Intelligent judgment system for damage of fan blade

Abstract

The invention provides a fan blade damage intelligent judgment system, which comprises an image acquisition module used for acquiring image data of a fan blade; the image processing module is used for preprocessing the image data to generate a standardized image; the damage analysis module is used for extracting defect features of the fan blade from the standardized image, positioning and classifying blade damage based on the defect features, and generating a judgment result containing a damage position and a damage type; the man-machine interaction module is used for displaying the image data and the judgment result on a display interface; according to the judgment result, the damage position is identified in the video picture in the form of the marking box, and the damage type is displayed in an associated mode, so that the efficiency and accuracy of damage detection are remarkably improved, the labor intensity and the misjudgment rate of manual detection are reduced, and the safe operation level and the operation and maintenance efficiency of the wind turbine generator are improved.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for wind power generation equipment, and in particular to an intelligent judgment system for wind turbine blade damage. Background Technology

[0002] As a key piece of equipment in clean energy, the long-term stable operation of wind turbines is crucial for ensuring power generation efficiency. Wind turbine blades, as the core components for capturing wind energy, are constantly exposed to a complex and ever-changing natural environment. They are inevitably affected by various factors such as lightning strikes, rain erosion, icing, and fatigue loads, leading to various types of damage on their surface, including cracks, corrosion, delamination, and perforation. If these damages are not detected and addressed in a timely manner, they will gradually expand, not only significantly reducing power generation efficiency but also potentially causing catastrophic accidents such as blade breakage, resulting in huge economic losses and safety risks.

[0003] Currently, the inspection and damage identification of wind turbine blades mainly rely on manual telescope observation or close-range inspection at high altitudes. This method is not only inefficient and costly, but also greatly affected by personnel experience and subjective factors, which can easily lead to missed inspections and misjudgments. It is difficult to meet the needs of modern wind farms for large-scale, high-frequency intelligent operation and maintenance. Summary of the Invention

[0004] This invention provides an intelligent judgment system for wind turbine blade damage, which solves the technical problems of low efficiency and insufficient accuracy in wind turbine blade damage detection in the prior art.

[0005] On one hand, the present invention provides an intelligent judgment system for wind turbine blade damage, comprising:

[0006] The image acquisition module is used to acquire image data of the wind turbine blades;

[0007] The image processing module is used to preprocess image data and generate standardized images;

[0008] The damage analysis module is used to extract defect features of wind turbine blades from standardized images, and to locate and classify blade damage based on the defect features, generating a judgment result that includes the location and type of damage.

[0009] The human-computer interaction module is used to display image data and judgment results on the display interface; the judgment results are marked in the video screen in the form of annotation boxes to identify the damage location and display the damage type.

[0010] According to the present invention, an intelligent judgment system for wind turbine blade damage includes a damage analysis module comprising:

[0011] The feature extraction unit is used to perform multi-scale morphological filtering on the standardized image to extract strip-shaped and patch-shaped abnormal regions respectively.

[0012] The crack screening unit is used to screen out strip-shaped abnormal regions whose directions deviate from the blade texture direction as candidate crack regions based on the directional field analysis of the blade texture.

[0013] The surface damage screening unit is used for consistency analysis based on surface texture to screen out patchy abnormal regions with texture entropy exceeding a preset threshold as candidate surface damage regions.

[0014] The damage determination unit is used to determine the candidate crack region and the candidate surface damage region as real crack damage and real surface damage respectively if the position offset and shape change rate of the candidate crack region and the candidate surface damage region remain within their respective preset threshold ranges within N consecutive frames; N is an integer greater than 1.

[0015] According to the intelligent judgment system for wind turbine blade damage provided by the present invention, the crack screening unit is further used for:

[0016] Determine the dominant local texture direction in the leaf region of the standardized image;

[0017] Extract the skeleton line of the strip-shaped anomaly region and determine the main direction of the skeleton line;

[0018] Determine the angle between the main direction of the skeleton line and the local dominant texture direction of the leaf area where the skeleton line is located;

[0019] When the included angle is greater than the first angle threshold, or when the strip-shaped abnormal region crosses the macroscopic feature line on the blade structure, the strip-shaped abnormal region is identified as a candidate crack region.

[0020] Among them, macroscopic feature lines include the blade cap line or leading edge bonding line.

[0021] According to the intelligent judgment system for wind turbine blade damage provided by the present invention, the crack screening unit is further used for:

[0022] The three-dimensional structural model of the wind turbine blade is combined with the UAV pose information and projected onto the image coordinate system of the current video frame to obtain the projection range of the beam cap area;

[0023] If a candidate crack region is located within the projection range, and the angle between the main direction of its skeleton line and the direction of the beam cap is less than the second angle threshold, then the processing priority of the candidate crack region is increased.

[0024] According to the intelligent judgment system for wind turbine blade damage provided by the present invention, the dominant local texture direction of the blade region in a standardized image is determined, including:

[0025] An orientation field estimation algorithm based on gradient structure tensor is used to calculate the gradient direction and gradient magnitude of each pixel in the image.

