Fan blade deformation degree detection method and system

Abstract

The invention relates to the technical field of wind power generation, and discloses a fan blade deformation degree detection method and system, and the method comprises the steps: obtaining laser radar point cloud data, an adaptive image and a high-frequency vibration signal of a fan blade; generating a real-time three-dimensional model of the fan blade; calculating the displacement of the preset key feature points, and generating a global deformation thermodynamic diagram; constructing a dynamic attractor sequence based on the high-frequency vibration signal, and extracting multi-scale fingerprint features through the dynamic attractor sequence; and according to the global deformation thermodynamic diagram, the deformation degree of the fan blade is determined in combination with curvature change, and based on a comparison result of the multi-scale fingerprint features and a preset fingerprint threshold value, a micro-deformation area is positioned and damage early warning is realized. According to the method, the problems of poor data quality, low fusion degree, inaccurate deformation quantification, early warning lag and the like in the prior art are effectively solved, the detection precision, efficiency and reliability are remarkably improved, and comprehensive technical support is provided for preventive maintenance of wind power equipment.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and more specifically, to a method and system for detecting the degree of deformation of wind turbine blades. Background Technology

[0002] As the core component for energy conversion in wind turbine units, wind turbine blades endure complex operating conditions such as alternating loads, extreme temperatures, and dust erosion over long periods, making them prone to progressive damage ranging from microscopic cracks to macroscopic deformation. If deformation is not detected in time, it can lead to major accidents such as blade breakage and turbine shutdown. Therefore, accurate monitoring of the degree of deformation is a core requirement for ensuring the safe operation of wind power equipment. Current mainstream inspection technologies include manual inspection, UAV visual inspection, laser point cloud modeling, and vibration signal analysis. Manual inspection relies on close-range observation by operators, resulting in low efficiency (2-3 hours per blade), high-altitude operation risks, and significant subjective judgment errors. UAV visual inspection is significantly affected by lighting conditions; the sunlit side of the blade is prone to overexposure, while the shaded side is prone to underexposure, leading to the loss of surface texture details and subsequent defect identification accuracy of less than 70%. While laser point cloud modeling can acquire three-dimensional geometric data, it lacks surface texture information fusion, making it difficult to capture microscopic deformation features such as coating peeling and fiber warping.

[0003] In other words, existing technologies are limited not only in single-dimensional detection but also lack an effective linkage mechanism between macroscopic and microscopic features. Laser point cloud and image data often suffer from spatial registration errors, leading to a disconnect between geometric and textural information, making it difficult to achieve coordinated analysis of shape changes and surface conditions. While vibration signal analysis can capture the dynamic characteristics of early micro-deformations, traditional methods struggle to correlate vibration anomalies with the specific spatial location of the blade, resulting in ambiguous early warning area positioning. Furthermore, the separation between macroscopic deformation variability and microscopic damage early warning makes it difficult for detection results to support full lifecycle risk assessment from minor deformation to potential fracture, leading to problems such as high false alarm rates, missed detections, and delayed warnings, failing to meet the needs of large-scale, refined operation and maintenance in large wind farms.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In view of the problems in the related technologies, the present invention proposes a method and system for detecting the degree of deformation of wind turbine blades, so as to overcome the above-mentioned technical problems existing in the existing related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for detecting the degree of deformation of wind turbine blades is provided, comprising the following steps: S1. Control the UAV to fly autonomously according to the preset flight route, and acquire lidar point cloud data, adaptive images and high-frequency vibration signals of the wind turbine blades during the flight. S2. By fusing lidar point cloud data and adaptive images, a real-time 3D model of the wind turbine blade is generated, and the real-time 3D model of the wind turbine blade is aligned with the reference model using a data registration method. S3. Based on the real-time 3D model and benchmark model of the aligned wind turbine blades, calculate the displacement of preset key feature points, and generate a global deformation heat map by combining the texture deformation in the adaptive image. S4. Based on high-frequency vibration signals, dynamic attractor sequences are constructed by combining a sliding window to update the embedding dimension and delay time in real time, and multi-scale fingerprint features are extracted through the dynamic attractor sequences. S5. Based on the global deformation heat map and the curvature change, determine the degree of deformation of the wind turbine blades. Based on the comparison results of multi-scale fingerprint features and preset fingerprint thresholds, locate the micro-deformation area and realize damage early warning.

[0007] Furthermore, controlling the UAV to fly autonomously according to a preset flight route, and acquiring lidar point cloud data, adaptive images, and high-frequency vibration signals of the wind turbine blades during flight, includes the following steps: S11. Plan the flight route of the UAV based on the wind turbine blades in the target area, control the UAV to fly autonomously based on the flight route, and acquire the lidar point cloud data of the wind turbine blades through lidar during the flight. S12. Use a camera to identify the light intensity in different areas of the wind turbine blades and determine the corresponding exposure parameters based on the light intensity in different areas; based on the determined exposure parameters, use the camera to acquire adaptive images of the wind turbine blades. S13. The high-frequency vibration signal of the wind turbine blade is continuously collected by the vibration sensor installed at the root of the wind turbine blade for a preset period of time, and the high-frequency vibration signal is subjected to wavelet threshold denoising processing; the lidar point cloud data and adaptive image are processed to ensure that the texture corresponds to the point cloud.

[0008] Furthermore, the camera is used to identify the light intensity in different areas of the wind turbine blades, and the corresponding exposure parameters are determined based on the light intensity in different areas. Based on the determined exposure parameters, the camera is used to acquire adaptive images of the wind turbine blades, including the following steps: S121. Use a camera to acquire initial low-resolution images of the wind turbine blades in real time, identify the outline of the wind turbine blades through an edge detection algorithm, and divide the wind turbine blades into several regions based on geometric features. S122. Extract the average pixel grayscale value of each region, and convert the average pixel grayscale value into the actual illumination level according to the mapping relationship between grayscale and illumination intensity. S123. Determine the exposure parameters for different areas of the wind turbine blade surface according to the preset mapping table of illumination level and exposure parameters. S124. Based on the exposure parameters of different regions, use a camera to capture single-frame images of the corresponding regions, stitch together the images of each region, and use an edge fusion algorithm to eliminate the stitching seams caused by parameter differences, generating a clear wind turbine blade adaptive image across the entire area.

[0009] Furthermore, by fusing lidar point cloud data and adaptive images, a real-time 3D model of the wind turbine blade is generated, and the real-time 3D model of the wind turbine blade is aligned with the baseline model using a data registration method, including the following steps: S21. Based on the camera extrinsic parameters and lidar calibration parameters, establish a mapping relationship between point cloud coordinates and image pixel coordinates, and associate the pixel color of the adaptive image with the three-dimensional points of the point cloud through the mapping relationship to achieve the binding of geometric coordinates and texture information. S22. Using the Poisson surface reconstruction algorithm, the initial mesh model of the wind turbine blade is generated by combining the associated point cloud, and the texture mapping of the adaptive image is attached to the surface of the initial mesh model to form a textured real-time 3D model, thus obtaining the real-time 3D model of the wind turbine blade. S23. Obtain the 3D CAD model of the wind turbine blade during the design phase and convert it into point cloud format to obtain the reference model of the wind turbine blade, and extract key geometric features as a set of registration reference points; based on the set of registration reference points, achieve the alignment of the real-time 3D model with the reference model through the data registration method.

