstruction algorithm is used for image stitching, SFIT feature extraction and matching is performed on the image sequence, and feature extraction and matching is performed on the points, edges, and contours in the image sequence; linear triangulation is used to obtain the three-dimensional space point coordinate cloud, and a sparse point cloud is reconstructed; bundle adjustment is used for optimization; dense space directed point clouds or patches are obtained through the classic dense algorithm PMVS; erroneous point cloud or patch data is filtered out to obtain a complete macroscopic three-dimensional image of the blade as a three-dimensional image of surface damage. Since some blades are large in size or the shooting range of the camera is limited, a complete macroscopic three-dimensional image of the blade cannot be obtained by only shooting once. The preferred solution provided by the present invention can shoot the blade multiple times, and then use multi-view Figure 3 The three-dimensional reconstruction algorithm processes multiple images to obtain a complete macroscopic three-dimensional image of the leaf.

[0035] Optionally, the color type of the three-dimensional stress cloud map is set to "Rainbow". In the three-dimensional stress cloud map, the nodes that tend to red represent larger stress values, and the nodes that tend to blue represent smaller stress values, that is, the areas prone to deep damage range from red to blue.

[0036] Optionally, fusing the deep damage 3D image and the surface damage 3D image to obtain the damage fusion image specifically includes the following steps: binarizing the deep damage 3D image to obtain the deep damage binary image; binarizing the surface damage 3D image to obtain the surface damage binary image; using the deep damage binary image as a matrix, mapping the surface damage binary image to the deep damage binary image in proportion to form the damage fusion image. Fusion after binarization can make the processing simpler, reduce the amount of data processing and compression, and increase efficiency.

[0037] Optionally, binarization of the deep damage three-dimensional image to obtain the deep damage binary image specifically includes the following steps: extracting each pixel in the deep damage three-dimensional image; obtaining the color component of each pixel; comparing the threshold range of the color component of each pixel with the first set threshold range, and modifying one of the pixels within the first set threshold range and the pixel outside the first set threshold range to white and the other to black. In this way, the deep damage three-dimensional image is binarized using the color component, and the division effect of the deep damage area and the non-deep damage area is better and the division efficiency is higher.

[0038] In a specific optional embodiment of the binarization processing, the range corresponding to red and yellow is confirmed as the deep damage area, that is, the threshold range of the color components of the red area and the yellow area is set to the first set threshold range, so that the node colors corresponding to the red area and the yellow area are changed to white, and the nodes corresponding to other colors are changed to black, and the white area is the deep damage area.

[0039] In another specific optional embodiment of the binarization process, nodes with an average value of color components RGB greater than 128 are changed to white, and nodes with an average value less than 128 are changed to black, and the white area is a deep damage area.

[0040] In another specific optional embodiment, through the finite element analysis software ABAQUS "Tools" → "Query" → "Query Value", the coordinates of each node in the three-dimensional stress cloud map and the corresponding Excel data file of the stress value can be obtained, and then the stress value is screened, and the nodes greater than the allowable stress of the blade material are screened out as target points. The area occupied by the set of all target points is the area where the stress is more concentrated, representing the deep damage area.

[0041] In other optional embodiments not disclosed in the present invention, the threshold range for dividing the deep damage area is also changed according to actual needs.

[0042] In other optional embodiments not disclosed in the present invention, other methods are selected according to actual needs, as long as the deep damage area and the non-deep damage area can be divided.

[0043] Optionally, binarization of the three-dimensional image of surface damage is performed to obtain the binary image of surface damage, specifically including the following steps: obtaining a grayscale image of the three-dimensional image of surface damage; extracting each pixel in the grayscale image; obtaining the grayscale value of each pixel; dividing the pixel whose grayscale value is greater than the second set threshold range and the pixel whose grayscale value is less than the second set threshold range into two groups, one group is modified to black, and the other group is modified to white. In this way, the three-dimensional image of surface damage is binarized using the threshold segmentation technology, and the division effect of the surface damage area and the non-surface damage area is better, and the division efficiency is higher.