[0026] A sliding window is set within the leaf area, and all pixels within the window are weighted and voted on based on their gradient magnitude; where pixels with larger gradient magnitudes have higher voting weights in their gradient direction.

[0027] The gradient direction distribution after all weighted votes within the sliding window is statistically analyzed, and the gradient direction that receives the highest number of weighted votes is determined as the dominant texture direction in the local region of that window.

[0028] By traversing the entire blade region through a sliding window, a mapping relationship is established between each window position and its local dominant texture direction, forming a complete blade surface orientation field distribution map.

[0029] According to the present invention, an intelligent judgment system for wind turbine blade damage extracts the skeleton line of a strip-shaped abnormal region and determines the main direction of the skeleton line, including:

[0030] Binarize the strip-shaped abnormal regions in the standardized image;

[0031] Morphological operations were used to refine the binarized strip-shaped abnormal regions and extract their skeleton lines.

[0032] Select multiple key points on the skeleton line and calculate the local tangent direction of each key point;

[0033] Statistical analysis was performed on the local tangent directions of all key points, and the principal component analysis method was used to determine the main direction of the skeleton line.

[0034] According to the intelligent judgment system for wind turbine blade damage provided by the present invention, the surface damage screening unit is further used for:

[0035] Within the patchy abnormal region, calculate the similarity of its texture features with the preset surrounding normal leaf region at multiple scales;

[0036] Determine the average lightness difference between the patchy abnormal area and the preset surrounding normal leaf area;

[0037] If the texture feature similarity is lower than the first preset threshold, the patchy abnormal region is determined as a candidate surface damage region.

[0038] For candidate surface damage areas, if the absolute value of their brightness difference is higher than the second preset threshold, they are classified as material defects.

[0039] If the absolute value of the difference in brightness is lower than the second preset threshold, it is classified as stain coverage damage.

[0040] According to the intelligent judgment system for wind turbine blade damage provided by the present invention, the damage judgment unit is further used for:

[0041] In N consecutive frames of images, spatial alignment and deformation registration are performed on the same candidate crack region or candidate surface damage region to obtain a temporal damage region sequence.

[0042] Based on the temporal damage region sequence, time series data of its location, length, width, area and shape factor are constructed;

[0043] The sliding window statistical method is used to calculate the slope of the trend and the variance of the fluctuation of time series data within the current window.

[0044] When the slope of the change trend exceeds the preset growth threshold, or the fluctuation variance remains below the preset noise level threshold, the confidence weight of the candidate region is dynamically adjusted.

[0045] If the confidence level of the candidate region is higher than the judgment threshold in N consecutive frames, it is confirmed as a real damage.

[0046] According to the intelligent judgment system for wind turbine blade damage provided by the present invention, the human-computer interaction module is further used for:

[0047] It provides a multi-mode visual interface and supports real-time inspection mode and historical playback mode;

[0048] In real-time inspection mode, the current video frame, the marked damage location box, the damage type label, and the detection timestamp are displayed simultaneously and overlaid; different types of damage are distinguished by different colored or graphic style label boxes.

[0049] In the historical playback mode, it supports searching historical inspection records by wind turbine number, blade serial number, inspection time or multi-dimensional filtering conditions, and displays the evolution process of the same damaged area in multiple inspections in the form of a timeline.

[0050] According to the present invention, an intelligent judgment system for wind turbine blade damage includes an image processing module comprising:

[0051] The illumination enhancement unit is used to improve the image contrast and sharpness of image data through an adaptive histogram equalization algorithm to obtain the processed image.

[0052] Size normalization unit is used to uniformly scale the processed image to a fixed resolution;

[0053] The pixel normalization unit is used to normalize the pixel values ​​of an image to a preset range to generate a standardized image.

[0054] The intelligent damage assessment system for wind turbine blades provided by this invention acquires video streams of wind turbine blades through an image acquisition module, standardizes the video streams through an image processing module, and accurately extracts blade defect features and locates and classifies the damage through a damage analysis module. Finally, a human-computer interaction module displays the damage location and type intuitively in the video frame using labeled boxes. This system automates and intelligently detects wind turbine blade damage, significantly improving the efficiency and accuracy of damage detection, reducing the labor intensity and error rate of manual inspection, providing an efficient and reliable detection method for the maintenance and management of wind turbine blades, and improving the safe operation and maintenance efficiency of wind turbine units. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0056] Figure 1 This is one of the structural schematic diagrams of the intelligent judgment system for wind turbine blade damage provided in the embodiments of the present invention;

[0057] Figure 2 This is the second schematic diagram of the intelligent judgment system for wind turbine blade damage provided in this embodiment of the invention;

[0058] Figure 3 This is a flowchart illustrating the intelligent method for judging wind turbine blade damage provided in an embodiment of the present invention.

[0059] Figure 4 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0061] Figure 1 This is one of the structural schematic diagrams of the intelligent judgment system for wind turbine blade damage provided in the embodiments of the present invention; Figure 2 This is the second structural schematic diagram of the intelligent judgment system for wind turbine blade damage provided in the embodiment of the present invention.