[0010] Furthermore, based on the real-time 3D model and baseline model of the aligned wind turbine blades, the displacement of preset key feature points is calculated, and a global deformation heat map is generated by combining the texture deformation in the adaptive image, including the following steps: S31. Based on the real-time three-dimensional model and the reference model of the aligned wind turbine blades, extract the preset key points and calculate the coordinate difference of the corresponding points in the three-dimensional coordinate system, and determine the absolute displacement of each key point based on the Euclidean distance. S32. Perform texture comparison on the regions in the adaptive image corresponding to the preset key points, and combine the feature matching algorithm to identify the degree of texture deformation, and quantify the degree of texture deformation into texture deformation coefficient. S33. Based on the mesh nodes of the real-time 3D model, combined with the absolute displacement of each key point, the deformation of the unmeasured points in the whole domain is estimated by using the Kriging interpolation method, and the texture deformation coefficient is incorporated to correct the interpolation results, generating a deformation matrix covering the entire blade. S34. Map the deformation matrix to a color gradient to generate an initial global deformation heatmap. Then, calibrate the color mapping accuracy of the initial global deformation heatmap by comparing the deformation regions in the adaptive image to obtain the global deformation heatmap of the wind turbine blade.

[0011] Furthermore, texture comparison is performed on the regions in the adaptive image corresponding to preset key points, and the degree of texture deformation is identified by combining feature matching algorithms. The degree of texture deformation is then quantified into texture deformation coefficients, including the following steps: S321. Based on the camera intrinsic parameter matrix, the three-dimensional coordinates of the preset key points in the real-time three-dimensional model are projected into the pixel coordinates of the adaptive image, and a square region with a predetermined side length is extracted as the texture analysis window with the pixel coordinates of the projected adaptive image as the center. S322. Extract the region corresponding to the current window position from the historical adaptive image corresponding to the baseline model as the baseline texture region, and extract the texture analysis window region determined in step S321 from the current adaptive image as the current texture region. S323. Use a feature matching algorithm to extract feature points of the baseline texture region and the current texture region, and each feature point is represented by several feature point descriptors; calculate the similarity of the feature point descriptors by Euclidean distance, and select matching point pairs based on the similarity calculation results; S324. Based on the selected matching point pairs, fit the affine transformation matrix using the least squares method, and calculate the deformation parameters according to the affine transformation matrix, where the deformation parameters include scaling factor, rotation angle and shear coefficient. S325. Determine the texture deformation coefficient based on the scaling factor, rotation angle and shearing coefficient, combined with the preset weight, and correct the texture deformation coefficient based on the gray-scale variance change rate of the region when the number of matching point pairs is less than the preset threshold. The formula for calculating the texture deformation coefficient is as follows: In the formula, C Represents the texture deformation coefficient. s Indicates the scaling factor. θ Indicates the rotation angle. k Indicates the shear coefficient. These represent the weights of the scaling factor, rotation angle, and shear coefficient, respectively. a , d , e These represent different linear deformation parameters of the texture region from the baseline state to the current state. This represents the corrected texture deformation coefficient. This represents the rate of change of the gray-scale variance of the region.

[0012] Furthermore, based on the mesh nodes of the real-time 3D model and combined with the absolute displacement of each key point, the deformation of the unmeasured points across the entire domain is estimated using the Kriging interpolation method. Texture deformation coefficients are then incorporated to correct the interpolation results, generating a deformation matrix covering the entire blade. This process includes the following steps: S331. Extract the mesh node set of the real-time 3D model and select preset key points as sample points with known displacement. S332. Fit the variogram function based on the spatial distance of the sample points, and construct the Kriging equation system based on the calculation results of the variogram function; solve the weight coefficients through the Kriging equation system, and estimate the initial deformation of each grid node based on the weight coefficients. S333. Obtain the texture region and texture deformation coefficient corresponding to each grid node in the adaptive image, calculate the correction weight based on the texture deformation coefficient, and correct the initial deformation amount through the correction weight to obtain the corrected deformation amount. S334. Integrate the deformation amounts after correction of all mesh nodes to generate a deformation matrix covering the entire blade; The expression for the corrected deformation is as follows: In the formula, Z j Indicates the first j The deformation amount after correction for each sample point Indicates the first j The initial deformation of each sample point Indicates the correction strength. Indicates the first j Corrected weights for each sample point K Indicates the number of feature points. Indicates the weighting coefficient. Indicates the first i The absolute displacement of a sample point in the Z direction. The weighting factor represents the texture deformation coefficient. Indicates the first j Texture deformation coefficients of each sample point h Indicates the spatial distance between sample points. This represents the distance attenuation coefficient.

[0013] Furthermore, based on high-frequency vibration signals, and combined with a sliding window to update the embedding dimension and delay time in real time, a dynamic attractor sequence is constructed, and multi-scale fingerprint features are extracted through the dynamic attractor sequence, including the following steps: S41. The high-frequency vibration signal is segmented and processed, a sliding window is set, and the signal in each window is denoised and standardized; the phase space reconstruction technique is used to determine the real-time optimal delay time, and the embedding dimension is updated by combining the false nearest neighbor method. S42. Using the updated window embedding dimension, the one-dimensional sub-signal is reconstructed into an attractor in a high-dimensional phase space, and the attractors of the continuous window are arranged according to the time order to form a dynamic attractor sequence. Each attractor in the dynamic attractor sequence corresponds to the local dynamic characteristics of the vibration signal, reflecting the temporal evolution of the micro-deformation of the wind turbine blade. S43. Based on the dynamic attractor sequence, the global scale is extracted to calculate the maximum Lyapunov exponent and fractal dimension, and the density entropy and local wrinkle curvature of the attractor are calculated by the sliding window to obtain multi-scale fingerprint features. The formula for calculating the curvature of local folds is as follows: In the formula, Indicates the curvature of a local fold. M Indicates the number of phase points. X k,i Indicates the first k The first window i One phase point, , denoted as the first and second derivative moduli of the phase point, respectively.

[0014] Furthermore, based on the global deformation heat map and the curvature change, the degree of deformation of the wind turbine blades is determined. Based on the comparison results of multi-scale fingerprint features and preset fingerprint thresholds, the micro-deformation areas are located and damage early warning is achieved, including the following steps: S51. Extract the deformation values ​​of each region based on the global deformation heat map, calculate the local wrinkle curvature of the real-time three-dimensional model and the benchmark model, and determine the macroscopic deformation level of the wind turbine blades based on the deformation and local wrinkle curvature. S52. Compare the multi-scale fingerprint feature sequence with the preset fingerprint threshold. When any multi-scale fingerprint feature sequence exceeds the window of the preset fingerprint threshold, locate the micro-deformation area of ​​the corresponding wind turbine blade through the mapping relationship between the sliding window and the spatial position of the wind turbine blade. When a preset number of consecutive windows trigger the threshold, it is determined to be a potential damage and a micro-deformation warning area is generated. S53. Associate macroscopic deformation level with micro-deformation warning area. When a severe deformation area overlaps with a micro-deformation warning area, it is marked as a high-risk area. When a slightly deformed area has a micro-deformation warning, it is marked as a region to be reviewed. The final output includes a comprehensive result that includes deformation degree classification, risk area location and warning level.