[0044] The second set threshold is determined by combining a custom threshold method and an iterative threshold method, including the following steps:

[0045] S11, select an initial threshold G, G = (Gmin + Gmax) / 2, where Gmin is the minimum grayscale value of the grayscale image, and Gmax is the maximum grayscale value of the grayscale image;

[0046] S12, according to the initial threshold G, the grayscale image is divided into two groups, denoted as P1 and P2; P1 is composed of all pixels with grayscale values ​​greater than G, and P2 is composed of all pixels with grayscale values ​​less than or equal to G;

[0047] S13, calculate the mean values ​​μ1 and μ2 of P1 and P2; calculate a new threshold G1, and G1=(μ1+μ2) / 2;

[0048] S14, if G1 does not change, then G1 is the second set threshold; otherwise, repeat S2 and S3, recalculate the new threshold G1, until the difference of the new threshold G1 in the continuous iteration is less than the predetermined value, and the new threshold G1 obtained by the latest iteration is used as the second set threshold. In this way, the image obtained by combining the custom threshold method and the iterative threshold method presents a more obvious black and white effect, which is conducive to subsequent further processing.

[0049] Optionally, the surface damage binary image is mapped proportionally to the depth damage binary image with the depth damage binary image as the matrix to form a damage fusion image, which specifically includes the following steps: before applying stress to the three-dimensional simulation analysis model for stress analysis, the original three-dimensional stress cloud map of the three-dimensional simulation analysis model is obtained, and the original coordinates of each node in the original three-dimensional stress cloud map are recorded; the coordinates of the leading edge endpoint and the trailing edge endpoint in the depth damage binary image are obtained through the original coordinates of each node in the original three-dimensional stress cloud map; the coordinates of the leading edge endpoint and the trailing edge endpoint in the depth damage binary image are used to calculate the blade in the depth damage binary image. The first chord length of the blade; obtain the coordinates of the leading edge endpoint and the trailing edge endpoint in the surface damage binary image; use the coordinates of the leading edge endpoint and the trailing edge endpoint in the surface damage binary image to calculate the second chord length of the blade in the surface damage binary image; divide the depth damage binary image into N first small segments according to the first chord length to form 2N first mapping points; divide the surface damage binary image into N second small segments according to the second chord length to form 2N second mapping points; correspond the 2N second mapping points to the 2N first mapping points one by one, fuse the surface damage binary image with the depth damage binary image, and obtain a damage fusion image. Considering that the images taken by the drone have certain distortion, resulting in the size of the surface damage binary image not matching the actual size, the surface damage binary image can be mapped to the depth damage binary image to solve this technical problem.

[0050] Optionally, extracting all the damaged areas in the damaged fusion image specifically includes the following steps: extracting the boundary of the damaged area based on the alphashapes algorithm, in a spatial discrete point set A, there are m discrete points in total, and the m points can be connected into m*(m-1) / 2 line segments; for all discrete points, draw a circle with alpha as the radius, and the circle must pass through points Q1 and Q2. If there are no other discrete points in the circle, Q1 and Q2 must be on the boundary, and the line segment Q1-Q2 belongs to the boundary line segment. The specific calculation steps are as follows: S21, traverse all boundary line segments Q1-Q2; S22, if the length of Q1-Q2 is greater than alpha, which is twice the diameter, skip it; S23, according to the geometric principle, find the centers R1 and R2 of the two circles; S24, if one of the two circles does not contain any other points in the point set A, then the line segment Q1-Q2 is the boundary; use multiple smooth straight lines to connect the scattered boundaries to form the contour line of the damaged area, that is, obtain multiple irregular polygons, each polygon corresponding to a damaged area. In this way, the entire damaged area can be obtained efficiently and accurately.

[0051] Optionally, in the technical solution of first extracting the deep damage area in the deep damage three-dimensional image and then fusing the deep damage three-dimensional image with the surface damage three-dimensional image, extracting the deep damage area in the deep damage three-dimensional image specifically includes the following steps: screening and marking the node coordinates whose stress is greater than the allowable stress of the blade material from the three-dimensional stress cloud map, and the collection of node coordinates forms the deep damage area.

[0052] Optionally, in the technical solution of first extracting the deep damage area in the deep damage three-dimensional image and then fusing the deep damage three-dimensional image with the surface damage three-dimensional image, extracting the surface damage area in the surface damage three-dimensional image specifically includes the following steps: marking the surface damage area and its three-dimensional coordinates in the surface damage three-dimensional image through image recognition.

[0053] In optional embodiments not provided in the present invention, in the technical solution of first extracting the deep damage area in the deep damage three-dimensional image and then fusing the deep damage three-dimensional image with the surface damage three-dimensional image, other methods may also be used to extract the deep damage area and the surface damage area, as long as the goal of extracting the damage area can be achieved.

[0054] Optionally, during image recognition, damage features including cracks, sand holes, pits, and spalling are extracted from the image; the damage features are pre-input into the deep learning algorithm model for feature recognition training, and the trained deep learning algorithm model is used for image recognition. In this way, the surface damage area can be quickly identified and extracted.

[0055] Optionally, taking the deep damage three-dimensional image with deep damage areas as the base, the surface damage three-dimensional image with surface damage areas is proportionally mapped to the deep damage three-dimensional image with deep damage areas to obtain a damage fusion image with all damage areas.

[0056] Optionally, before performing image stitching and / or image recognition, the image information is preprocessed, specifically including the following steps: performing histogram equalization processing and / or median filtering to eliminate some noise in the image. In this way, after preprocessing, it is more conducive to improving the accuracy and efficiency of subsequent image stitching or image recognition.

[0057] Optionally, when using a drone to photograph the blade to be inspected, the flight speed of the drone is controlled to be less than or equal to 4 m / s, so as to avoid unclear images due to excessively high flight speed of the drone.

[0058] Optionally, the distance between the drone and the blade is controlled to be less than or equal to 0.5 m. In this way, it is possible to avoid the drone and the blade being too far away from each other, resulting in unclear shooting images, and to avoid the drone and the blade being too close, resulting in the drone being easily collided with the blade when affected by external factors such as wind.

[0059] Optionally, when controlling the drone to fly along a preset route, the drone's location information is recorded, and according to the location information of the wind turbine blades, the blade region information where the drone was located when taking each image is determined, and the stored image and location information are transmitted to the staff operating the drone for image processing and analysis in the background to obtain a three-dimensional image of the surface damage. In this way, the drone's location information can be used to splice the images, resulting in better splicing effects and higher splicing efficiency.

[0060] Optionally, the blade damage inspection map construction method further includes the following steps: defining the damage area as the area to be detected in the damage inspection map; defining the area outside the damage area as the walking area in the damage inspection map; the area to be detected and the walking area together form the damage inspection map. In this way, the damage inspection map constructed by the blade damage inspection map construction method provided by the present invention can guide the drone to perform key inspections at the position corresponding to the damage area and quickly pass through the non-damaged area, thereby improving the inspection efficiency of the drone.

[0061] According to another aspect of the present invention, a blade damage inspection method is provided, comprising the following steps: planning the inspection route of the drone according to the coordinates of the area to be inspected and the coordinates of the walking area in the blade damage inspection map, so that the drone takes the shortest time to pass through all the areas to be inspected; wherein the blade damage inspection map is constructed according to the above or below blade damage inspection map construction method; controlling the drone to fly along the inspection route; when the drone passes through the area to be inspected, controlling the drone to take pictures of the blades to be inspected to obtain inspection pictures. In this way, the inspection route of the drone is reasonably planned using the blade damage inspection map, so that the drone takes the shortest time to pass through all the areas to be inspected, thereby being able to efficiently, accurately and reliably inspect both the deep damage area and the surface damage area.

[0062] Optionally, the blade damage inspection method further includes the following steps: regularly updating the blade damage inspection map according to the above or below blade damage inspection map construction method; replanning the inspection route of the drone using the updated blade damage inspection map to update the inspection route; after the inspection route is updated, controlling the drone to fly along the updated inspection route. In this way, after the initial establishment of the damage inspection map for the blades to be inspected, new damage areas may appear. This problem can be solved by regularly updating the blade damage inspection map to avoid missed inspections.

[0063] Optionally, the blade damage inspection method further comprises the following steps: judging whether the blade needs maintenance according to the inspection image, so that the blade can be repaired in time.

[0064] The blade damage inspection map construction method provided by the present invention can perform dual detection of deep damage and surface damage to blades, thereby facilitating efficient repair of wind turbine blade damage. Deep damage detection is performed on vulnerable parts of the wind turbine based on the three-dimensional stress cloud map, and the damaged area is modeled in the form of three-dimensional coordinates; the blade surface damage images taken by the drone are spliced, and then the deep damage images and surface damage images are rasterized and fused to obtain a blade damage inspection map for unmanned systems. The inspection unmanned system performs inspections according to the above inspection map. This method can greatly improve the efficiency of blade inspections and achieve the effect of full inspection of deep damage and surface damage.

[0065] The present invention is aimed at the field of wind turbine blade inspection and is used to solve the problem of unmanned system navigation map construction in the process of blade damage inspection of large-scale onshore and offshore wind turbines. It is an unmanned system inspection map construction method suitable for simultaneous detection of deep damage and surface damage of wind turbine blades, providing support for high-efficiency path planning for comprehensive inspection and diagnosis of blade damage. It solves the technical problem of the related art that inspection of both deep damage and surface damage is not achieved.