[0015] According to another aspect of the present invention, a wind turbine blade deformation degree detection system is provided, including a data acquisition module, a real-time three-dimensional model generation module, a global deformation heat map generation module, a multi-scale fingerprint feature extraction module, and a deformation degree determination module; The data acquisition module is used to control the UAV to fly autonomously according to the preset flight route, and to acquire lidar point cloud data, adaptive images and high-frequency vibration signals of the wind turbine blades during the flight. The real-time 3D model generation module is used to fuse lidar point cloud data and adaptive images to generate a real-time 3D model of the wind turbine blades, and to align the real-time 3D model of the wind turbine blades with the reference model through data registration. The global deformation heat map generation module is used to calculate the displacement of preset key feature points based on the real-time 3D model and benchmark model of the aligned wind turbine blades, and generate a global deformation heat map by combining the texture deformation in the adaptive image. The multi-scale fingerprint feature extraction module is used to construct a dynamic attractor sequence based on high-frequency vibration signals, combined with a sliding window to update the embedding dimension and delay time in real time, and to extract multi-scale fingerprint features through the dynamic attractor sequence. The deformation degree determination module is used to determine the deformation degree of the wind turbine blades based on the global deformation heat map and the curvature change. Based on the comparison results of multi-scale fingerprint features and preset fingerprint thresholds, it locates micro-deformation areas and realizes damage early warning.

[0016] The beneficial effects of this invention are as follows: 1) This invention achieves a complete innovation in the entire process of wind turbine blade deformation detection from data acquisition to accurate early warning through multi-source data fusion, adaptive parameter optimization, multi-scale feature extraction and linkage early warning mechanism. It effectively solves the problems of poor data quality, low fusion degree, inaccurate deformation quantification and early warning lag in the existing technology, significantly improves the accuracy, efficiency and reliability of detection, and provides comprehensive technical support for the preventive maintenance of wind power equipment.

[0017] 2) This invention can not only generate adaptive images by dynamically adjusting exposure parameters to solve the image quality problem caused by uneven illumination, but also simultaneously acquire laser point clouds and high-frequency vibration signals, laying a high-quality data foundation for subsequent multi-dimensional analysis. Moreover, the point cloud and image fusion modeling can also significantly improve the accuracy of real-time 3D models by accurately binding geometric coordinates and texture information, combined with Poisson surface reconstruction and feature point registration technology, ensuring that the alignment error with the benchmark model is controlled at the millimeter level, and providing a reliable geometric benchmark for deformation quantization.

[0018] 3) This invention can generate a global deformation heatmap through Kriging interpolation and texture deformation coefficient correction, achieving accurate estimation of deformation from key feature points to the entire domain. Combined with color gradient mapping and texture calibration, it improves the visualization accuracy of the deformation area, overcoming the shortcomings of traditional geometric modeling that ignores micro-texture deformation. Furthermore, it can capture the chaotic characteristics of vibration signals by updating embedded parameters in real time to construct dynamic attractor sequences, extract multi-scale fingerprint features, and utilize micro-features such as local fold curvature to achieve early identification of micro-deformation, making up for the lack of spatial positioning in traditional vibration analysis.

[0019] 4) This invention can perform macroscopic and microscopic deformation linkage analysis by associating heat map levels with fingerprint feature thresholds, realizing multi-dimensional verification of geometric deformation, texture changes and vibration anomalies, significantly reducing the false alarm rate of single indicators, accurately locating high-risk areas, and solving the problems of lagging or ambiguous positioning in existing early warning mechanisms. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram illustrating the principle of a method for detecting the degree of deformation of wind turbine blades according to an embodiment of the present invention. Detailed Implementation

[0022] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0023] According to an embodiment of the present invention, a method and system for detecting the degree of deformation of wind turbine blades are provided.

[0024] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a method for detecting the degree of deformation of wind turbine blades is provided, comprising the following steps: S1. Control the UAV to fly autonomously according to the preset flight route, and acquire lidar point cloud data, adaptive images and high-frequency vibration signals of the wind turbine blades during the flight. The process of controlling the UAV to fly autonomously according to a preset flight route and acquiring lidar point cloud data, adaptive images, and high-frequency vibration signals of the wind turbine blades during flight includes the following steps: S11. Plan the flight route of the UAV based on the wind turbine blades in the target area, control the UAV to fly autonomously based on the flight route, and acquire the LiDAR point cloud data of the wind turbine blades through LiDAR during the flight, that is, record the three-dimensional coordinates of the wind turbine blade surface, including the geometric shape of the blade in the current state. In this embodiment, the UAV is equipped with a 16-line lidar (point cloud density ≥ 200 points / m²). 2 ), 20-megapixel high-definition camera (with anti-reflective lens), GNSS-RTK module (positioning accuracy ≤5cm), ensuring battery life meets the requirements for single-leaf inspection (approximately 15 minutes / leaf).

[0025] A multi-angle, circling shooting path was employed: three circling tracks (5-8m from the blade surface) were set with the blade root as the center, corresponding to the leading, middle, and trailing edges of the blade, respectively; simultaneously, intersecting tracks perpendicular to the blade axis were set at the blade tip, half-blade length, and one-quarter-blade length positions to ensure no blind spots. The LiDAR sampling frequency was set to 10Hz, and the camera shutter was synchronously triggered (one image every 0.5 seconds), with a track overlap rate of ≥80% (to ensure the accuracy of point cloud and image stitching).

[0026] S12. Use a camera to identify the light intensity in different areas of the wind turbine blades and determine the corresponding exposure parameters based on the light intensity in different areas; based on the determined exposure parameters, use the camera to acquire adaptive images of the wind turbine blades. Specifically, the camera is used to identify the light intensity in different areas of the wind turbine blades, and the corresponding exposure parameters are determined based on the light intensity in different areas; based on the determined exposure parameters, the camera is used to acquire adaptive images of the wind turbine blades, including the following steps: S121. Use a camera to acquire initial low-resolution images of the wind turbine blades in real time, identify the outline of the wind turbine blades through an edge detection algorithm, and divide the wind turbine blades into several regions based on geometric features. In this embodiment, the camera acquires the initial low-resolution image of the leaf in real time, identifies the leaf outline through an edge detection algorithm (such as the Canny operator), and divides the leaf into independent regions according to geometric features (leading edge / tailing edge / leaf tip / leaf root) or a preset grid (5×5 pixel block), with each region serving as a light detection unit.