[0066] In a specific embodiment of the present invention, a method for constructing a blade damage inspection map is provided, comprising the following specific steps: S1. Considering the environmental conditions, unit geometric parameters, and unit control parameters faced by the wind turbine, a finite element simulation analysis model for the wind turbine blade is established to obtain a three-dimensional stress cloud map of the blade. The node coordinates whose stress is greater than the allowable stress of the blade material are screened and marked from the three-dimensional stress cloud map, and their RGB values ​​and three-dimensional coordinates are marked on the three-dimensional stress cloud map image; S2. The high-definition camera in the drone equipment is used to collect information on the macroscopic image of the blade, and then the image is spliced ​​based on a feature matching algorithm to obtain a three-dimensional image of the blade. The location of the damage on the blade surface and its three-dimensional coordinates are marked through image recognition;

[0067] S3, binarize the three-dimensional images of the wind turbine blades in S1 and S2 respectively, and then perform image mapping fusion processing on the two images, and finally obtain a three-dimensional image of the wind turbine blade with both depth and surface damage; S4, connect the boundary points scattered in the white area with multiple smooth straight lines to form the contour line of the damaged area, that is, to obtain multiple irregular polygons. Optionally, in S2, when performing image stitching based on the feature matching algorithm, apply multi-view Figure 3dimensional reconstruction algorithm for image stitching. Optionally, in S2, during image recognition, damage features in the stitched image are extracted, such as cracks, sand holes, pits, and peeling; the damage features are input into the deep learning algorithm model for feature recognition. Optionally, in S4, the boundary of the damaged area is extracted based on the alpha shapes algorithm. Optionally, in S2, before image stitching and / or image recognition, the image information is also preprocessed, including: histogram equalization processing and / or median filtering to eliminate part of the noise in the image.

[0068] In a specific embodiment of the present invention, SolidWorks is used to perform three-dimensional modeling of the fan blade, and the modeling is mainly completed by using the coordinate import method. The modeling process is as follows: Figure 1 shown.

[0069] S1. Considering the environmental conditions, geometric parameters and control parameters of the wind turbine, a finite element simulation analysis model of the wind turbine blade is established to obtain the three-dimensional stress cloud map of the blade. The coordinates of the nodes whose stress is greater than the allowable stress of the blade material are selected and marked from the three-dimensional stress cloud map, and their RGB values ​​and three-dimensional coordinates are marked on the three-dimensional stress cloud map image;

[0070] (1) Blade coordinate transformation

[0071] Firstly, the two-dimensional coordinates of the fan airfoil are converted into three-dimensional coordinates by combining the optimization algorithm with the three-dimensional coordinate conversion method. The overall coordinate data is stored in a .txt format file. In SolidWorks, the command "Generate Curves through XYZ Points" in the curve modeling menu is called to import the stored txt coordinate file. A fan airfoil cross-section can be drawn on the sketch. By repeating this operation, the fan contour curves of each cross-section are imported to generate multiple sketches along the airfoil expansion direction. Finally, the sketch is stretched section by section through the lofting operation to obtain the overall three-dimensional model of the fan blade.

[0072] The coordinates of the points that make up the figure are transformed by mathematical methods. Through the homogeneous matrix coordinate transformation method in matrix theory research, the relationship between the coordinate changes of the points in the figure can be expressed as:

[0073] [x1 y1 z1 1] T =[x0 y0 z o 1] T (1)

[0074] Where: [x0 y0 z o 1] T is the position matrix of the point in the figure that has not been translated or rotated; [x1 y1 z1 1] T The position of the point after coordinate transformation.

[0075] T is a 4th-order transformation matrix, expressed as:

[0076]

[0077] In the matrix T, the upper and lower corners on the left and the upper and lower corners on the right are respectively subjected to rotation transformation, translation transformation, perspective transformation and full scale transformation. The calculation process is as follows:

[0078] First, obtain the airfoil coordinates, move the section to the aerodynamic center, and solve the two-dimensional coordinates (x1, y1). Assume that the aerodynamic center coordinates are (0.25C, 0), where I is the chord length, and we can get:

[0079] (x1,y1)=(x0,y0)-(0.25C,0) (3)

[0080] Calculate the blade element coordinates based on the blade element chord length of each section:

[0081]

[0082] Combining formula (3) and formula (4) we can get:

[0083]

[0084] At this point, the three-dimensional conversion of the two-dimensional airfoil coordinates is completed, and the three-dimensional airfoil coordinates (x, y, z) are obtained.

[0085] (2) Blade 3D Modeling

[0086] The three-dimensional spatial coordinates of all blade element sections are obtained by the above method, and then the three-dimensional coordinates of each section are saved as a .txt file, and the coordinates are imported into SolidWorks to obtain each blade element section, thereby completing the three-dimensional modeling. Then, all sections are lofted to obtain the three-dimensional model of the wind turbine blade.

[0087] The present invention uses ABAQUS as a simulation analysis platform to perform finite element simulation analysis on wind turbine blades. The three-dimensional model built in SolidWorks is saved in ISG format and imported into ABAQUS software, and the unit type is defined, and solid186 entity unit is selected; then the blade model is meshed, and hexahedral meshing is used, and a total of 94610 units are divided; considering the gravity factor of the composite blade, the root of the blade is fixed with full constraints, and it is approximately regarded as a cantilever beam structure for analysis.

[0088] Set all parameters in ABAQUS, run the program in the job module for simulation analysis, and obtain the mises stress cloud map of the blade. In the ABAQUS stress cloud map, the warmth and coldness of the color indicates the size of the stress value, and the color type is selected as "Rainbow". The more red it is, the greater the stress value is, and the more blue it is, the smaller the stress value is, which means that the susceptible area changes from red to blue. From "Tools" → "Query" → "Query Value", export an Excel data file containing all node coordinates and corresponding stress values ​​of the blade three-dimensional model before and after deformation. Then simply process the data file, delete the deformed coordinates and other irrelevant node information, and retain the original coordinates before deformation. Filter the mises stress values, set the condition to be greater than or equal to a certain threshold, and you can get a set of all target points. The area occupied by these target points is an area where stress is more concentrated, which can represent the deep damage area.

[0089] S2. Combine the working condition of the blade and the specific location and terrain conditions, use the high-definition camera in the drone to record and store the macro image of the blade, and splice the captured blade images to obtain a three-dimensional image of the blade. Use the algorithm to identify and mark the location of the blade surface damage and its three-dimensional coordinates;

[0090] The inspection of wind turbine blades by drones is divided into three modules: staff, drones, and image processing. The main function of the staff part is to control the drone to fly at high altitudes, control the drone to fly to the designated position around the wind turbine blades, record the location information of the drone at this time, and design a reasonable cruise route so that the drone can fly through the preset cruise route. The drone part is mainly used to execute the inspection route set in the system and take pictures of the surface of the wind turbine blades from four directions. During the inspection process, the drone will store and transmit the taken blade images to the staff. Image processing is mainly used to extract the pictures taken by wind turbine blades for detection, identification, classification and splicing; extract the pictures of the surface of the wind turbine blades taken, pre-process the pictures obtained, and then detect and classify the defects of the wind turbine blades; finally, splice the images of each part of the wind turbine blades.

[0091] The process of acquiring wind turbine blade surface damage images based on drones, such as Figure 2As shown. First, use the drone to check the surrounding environment and influencing factors of the wind turbine blades, check the relevant data, and then set the drone system program to ensure that there are no dense trees and tall buildings in the working area of ​​the drone, so as to create good conditions for the normal operation of the drone. Set a reasonable flight speed according to the local terrain characteristics and climatic conditions. When detailed inspection of the transmission line is required, the flight speed of the drone can be appropriately reduced and controlled within the range of 4m / s. The overall process is that the staff controls the drone to fly to the designated position of the wind turbine blade, calculates the distance between the drone and the wind turbine blade, and controls the distance between the drone and the wind turbine blade within 0.5m, and records the location information of the drone at this time. The drone flies along the preset route, and uses the high-definition camera carried by the drone to shoot the pressure surface, suction surface, leading edge and trailing edge of the wind turbine blade. Figure 3 The image group taken by the drone is processed based on features, and the features of points, edges, contours, etc. in the image are extracted and matched to obtain a complete macroscopic three-dimensional image of the blade. After the drone inspection is completed, the drone can be operated to return. The staff will send a return instruction to the drone, requiring the drone to return along a reasonable and fast route.

[0092] S3. First, the three-dimensional images of the wind turbine blades in steps 1 and 2 are binarized respectively; then, since there is a certain degree of distortion in the images taken by the camera, the blade in step 1 is used as the base, and each blade segment in step 2 corresponds to it one by one, and the two are superimposed and fused, and finally a three-dimensional image of the wind turbine blade with both depth and surface damage is obtained.

[0093] (1) Binarization of deeply damaged leaf images

[0094] ① Input the image that needs to be converted to black and white;

[0095] ② Take out each pixel in the image;

[0096] ③ Calculate the color of these pixels. If the average value of the three primary colors of RGB is greater than 128, change the color of the pixel to white, and if it is less than 128, change it to black (you can also change the color of the pixel greater than 128 to black and less than 128 to white).