[0027] S122. Extract the average pixel grayscale value of each region, and convert the average pixel grayscale value into the actual illumination level according to the mapping relationship between grayscale and illumination intensity. In this embodiment, for each segmented region, the average pixel grayscale value (0-255) is extracted and converted into the actual illumination level (e.g., level 1-5) through a grayscale-light intensity mapping model (pre-calibrated, for example, grayscale value >200 corresponds to strong light and <50 corresponds to weak light).

[0028] S123. Determine the exposure parameters for different areas of the wind turbine blade surface according to the preset mapping table of illumination level and exposure parameters. In this embodiment, the mapping table between illumination level and exposure parameters is as follows: strong light area (level 4-5), shutter speed 1 / 1000-1 / 2000s, ISO 100-200, to avoid overexposure; weak light area (level 1-2), shutter speed 1 / 100-1 / 200s, ISO 400-800, to increase brightness; medium light area (level 3), shutter speed 1 / 500s, ISO 200, to balance sharpness and noise.

[0029] S124. Based on the exposure parameters of different regions, use a camera to capture single-frame images of the corresponding regions, stitch together the images of each region, and use an edge fusion algorithm to eliminate the stitching seams caused by parameter differences, generating a clear wind turbine blade adaptive image across the entire area.

[0030] In this embodiment, after capturing a single frame image according to the matching parameters, it is immediately detected whether the grayscale value of each region is within the target range (80-200). If the deviation of a certain region exceeds 30% (for example, the grayscale value of the strong light area is still >230), the parameters are locally fine-tuned (such as increasing the shutter speed by 10%), and the sub-image of that region is recaptured. The optimized sub-images of each region are stitched together, and the stitching seams caused by parameter differences are eliminated by the edge fusion algorithm to generate a clear leaf adaptive image across the entire domain, ensuring that texture details (such as glue marks and micro-cracks) are completely preserved.

[0031] S13. The high-frequency vibration signal of the wind turbine blade is continuously collected by the vibration sensor installed at the root of the wind turbine blade for a preset period of time, and the high-frequency vibration signal is subjected to wavelet threshold denoising processing; the lidar point cloud data (denoising and segmentation) and adaptive image (local exposure compensation data is fused during stitching to avoid stitching seams) are processed to ensure that the texture corresponds to the point cloud.

[0032] In this embodiment, wavelet packet decomposition combined with chaotic threshold screening is used to subdivide high-frequency signals (2000-5000Hz) into 100Hz frequency bands, retaining only frequency bands with significant chaotic characteristics (judged by sample entropy > 0.8); low-frequency signals and high-frequency signals are subjected to coherence verification, and interference components caused by the overall vibration of the wind turbine (such as signals below 10Hz caused by tower resonance) are removed, and finally pure chaotic signals that are strongly correlated with the micro-deformation of the blades are retained.

[0033] S2. By fusing lidar point cloud data and adaptive images, a real-time 3D model of the wind turbine blade is generated, and the real-time 3D model of the wind turbine blade is aligned with the reference model using a data registration method. The process of fusing lidar point cloud data and adaptive images to generate a real-time 3D model of the wind turbine blades, and aligning the real-time 3D model of the wind turbine blades with the baseline model using a data registration method, includes the following steps: S21. Based on the camera extrinsic parameters (focal length, distortion coefficient) and lidar calibration parameters, establish a mapping relationship between point cloud coordinates and image pixel coordinates, and associate the pixel color of the adaptive image with the three-dimensional points of the point cloud through the mapping relationship to achieve the binding of geometric coordinates and texture information. S22. Using the Poisson surface reconstruction algorithm, the initial mesh model of the wind turbine blade is generated by combining the associated point cloud, and the texture mapping of the adaptive image is attached to the surface of the initial mesh model to form a textured real-time 3D model, thus obtaining the real-time 3D model of the wind turbine blade. S23. Obtain the 3D CAD model of the wind turbine blade during the design phase and convert it into point cloud format to obtain the reference model of the wind turbine blade. Extract key geometric features (such as blade axis, root flange edge, and blade tip) as a registration reference point set. Based on the registration reference point set, align the real-time 3D model with the reference model using a data registration method (such as the ICP algorithm). Specifically, a local coordinate system is established with the center of the flange of the reference model as the origin. Corresponding feature points are selected in the real-time model. The rotation matrix and translation vector are calculated iteratively to minimize the error of the feature point set between the two models (root mean square error ≤ 0.5 mm). The global distance deviation between the two models after registration is calculated. If the proportion of the area exceeding 1 mm is ≤ 5%, the registration is deemed valid. Otherwise, the feature point set is reselected (such as adding the leading edge inflection point in the leaf) and the registration is repeated until the accuracy requirements are met.

[0034] S3. Based on the real-time 3D model and benchmark model of the aligned wind turbine blades, calculate the displacement of preset key feature points, and generate a global deformation heat map by combining the texture deformation in the adaptive image. The process of calculating the displacement of preset key feature points based on the real-time 3D model and baseline model of the aligned wind turbine blades, and generating a global deformation heatmap by combining the texture deformation in the adaptive image, includes the following steps: S31. Based on the real-time three-dimensional model and the reference model of the aligned wind turbine blades, extract the preset key points and calculate the coordinate difference of the corresponding points in the three-dimensional coordinate system, and determine the absolute displacement of each key point based on the Euclidean distance. S32. Perform texture comparison on the regions in the adaptive image corresponding to the preset key points, and combine the feature matching algorithm to identify the degree of texture deformation (including stretching, compression or wrinkling changes, etc.), and quantify the degree of texture deformation into texture deformation coefficient. Specifically, the process of performing texture comparison on regions corresponding to preset key points in the adaptive image, and identifying the degree of texture deformation using a feature matching algorithm, and quantifying the degree of texture deformation into texture deformation coefficients, includes the following steps: S321. Based on the camera intrinsic parameter matrix, the three-dimensional coordinates of the preset key points in the real-time three-dimensional model are projected into the pixel coordinates of the adaptive image, and a square region with a predetermined side length is extracted as the texture analysis window with the pixel coordinates of the projected adaptive image as the center. In this embodiment, a square region with a side length of 2n+1 pixels (e.g., n=20, i.e., 41×41 pixels) is selected as the texture analysis window to ensure that typical textures (such as coating textures and adhesive marks) around the key point are included.

[0035] S322. Extract the region corresponding to the current window position from the historical adaptive image corresponding to the baseline model as the baseline texture region, and extract the texture analysis window region determined in step S321 from the current adaptive image as the current texture region. S323. Use a feature matching algorithm to extract feature points from the baseline texture region and the current texture region, and each feature point is represented by a 128-dimensional feature point descriptor; calculate the similarity of the feature point descriptors using Euclidean distance, and select matching point pairs based on the similarity calculation results; S324. Based on the filtered matching point pairs, fit the affine transformation matrix using the least squares method, and calculate the deformation parameters according to the affine transformation matrix. The deformation parameters include scaling factor (used to describe the degree of texture stretching / compression), rotation angle (used to describe the degree of texture rotation), and shear coefficient (used to describe the degree of texture shear deformation). S325. Determine the texture deformation coefficient based on the scaling factor, rotation angle, and shearing coefficient, combined with the preset weight. When the number of matching point pairs is less than 5, correct the texture deformation coefficient based on the gray-scale variance change rate of the region. The formula for calculating the texture deformation coefficient is as follows: In the formula, C This represents the texture deformation coefficient (values ​​0-1, where 1 indicates no deformation and 0 indicates severe deformation). s Indicates the scaling factor. θ Indicates the rotation angle. kIndicates the shear coefficient. These represent the weights of the scaling factor, rotation angle, and shear coefficient, respectively. a , d , e These represent different linear deformation parameters of the texture region from the baseline state to the current state. This represents the corrected texture deformation coefficient. This represents the rate of change of the gray-scale variance of the region.

[0036] S33. Based on the mesh nodes of the real-time 3D model, combined with the absolute displacement of each key point, the deformation of the unmeasured points in the whole domain is estimated by using the Kriging interpolation method, and the interpolation results are corrected by incorporating the texture deformation coefficient (for example, enhancing the local deformation weight in the texture wrinkle area) to generate a deformation matrix covering the entire blade. Specifically, based on the mesh nodes of the real-time 3D model and combined with the absolute displacement of each key point, the deformation of unmeasured points across the entire domain is estimated using the Kriging interpolation method. Texture deformation coefficients are then incorporated to correct the interpolation results, generating a deformation matrix covering the entire blade. The process includes the following steps: S331. Extract the mesh node set of the real-time 3D model and select preset key points as sample points with known displacement, ensuring that the sample points are evenly distributed throughout the entire blade (e.g., the tip, middle, and root of the blade each occupy 1 / 3). S332. Fit the variogram function based on the spatial distance of the sample points, and construct the Kriging equation system based on the calculation results of the variogram function; solve the weight coefficients through the Kriging equation system, and estimate the initial deformation of each grid node based on the weight coefficients. In this embodiment, the spatial distance between sample points is the basic data, which can quantify the spatial positional relationship between sample points. Based on this positional relationship, a variogram is fitted to reveal the spatial correlation of the deformation of sample points (i.e., sample points that are closer together have more similar deformation characteristics). This spatial correlation is the core basis for constructing the Kriging equations. The coefficients of the equations are calculated from the variogram, and the goal of solving the equations is to obtain the weighting coefficients. These coefficients can reflect the degree of influence of each sample point on the deformation of unknown grid nodes. By using the obtained weighting coefficients to perform a weighted summation of the known displacements of the sample points, the initial deformation of each grid node can be estimated. Since the entire process is based on spatial correlation, the final estimated global deformation can naturally reflect the continuous trend of blade deformation in space (e.g., the gradual change from the blade root to the blade tip).

[0037] S333. Obtain the texture region and texture deformation coefficient corresponding to each grid node in the adaptive image, calculate the correction weight based on the texture deformation coefficient, and correct the initial deformation amount through the correction weight to obtain the corrected deformation amount. S334. Integrate the deformation amounts after correction of all mesh nodes to generate a deformation matrix covering the entire blade; The expression for the corrected deformation is as follows: In the formula, Z j Indicates the first j The deformation amount after correction for each sample point Indicates the first j The initial deformation of each sample point Indicates the correction strength. Indicates the first j Corrected weights for each sample point K Indicates the number of feature points. Indicates the weighting coefficient. Indicates the first i The absolute displacement of a sample point in the Z direction. The weighting factor represents the texture deformation coefficient. Indicates the first j Texture deformation coefficients of each sample point h Indicates the spatial distance between sample points. This represents the distance attenuation coefficient.

[0038] S34. Map the deformation matrix to a color gradient to generate an initial global deformation heatmap. Then, calibrate the color mapping accuracy of the initial global deformation heatmap by comparing the deformation areas (e.g., the bending segments visible to the naked eye) in the adaptive image to obtain the global deformation heatmap of the wind turbine blade.

[0039] S4. Based on high-frequency vibration signals, dynamic attractor sequences are constructed by combining a sliding window to update the embedding dimension and delay time in real time, and multi-scale fingerprint features are extracted through the dynamic attractor sequences. The process of constructing a dynamic attractor sequence based on high-frequency vibration signals, combined with a sliding window that updates the embedding dimension and delay time in real time, and extracting multi-scale fingerprint features using the dynamic attractor sequence includes the following steps: S41. The high-frequency vibration signal is segmented and processed. A sliding window is set (the window size is dynamically adjusted with the blade rotation speed, for example, the window is 1 second when the rotation speed is 10 r / min, with a step size of 0.2 seconds). The signal in each window is denoised (wavelet soft threshold denoising) and standardized (zero mean). The phase space reconstruction technology is used to determine the real-time optimal delay time, and the embedding dimension is updated by combining the false nearest neighbor method. In this embodiment, for each window's sub-signal, the optimal delay time (determined by the minimum mutual information entropy) is calculated in real time using the CC algorithm (i.e., a method for determining phase space reconstruction parameters in a time series). The embedding dimension is updated in conjunction with the false nearest neighbor method (ensuring that attractors in the phase space do not fold), so that the embedding parameters (delay time, dimension) are adaptively adjusted according to the dynamic characteristics of the signal, ensuring the accuracy of phase space reconstruction.

[0040] S42. Using the updated window embedding dimension, the one-dimensional sub-signal is reconstructed into an attractor in a high-dimensional phase space (e.g., 5-7 dimensions), and the attractors of the continuous window are arranged according to the time order to form a dynamic attractor sequence. In the dynamic attractor sequence, each attractor corresponds to the local dynamic characteristics of the vibration signal, reflecting the temporal evolution of the micro-deformation of the wind turbine blade. S43. Based on the dynamic attractor sequence, the maximum Lyapunov exponent (reflecting the degree of chaos) and fractal dimension (describing geometric complexity) are extracted at the global scale. The density entropy (point distribution uniformity) and local wrinkle curvature (surface irregularity) of the attractors are calculated through a sliding window to obtain multi-scale fingerprint features. The formula for calculating the curvature of local folds is as follows: In the formula, Indicates the curvature of a local fold. M Indicates the number of phase points. X k,i Indicates the first k The first window i One phase point, , denoted as the first and second derivative moduli of the phase point, respectively.

[0041] S5. Based on the global deformation heat map and the curvature change, determine the degree of deformation of the wind turbine blades. Based on the comparison results of multi-scale fingerprint features and preset fingerprint thresholds, locate the micro-deformation area and realize damage early warning.