[0097] (2) Binarization processing of surface damaged leaf images

[0098] Threshold segmentation is an image segmentation technology based on regions. The basic principle is to divide the image pixels into several categories by setting different thresholds. Common features include: grayscale or color features from the original image; features obtained by transforming the grayscale or color values ​​in the image. In order to present an obvious black and white effect for the entire image, the grayscale image with 256 brightness levels is selected by appropriate thresholds to obtain a binary image that can still reflect the overall and local features of the image. The collective properties of these binarized images are only related to the positions of points with pixel values ​​of 0 or 255, and no longer involve multi-level values ​​of pixels, making the processing simpler, and the amount of data processing and compression is small. It is also more conducive to the development of subsequent work. The image after leaf segmentation and extraction is binarized by combining the custom threshold method and the iterative threshold method; because the image after leaf extraction removes the complex background and only retains the leaf and defect information, this method can effectively retain the accumulation core features of the leaf, and can better reflect the overall and local features of the image, thereby achieving the goal of segmenting defects. The implementation steps of the algorithm are as follows:

[0099] ① Select an initial threshold G, G = (Gmin + Gmax) / 2, where Gmin is the minimum grayscale value of the defect image and Gmax is the maximum grayscale value of the defect image;

[0100] ② According to the threshold G, the given image is divided into two groups of images, denoted as P1 and P2; P1 consists of all pixels with grayscale values ​​greater than G, and P2 consists of all pixels with grayscale values ​​less than or equal to G.

[0101] ③ Calculate the mean values ​​μ1 and μ2 of P1 and P2; calculate the new threshold G1, and G1=(μ1+μ2) / 2;

[0102] ④ If G1 no longer changes, G1 is the threshold; otherwise, go to step (2) and recalculate until the difference of G in consecutive iterations is less than the predetermined value.

[0103] The process of using image segmentation algorithm for binarization is as follows: Figure 4 shown.

[0104] (3) Image mapping fusion processing

[0105] Considering that the images taken by the drone have certain distortion, the size of the blade 3D image obtained in step 2 may be different from the actual image (in step 1). Figure 5As shown, the upper contour is the blade model in step 1, and the lower contour is the blade model in step 2. According to the length of the blade chord (the line connecting the leading edge endpoint and the trailing edge endpoint), select a suitable ratio to divide the blade into five small sections and make point-to-point correspondences, such as 1-1', 2-2', ... 6-6'. Therefore, taking the blade in step 1 as the base, and the leading edge endpoint and the trailing edge endpoint as the base points, the size and length of each blade section are roughly equal. The depth and surface damage features contained in the wind turbine blades are mapped to the blade image in step 1 and reflected. In addition, areas without features or with lighter features are regarded as walkable areas, and areas with obvious features are regarded as areas to be detected.

[0106] S4. In the binarized image, the main body of the leaf is black, and the damage on the leaf is white. In order to make the damage features clearer and the maintenance more efficient, multiple smooth straight lines are used to connect the boundary points scattered in the white area to form the contour line of the damage area, that is, to obtain multiple irregular polygons.

[0107] The alpha shapes algorithm, also known as the rolling ball method, is a simple and effective algorithm for quickly extracting boundary points. It is suitable for extracting the coordinates of boundary points from discrete spatial point sets. The present invention introduces this method into the boundary extraction of hash-type damage areas to extract the coordinates of the boundaries. The area enclosed by the boundary is the target area for unmanned system detection and maintenance.

[0108] The basic principle of the algorithm is: in a spatial discrete point set A, there are m discrete points in total. Obviously, m points can be connected into m*(m-1) / 2 line segments. For all discrete points, draw a circle with alpha as the radius, and the circle must pass through points Q1 and Q2 (the radius is constant, and there are two circles passing through two certain points). If there are no other discrete points in the circle, Q1 and Q2 must be on the boundary, and the line segment Q1-Q2 belongs to the boundary segment.

[0109] The specific calculation steps are as follows: Sep1, traverse all boundary line segments Q1-Q2; Sep2, if the length of Q1-Q2 is greater than alpha which is twice the diameter, skip it (it is easy to know that there is no circle passing through these two points); Sep3, according to geometric principles, find the centers R1 and R2 of the two circles; Sep4, if one of the two circles does not contain any other points in the point set A, then the edge Q1-Q2 is the boundary.

[0110] The blade damage inspection map construction method provided by the present invention takes the places with more concentrated stress in the mises cloud map as the deep damage area; multiple local photos of the wind turbine blade are taken and data information is collected, and then the whole blade image is spliced. The whole image is identified based on the deep learning algorithm to determine the surface damage area on the wind turbine blade.

[0111] Compared with the prior art, the present invention has the following beneficial effects: the blade damage inspection map construction method and the blade damage inspection method provided by the present invention can achieve full inspection of deep and surface damage, and both inspection efficiency and safety are improved.

[0112] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0113] Unless otherwise defined, the technical terms or scientific terms used in the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, similar words such as "one", "one" or "the" do not indicate quantity restrictions, but indicate that there is at least one. Similar words such as "include" or "comprise" mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Similar words such as "connect" or "connected" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

Claims

1. A method for constructing a blade damage inspection map, characterized in that: The following steps are involved: Obtaining a three-dimensional image of the depth damage of the blade to be inspected; Acquire a three-dimensional image of the surface damage of the blade to be inspected; Firstly, the deep damage 3D image and the surface damage 3D image are fused to obtain a damage fusion image, and then all damage areas in the damage fusion image are extracted; or firstly, the deep damage area in the deep damage 3D image is extracted, and the surface damage area in the surface damage 3D image is extracted, and then the deep damage 3D image and the surface damage 3D image are fused to obtain a damage fusion image with all damage areas; The method of obtaining the deep damage image of the blade to be inspected specifically comprises the following steps: establishing a three-dimensional simulation analysis model of the wind turbine blade to be inspected; performing stress analysis on the three-dimensional simulation analysis model to obtain a three-dimensional stress cloud map; and using the three-dimensional stress cloud map as the deep damage three-dimensional image; The method of obtaining the surface damage image of the blade to be inspected specifically comprises the following steps: using a drone to photograph the blade to be inspected to obtain a three-dimensional surface damage image of the blade to be inspected; The step of fusing the deep damage three-dimensional image and the surface damage three-dimensional image to obtain the damage fused image specifically includes the following steps: Binarizing the three-dimensional image of the depth damage to obtain a binary image of the depth damage; Binarizing the three-dimensional image of surface damage to obtain a binary image of surface damage; Taking the depth damage binary image as a matrix, the surface damage binary image is mapped to the depth damage binary image in proportion to form a damage fusion image; The binarization process of the depth damage three-dimensional image to obtain the depth damage binary image specifically includes the following steps: extracting each pixel point in the depth damage three-dimensional image; obtaining the color component of each pixel point; comparing the threshold range of the color component of each pixel point with the first set threshold range, and modifying one of the pixel points within the first set threshold range and the pixel points outside the first set threshold range to white and the other to black; The three-dimensional image of surface damage is binarized to obtain the binary image of surface damage, which specifically includes the following steps: obtaining a grayscale image of the three-dimensional image of surface damage; extracting each pixel in the grayscale image; obtaining the grayscale value of each pixel; dividing the pixel points whose grayscale values ​​are greater than a second set threshold range and the pixel points whose grayscale values ​​are less than the second set threshold range into two groups, one group is modified to black, and the other group is modified to white; The second set threshold is determined by combining a custom threshold method and an iterative threshold method, including the following steps: S11, select an initial threshold G, G = (Gmin + Gmax) / 2, where Gmin is the minimum grayscale value of the grayscale image, and Gmax is the maximum grayscale value of the grayscale image; S12, divide the grayscale image into two groups according to the initial threshold G, denoted as and ; It consists of all pixels with gray value greater than G. It consists of all pixels with gray value less than or equal to G; S13, calculation and Mean and ; Calculate the new threshold ,and ; S14, if No longer changes, This is the second set threshold; otherwise, repeat S2 and S3 to recalculate the new threshold , until the new threshold in successive iterations Until the difference is less than the predetermined value, the new threshold obtained by the latest iteration As a second set threshold; Taking the depth damage binary image as a matrix, mapping the surface damage binary image to the depth damage binary image in proportion to form a damage fusion image specifically includes the following steps: Before applying stress to the three-dimensional simulation analysis model for stress analysis, an original three-dimensional stress cloud map of the three-dimensional simulation analysis model is obtained, and the original coordinates of each node in the original three-dimensional stress cloud map are recorded; the coordinates of the leading edge endpoint and the trailing edge endpoint in the deep damage binary image are obtained through the original coordinates of each node in the original three-dimensional stress cloud map; the first chord length of the blade in the deep damage binary image is calculated using the coordinates of the leading edge endpoint and the trailing edge endpoint in the deep damage binary image; Acquire the coordinates of the leading edge endpoint and the trailing edge endpoint in the surface damage binary image; calculate the second chord length of the blade in the surface damage binary image using the coordinates of the leading edge endpoint and the trailing edge endpoint in the surface damage binary image; Divide the depth damage binary image into N first small segments according to the first chord length, forming 2N first mapping points; divide the surface damage binary image into N second small segments according to the second chord length, forming 2N second mapping points; The 2N second mapping points are respectively matched one by one with the 2N first mapping points, and the surface damage binary image and the depth damage binary image are fused to obtain the damage fused image; The blade damage inspection map construction method further comprises the following steps: The damage area is defined as the area to be inspected in the damage inspection map; The area outside the damage area is defined as the walking area in the damage inspection map; The area to be inspected and the walking area together form a damage inspection map.