[0042] The process of determining the degree of deformation of the wind turbine blades based on the global deformation heat map and curvature changes, and locating micro-deformation areas and achieving damage early warning based on the comparison results of multi-scale fingerprint features and preset fingerprint thresholds, includes the following steps: S51. Extract the deformation values ​​of each region based on the global deformation heat map, calculate the local wrinkle curvature of the real-time three-dimensional model and the benchmark model, and determine the macroscopic deformation level of the wind turbine blades based on the deformation and local wrinkle curvature. Specifically, when the deformation is less than 10 mm and the local wrinkle curvature is less than 0.01... -1 When the deformation is less than 30mm or 10mm, it is considered a minor deformation; when the deformation is less than 30mm or 0.01mm, it is considered a minor deformation. -1≤Local wrinkle curvature <0.05 -1 When the deformation is greater than 30 mm or the local wrinkle curvature is greater than 0.05, it is judged as moderate deformation; when the deformation is greater than 30 mm or the local wrinkle curvature is greater than 0.05, it is judged as moderate deformation. -1 At that time, it was determined to be severely deformed.

[0043] S52. Compare the multi-scale fingerprint feature sequence with the preset fingerprint threshold. When any multi-scale fingerprint feature sequence exceeds the window of the preset fingerprint threshold, locate the micro-deformation area of ​​the corresponding wind turbine blade through the mapping relationship between the sliding window and the spatial position of the wind turbine blade. When the threshold is triggered by three consecutive windows, it is determined to be a potential damage and a micro-deformation warning area is generated. S53. Associate macroscopic deformation level with micro-deformation warning area. When a severe deformation area overlaps with a micro-deformation warning area, it is marked as a high-risk area. When a slightly deformed area has a micro-deformation warning, it is marked as a region to be reviewed. The final output includes a comprehensive result that includes deformation degree classification, risk area location and warning level.

[0044] To better understand the above-mentioned technical solutions of the present invention, the following description is provided in conjunction with specific application scenarios: Scenario Background: A coastal wind farm with an installed capacity of 50MW comprises 25 2MW wind turbines with blades 56.7m in length, made of fiberglass composite material, and has been in operation for 3 years. Recently, the SCADA system showed abnormal vibration amplitude at the blade tip of turbine #12 (15% higher than the historical average), suggesting possible deformation. This technical solution was used to conduct a detailed inspection of the blades of turbine #12. The inspection took place at 10:00 AM on [Date] (Weather: Sunny, Wind speed 6m / s, Light intensity 8000-12000 lux, Blade tip facing the sun, Blade root in the shade).

[0045] Step 1: Multi-source data acquisition: Equipment and Flight Path Planning: The UAV is equipped with a 16-line lidar (point cloud density 220 points / m²), a 20-megapixel anti-reflective camera, a GNSS-RTK module (positioning accuracy 3cm), and a battery life of 20 minutes (enough for single blade inspection). The flight path is designed as three circular tracks (5-8m from the blade) corresponding to the leading edge, middle, and trailing edge. The intersecting flight paths are set at the blade tip (56.7m), 28.35m (1 / 2 blade length), and 14.18m (1 / 4 blade length), with an overlap rate of 85%. The lidar sampling frequency is 10Hz, and the camera triggers shooting every 0.5 seconds.

[0046] Adaptive Image Acquisition: The initial low-resolution image was segmented into four regions using the Canny operator: leaf tip (50-56.7m), leaf middle (28-50m), upper leaf root (14-28m), and lower leaf root (0-14m). Gray-scale Mean Detection: Gray-scale value of 220 for the leaf tip region (strong light, level 5), 45 for the lower leaf root region (weak light, level 1), and 120 for the leaf middle region (medium light, level 3). Exposure Parameter Matching: Leaf tip (shutter speed 1 / 1600s, ISO 200), lower leaf root (shutter speed 1 / 150s, ISO 600), leaf middle (shutter speed 1 / 500s, ISO 200). Feedback Calibration: The initial image of the leaf tip had a gray-scale value of 235 (30% overexposure), the shutter speed was finely adjusted to 1 / 2000s, and the gray-scale value was 190 after reshooting (target range 80-200); the stitched image was seamless, and the glue joint texture of the leaf tip was clear.

[0047] Vibration signal acquisition: A 10kHz high-frequency vibration sensor was installed at the root of the blade to continuously acquire signals for 15 minutes (including 5 rotation cycles). After wavelet threshold denoising and frequency band filtering (retaining 2000-5000Hz, sample entropy 0.85), a clean signal was obtained.

[0048] Step 2: 3D Modeling and Registration A mapping was established using camera extrinsic parameters (focal length 12mm, distortion coefficient k1=-0.01), binding image pixel colors to point clouds (3D coordinates X / Y / Z correspond to pixel u / v); a Poisson surface reconstruction was used to generate a mesh model (800,000 triangle faces), and texture UV mapping preserved details of the blade tip coating wrinkles and leading edge adhesive joints; the baseline model was the design CAD point cloud (error ≤0.1mm), from which feature point sets were extracted: flange center (origin), blade tip apex (P0), and 28.35m leading edge inflection point (P1). After 5 iterations of the ICP algorithm, the root mean square error was 0.3mm (areas exceeding 1mm accounted for 3%), indicating effective registration; Step 3: Deformation Calculation and Thermographic Diagram: Key feature point displacements: Blade tip apex (P0), real-time coordinates (12.35, 4.22, 56.68), reference coordinates (12.30, 4.20, 56.70), displacement Δ = 0.057m (57mm). 28.35m leading edge (P1), displacement 18mm; 14.18m trailing edge (P2), displacement 6mm; root flange (P3), displacement 1mm.

[0049] Texture deformation coefficient: Image area corresponding to leaf tip P0 (41×41 pixels), baseline texture spacing 0.5mm / pixel, currently stretched to 0.6mm / pixel, scaling factor. s =1.2; Affine matrix a =1.18, shear coefficient k=0.07. Texture deformation coefficient C=0.74 (medium deformation).

[0050] Deformation matrix and heat map: Variation function nugget value c0=0.01, sill value c=0.05, range a =5m; weighting coefficient , , Corrected deformation amount, blade tip area Z j =Initial value × (1 + (1 - 0.7) × 0.1) = 57 × 1.03 = 58.7 mm; Heat map: The area at the leaf tip 50-56.7 m is orange-red (50-60 mm), 28-50 m is yellow (10-30 mm), and 0-28 m is blue (<10 mm).

[0051] Step 4: Chaotic fingerprint feature extraction: Sliding window: blade rotation speed 12 r / min (cycle 5s), window length 2s (covering 2 / 5 cycles), step size 0.4s, generating a total of 225 windows.

[0052] Embedding parameters: optimal delay time τ=0.02s (minimum mutual information entropy), embedding dimension d=5 (FNN rate 3.2%<5%).

[0053] Multiscale features: Leaf tip corresponding window (windows 180-200): maximum Lyapunov index 0.09 (0.08 above the threshold), fractal dimension 2.38 (healthy value 2.3±0.05), local wrinkle curvature κ=0.62 (0.5 above the threshold).

[0054] Step 5: Degree of Deformation and Early Warning Macroscopic grade: Blade tip deformation 58.7mm > 30mm, curvature difference It was determined to be severely deformed.

[0055] Micro-deformation warning: Three consecutive window features (180-182) exceed the threshold, locating the 50-56.7m region at the leading edge of the blade tip (overlapping with macro-deformation).

[0056] Results output: Marked as a high-risk area, it is recommended to shut down the machine for testing within 48 hours, focusing on checking for micro-cracks in the blade tip fiber layer and peeling of the adhesive surface.

[0057] According to another embodiment of the present invention, a wind turbine blade deformation degree detection system is provided, including a data acquisition module, a real-time three-dimensional model generation module, a global deformation heat map generation module, a multi-scale fingerprint feature extraction module, and a deformation degree determination module; The data acquisition module is used to control the UAV to fly autonomously according to the preset flight route, and to acquire lidar point cloud data, adaptive images and high-frequency vibration signals of the wind turbine blades during the flight. The real-time 3D model generation module is used to fuse lidar point cloud data and adaptive images to generate a real-time 3D model of the wind turbine blades, and to align the real-time 3D model of the wind turbine blades with the reference model through data registration. The global deformation heat map generation module is used to calculate the displacement of preset key feature points based on the real-time 3D model and benchmark model of the aligned wind turbine blades, and generate a global deformation heat map by combining the texture deformation in the adaptive image. The multi-scale fingerprint feature extraction module is used to construct a dynamic attractor sequence based on high-frequency vibration signals, combined with a sliding window to update the embedding dimension and delay time in real time, and to extract multi-scale fingerprint features through the dynamic attractor sequence. The deformation degree determination module is used to determine the deformation degree of the wind turbine blades based on the global deformation heat map and the curvature change. Based on the comparison results of multi-scale fingerprint features and preset fingerprint thresholds, it locates micro-deformation areas and realizes damage early warning.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method of detecting a degree of deformation of a fan blade, characterized by, The method comprises the following steps: S1, controlling the unmanned aerial vehicle to fly autonomously according to a preset flight route, and acquiring laser radar point cloud data, adaptive images and high-frequency vibration signals of the fan blade during the flight; S2, fusing the laser radar point cloud data and the adaptive images to generate a real-time three-dimensional model of the fan blade, and aligning the real-time three-dimensional model of the fan blade with a reference model through a data registration method; S3, calculating the displacement of preset key feature points according to the real-time three-dimensional model and the reference model of the fan blade after alignment, and generating a global deformation heat map in combination with the texture deformation in the adaptive images; S4, based on the high-frequency vibration signals, combining a sliding window to update the embedding dimension and the delay time in real time, constructing a dynamic attractor sequence, and extracting multi-scale fingerprint features through the dynamic attractor sequence; S5, determining the deformation degree of the fan blade according to the global deformation heat map in combination with the curvature change, positioning the micro-deformation area based on the comparison result of the multi-scale fingerprint features and the preset fingerprint threshold, and realizing damage warning.

2. The method of claim 1, wherein The method comprises the following steps: S11, planning a flight route of the unmanned aerial vehicle according to the fan blade situation of the target area, controlling the unmanned aerial vehicle to fly autonomously based on the flight route, and acquiring laser radar point cloud data of the fan blade through a laser radar during the flight; S12, identifying the illumination intensity of different regions of the fan blade by using a camera, and determining the corresponding exposure parameters according to the illumination intensity of different regions; based on the determined exposure parameters, acquiring adaptive images of the fan blade by using the camera; S13, continuously collecting high-frequency vibration signals of the fan blade in a preset time period by using a vibration sensor installed at the root of the fan blade, and performing wavelet threshold denoising processing on the high-frequency vibration signals; processing the laser radar point cloud data and the adaptive images to ensure that the texture corresponds to the point cloud.

3. The method of claim 2, wherein The method comprises the following steps: S121, acquiring an initial low-resolution image of the fan blade by using the camera in real time, identifying the outline of the fan blade by using an edge detection algorithm, and dividing the fan blade into a plurality of regions according to geometric features; S122, extracting the pixel gray mean value of each region, and converting the pixel gray mean value into an actual illumination level according to the mapping relationship between the gray and the illumination intensity; S123, determining the exposure parameters of different regions of the surface of the fan blade according to a preset mapping table of illumination levels and exposure parameters; S124, based on the exposure parameters of different regions, using the camera to shoot single-frame images of the corresponding regions, splicing the images of each region, and eliminating the splicing seams caused by parameter differences through an edge fusion algorithm to generate a global clear adaptive image of the fan blade. The method comprises the following steps:

4. The method of claim 1, wherein ​ S21, establish a mapping relationship between point cloud coordinates and image pixel coordinates according to camera external parameters and laser radar calibration parameters, and associate pixel colors of the adaptive image with three-dimensional points of the point cloud through the mapping relationship to realize binding of geometric coordinates and texture information; S22, generate an initial mesh model of the fan blade by using a Poisson surface reconstruction algorithm in combination with the associated point cloud, and map the texture of the adaptive image to the surface of the initial mesh model to form a real-time three-dimensional model with texture, thereby obtaining the real-time three-dimensional model of the fan blade; S23, obtain a three-dimensional CAD model of the fan blade in the design stage and convert it into a point cloud format to obtain a reference model of the fan blade, and extract key geometric features as a registration reference point set; and realize alignment of the real-time three-dimensional model and the reference model through a data registration method based on the registration reference point set.

5. The method of claim 1, wherein The displacement of the preset key feature points is calculated according to the aligned real-time three-dimensional model and the reference model of the fan blade, and a global deformation heat map is generated in combination with the texture deformation in the adaptive image, including the following steps: S31, extract the preset key points from the aligned real-time three-dimensional model and the reference model of the fan blade, calculate the coordinate difference of the corresponding points in the three-dimensional coordinate system, and determine the absolute displacement of each key point based on the Euclidean distance; S32, perform texture comparison on the region corresponding to the preset key points in the adaptive image, and identify the deformation degree of the texture in combination with a feature matching algorithm, and quantize the deformation degree of the texture into a texture deformation coefficient; S33, based on the grid nodes of the real-time three-dimensional model, in combination with the absolute displacement of each key point, the deformation of the global unmeasured points is estimated by using the Kriging interpolation method, and the interpolation result is corrected by integrating the texture deformation coefficient to generate a deformation matrix covering the whole blade; S34, map the deformation matrix to a color gradient to generate an initial global deformation heat map, and calibrate the color mapping accuracy of the initial global deformation heat map by comparing the deformation region in the adaptive image to obtain the global deformation heat map of the fan blade.