2. The method for constructing a blade damage inspection map according to claim 1, characterized in that: The method of using a drone to photograph the blade to be inspected to obtain a three-dimensional image of the surface damage of the blade to be inspected specifically includes the following steps: Using a drone to shoot the leading edge, pressure side, trailing edge, and suction side of the blade, four images are obtained to form an image sequence; or using a drone to shoot the blade along a preset route that passes through the leading edge, pressure side, trailing edge, and suction side in sequence, multiple images are obtained to form an image sequence; A multi-view 3D reconstruction algorithm is used for image stitching, SFIT feature extraction and matching are performed on the image sequence, and feature extraction and matching are performed on the points, edges and contours in the image sequence; the linear triangulation method is used to calculate the 3D space point coordinate cloud, and a sparse point cloud is reconstructed; the bundle method is used for optimization; the classic dense algorithm PMVS is used to obtain a dense space directed point cloud or patch; erroneous point cloud or patch data is filtered out to obtain a complete macroscopic 3D image of the blade as the surface damage 3D image.

3. The method for constructing a blade damage inspection map according to claim 1, characterized in that: Extracting all damaged areas in the damage fusion image specifically includes the following steps: Based on the alpha shapes algorithm, the boundary of the damaged area is extracted. In a spatial discrete point set A, there are m discrete points in total, and the m points are connected into m* (m-1) / 2 line segments; for all discrete points, a circle is drawn with alpha as the radius, and the circle must pass through the point , Two points, if there are no other discrete points in the circle, then , Must be on the boundary, line segment Belongs to the boundary segment; The specific calculation steps are as follows: S21, traverse all boundary segments ; S22, if If the length is greater than alpha which is twice the diameter, it will be skipped; S23, according to the principle of geometry, find the center of two circles , ; S24, if one of the two circles does not contain any other points in point set A, then the line segment That is the boundary; The scattered boundaries are connected with multiple smooth straight lines to form the contour line of the damaged area, that is, multiple irregular polygons are obtained, and each polygon corresponds to a damaged area.

4. The method for constructing a blade damage inspection map according to claim 2, characterized in that: The method of extracting the deep damage area from the deep damage three-dimensional image specifically comprises the following steps: selecting and marking the node coordinates whose stress is greater than the allowable stress of the blade material from the three-dimensional stress cloud map, and the set of the node coordinates forms the deep damage area; The method of extracting the surface damage area in the surface damage three-dimensional image specifically includes the following steps: marking the surface damage area and its three-dimensional coordinates in the surface damage three-dimensional image by image recognition.

5. The method for constructing a blade damage inspection map according to claim 4, characterized in that: During image recognition, damage features in the image are extracted, including cracks, sand holes, pits, and spalling; the damage features are pre-input into the deep learning algorithm model for feature recognition training, and the trained deep learning algorithm model is used for image recognition; Before image stitching and / or image recognition, the image information is preprocessed, specifically including the following steps: performing histogram equalization processing and / or median filtering to eliminate part of the noise in the image; When using a drone to photograph the blades to be inspected, the flight speed of the drone is controlled to be less than or equal to 4 m / s; the distance between the drone and the blade is controlled to be less than or equal to 0.5 m; when the drone is controlled to fly along a preset route, the position information of the drone is recorded, and based on the position information of the wind turbine blades, the blade area information where the drone was located when taking each image is determined, and the stored image and position information are transmitted to the staff operating the drone for image processing and analysis in the background to obtain the three-dimensional image of the surface damage.

6. A blade damage inspection method using the blade damage inspection map construction method according to claim 1, characterized in that: The following steps are involved: Planning the inspection route of the drone according to the coordinates of the area to be inspected and the coordinates of the walking area in the blade damage inspection map, so that the drone takes the shortest time to pass through all the areas to be inspected; wherein the blade damage inspection map is constructed according to the blade damage inspection map construction method; Control the drone to fly along the inspection route; When the drone passes through the area to be inspected, the drone is controlled to take photos of the blades to be inspected to obtain inspection pictures.

Images