6. The method of claim 5, wherein The texture comparison on the region corresponding to the preset key points in the adaptive image, and the identification of the deformation degree of the texture in combination with the feature matching algorithm, and the quantization of the deformation degree of the texture into a texture deformation coefficient include the following steps: S321, based on the camera intrinsic matrix, project the three-dimensional coordinates of the preset key points in the real-time three-dimensional model into pixel coordinates of the adaptive image, and take the projected pixel coordinates of the adaptive image as the center to intercept a square region with a predetermined side length as a texture analysis window; S322, extract the region corresponding to the current window position from the historical adaptive image corresponding to the reference model as a reference texture region, and extract the texture analysis window region determined in step S321 from the current adaptive image as a current texture region; S323, extract feature points of the reference texture region and the current texture region by using a feature matching algorithm, and each feature point is represented by a plurality of feature point descriptors; calculate the similarity of the feature point descriptors by the Euclidean distance, and screen the matching point pairs according to the similarity calculation result; S324, based on the screened matching point pairs, an affine transformation matrix is fitted by using a least square method, and deformation parameters are calculated according to the affine transformation matrix, wherein the deformation parameters include a scaling factor, a rotation angle and a shearing coefficient; S325, according to the scaling factor, the rotation angle and the shearing coefficient, a texture deformation coefficient is determined in combination with a preset weight, and when the number of matching point pairs is less than a preset threshold, the texture deformation coefficient is corrected in combination with a regional gray variance change rate; wherein, the calculation formula of the texture deformation coefficient is: wherein, C denotes a texture deformation coefficient, s denotes a scaling factor, θ denotes a rotation angle, k denotes a shear coefficient, denotes a weight of the scaling factor, the rotation angle and the shear coefficient, respectively, a , d , e denotes a different linear deformation parameter of the texture region from a reference state to a current state, respectively, denotes a corrected texture deformation coefficient, denotes a region gray scale variance change rate.

7. The method of claim 5, wherein the method further comprises: The grid nodes of the real-time three-dimensional model are taken as the basis, the absolute displacement of each key point is combined, the Kriging interpolation method is used to estimate the deformation of the global unmeasured points, the interpolation result is modified by integrating the texture deformation coefficient, and a deformation matrix covering the whole blade is generated, including the following steps: S331, the grid node set of the real-time three-dimensional model is extracted, and a preset key point is selected as a sample point with known displacement; S332, a variogram is fitted based on the spatial distance of the sample points, and a Kriging equation set is constructed according to the calculation result of the variogram; the weight coefficient is solved by the Kriging equation set, and the initial deformation of each grid node is estimated according to the weight coefficient; S333, the texture region corresponding to each grid node in the adaptive image and the texture deformation coefficient are obtained, the correction weight is calculated according to the texture deformation coefficient, and the initial deformation is corrected by the correction weight to obtain the corrected deformation; S334, the corrected deformation of all grid nodes is integrated to generate a deformation matrix covering the whole blade; wherein, the expression of the corrected deformation is: In the formula, Z j Indicates the first j The deformation amount after correction for each sample point Indicates the first j The initial deformation of each sample point Indicates the correction strength. Indicates the first j Corrected weights for each sample point K Indicates the number of feature points. Indicates the weighting coefficient. Indicates the first i The absolute displacement of a sample point in the Z direction. The weighting factor represents the texture deformation coefficient. Indicates the first j Texture deformation coefficients of each sample point h Indicates the spatial distance between sample points. This represents the distance attenuation coefficient.

8. The method of claim 1, wherein Based on the high-frequency vibration signal, the embedding dimension and the delay time are updated in real time by combining the sliding window, a dynamic attractor sequence is constructed, and multi-scale fingerprint features are extracted from the dynamic attractor sequence, including the following steps: S41, the high-frequency vibration signal is segmented, a sliding window is set, and the signal in each window is denoised and standardized; the real-time optimal delay time is determined by using phase space reconstruction technology, and the embedding dimension is updated by combining the false neighbor point method; S42, the updated window embedding dimension is used to reconstruct one-dimensional sub-signals into attractors in a high-dimensional phase space, and the attractors of consecutive windows are arranged according to time sequence to form a dynamic attractor sequence, wherein each attractor in the dynamic attractor sequence corresponds to the local dynamic characteristics of the vibration signal, reflecting the time sequence evolution of the fan blade micro-deformation; S43, based on the dynamic attractor sequence, the maximum Lyapunov exponent and the fractal dimension are calculated in the global scale, and the density entropy and the local fold curvature of the attractor are calculated by the sliding window to obtain the multi-scale fingerprint features; wherein, the calculation formula of the local fold curvature is: In the formula, represents the local wrinkle curvature, M represents the number of phase points, X k,i represents the first k phase point of the first i window, , respectively represent the first-order and second-order derivative modules of the phase point.

9. The method of claim 1, wherein According to the global deformation thermal map, the deformation degree of the fan blade is determined by combining the curvature change, and based on the comparison result of the multi-scale fingerprint features and the preset fingerprint threshold, the micro-deformation area is located and damage warning is realized, including the following steps: S51, the deformation values of each region are extracted based on the global deformation thermal map, the local fold curvature of the real-time three-dimensional model and the reference model is calculated, and the macro-deformation grade of the fan blade is determined according to the deformation and the local fold curvature; S52, compare the multi-scale fingerprint feature sequence with the preset fingerprint threshold, when any multi-scale fingerprint feature sequence exceeds the window of the preset fingerprint threshold, then locate the micro-deformation area of the corresponding fan blade through the mapping relationship between the sliding window and the spatial position of the fan blade, when a continuous preset number of windows trigger the threshold, then determine a potential damage and generate a micro-deformation warning area; S53, associate the macro-deformation level with the micro-deformation warning area, when the severe deformation area and the micro-deformation warning area overlap, mark as a high-risk area, when there is a slight deformation but a micro-deformation warning area, mark as a to-be-reviewed area, and finally output a comprehensive result including the deformation degree classification, risk area positioning and warning level.

10. A wind turbine blade deformation detection system for implementing the steps of the wind turbine blade deformation detection method of any of claims 1-9, characterized by, The system comprises a data acquisition module, a real-time three-dimensional model generation module, a global deformation heat map generation module, a multi-scale fingerprint feature extraction module and a deformation degree determination module. The data acquisition module is configured to control the unmanned aerial vehicle to fly autonomously according to a preset flight route, and acquire laser radar point cloud data, adaptive images and high-frequency vibration signals of the fan blade during flight. The real-time three-dimensional model generation module is configured to fuse the laser radar point cloud data and the adaptive images to generate a real-time three-dimensional model of the fan blade, and align the real-time three-dimensional model of the fan blade with a reference model through a data registration method. The global deformation heat map generation module is configured to calculate the displacement of preset key feature points according to the aligned real-time three-dimensional model and the reference model of the fan blade, and generate a global deformation heat map in combination with the texture deformation in the adaptive images. The multi-scale fingerprint feature extraction module is configured to update the embedding dimension and the delay time in real time based on the high-frequency vibration signals, construct a dynamic attractor sequence, and extract multi-scale fingerprint features through the dynamic attractor sequence. The deformation degree determination module is configured to determine the deformation degree of the fan blade according to the global deformation heat map in combination with the curvature change, locate the micro-deformation area and realize damage warning based on the comparison result of the multi-scale fingerprint features and the preset fingerprint threshold.

Images