[0062] See Figure 1 and Figure 2 The intelligent judgment system 10 for wind turbine blade damage includes an image acquisition module 110, an image processing module 120, a damage analysis module 130, and a human-computer interaction module 140.

[0063] Image acquisition module 110 is used to acquire image data of wind turbine blades;

[0064] Specifically, the image acquisition module 110 can be implemented by using a drone equipped with a camera. For example, a fixed-wing drone or a rotary-wing drone can be used to carry high-definition camera equipment to take pictures of the wind turbine blades from multiple angles. The main purpose is to achieve continuous and comprehensive visual data acquisition.

[0065] Image processing module 120 is used to preprocess image data and generate standardized images;

[0066] Specifically, the preprocessing operations in the image processing module 120 may include a variety of techniques, such as improving image contrast through histogram equalization or removing noise interference through filtering algorithms, with the aim of providing stable image input for subsequent analysis.

[0067] The damage analysis module 130 is used to extract the defect features of the wind turbine blades from the standardized images, and to locate and classify the blade damage based on the defect features, generating a judgment result that includes the damage location and damage type.

[0068] Specifically, the defect feature extraction process in the damage analysis module 130 can be implemented using edge detection algorithms or region segmentation algorithms. For example, Canny edge detection can be used to extract abnormal boundaries on the blade surface, or K-means clustering can be used to segment abnormal regions in the image. The main purpose is to achieve accurate identification and classification of damage features.

[0069] The human-computer interaction module 140 is used to display image data and judgment results on the display interface; wherein, the judgment results are marked in the video screen in the form of annotation boxes to identify the damage location and display the damage type in association;

[0070] Specifically, the annotation box display function in the human-computer interaction module 140 can be achieved through graphic overlay technology, such as embedding a rectangular box in the video stream to mark the location of the damage and labeling the damage type with text labels. The purpose is to present the complex analysis results to the user in an intuitive form.

[0071] In this embodiment, the intelligent wind turbine blade damage judgment system 10 acquires video streams of the wind turbine blades through the image acquisition module 110, the image processing module 120 standardizes the video streams, and the damage analysis module 130 accurately extracts blade defect features and locates and classifies the damage. Finally, the human-computer interaction module 140 intuitively displays the damage location and type in the form of labeled boxes on the video screen. This system automates and intelligently detects wind turbine blade damage, significantly improving the efficiency and accuracy of damage detection, reducing the labor intensity and misjudgment rate of manual inspection, providing an efficient and reliable detection method for the maintenance and management of wind turbine blades, and improving the safe operation level and maintenance efficiency of wind turbine units.

[0072] In one embodiment of this specification, the damage analysis module 130 includes a feature extraction unit 131, a crack screening unit 132, a surface damage screening unit 133, and a damage determination unit 134.

[0073] Feature extraction unit 131 is used to perform multi-scale morphological filtering on the standardized image to extract strip-shaped abnormal regions and patch-shaped abnormal regions respectively.

[0074] The feature extraction unit 131 can be implemented by a combination of morphological operators such as opening and closing operations. The purpose is to adapt to damage features of different sizes, enhance the ability to capture small and large abnormal regions, and thus reduce the risk of feature omission or noise interference at a single scale.

[0075] The crack screening unit 132 is used for orientation field analysis based on blade texture to screen out strip-shaped abnormal regions whose orientation deviates from the blade texture orientation as candidate crack regions.

[0076] The crack screening unit 132 can be implemented by calculating the local dominant texture direction and combining it with the deviation of the main direction of the skeleton line. The purpose is to eliminate the interference of inherent structural features such as beam cap line or leading edge bonding line, and ensure that the candidate crack area focuses on potential damage.

[0077] The surface damage screening unit 133 is used for consistency analysis based on surface texture to screen out patchy abnormal regions with texture entropy exceeding a preset threshold as candidate surface damage regions.

[0078] The surface damage screening unit 133 can calculate the texture entropy value and compare it with a preset threshold to accurately distinguish between material defects and stains, thereby reducing the possibility of misclassification.

[0079] The damage determination unit 134 is used to determine the candidate crack region and the candidate surface damage region as real crack damage and real surface damage respectively if the position offset and shape change rate of the candidate crack region and the candidate surface damage region are both kept within their respective preset threshold ranges within N consecutive frames.

[0080] The damage determination unit 134 verifies whether a candidate region is a real damage by checking the temporal stability of multiple consecutive frames of images. This can be achieved by monitoring the changing trends of position offset and shape change rate, with the aim of filtering out instantaneous noise or temporary interference and improving the reliability of the recognition results. N can be an integer greater than 1, for example, any value between 3 and 10, or can be set according to actual needs.

[0081] In this embodiment, the feature extraction unit 131 uses multi-scale morphological filtering to process the standardized image, which can adapt to damage features of different sizes and provide a purer basis for strip-shaped and patchy abnormal regions for subsequent screening. The crack screening unit 132 screens candidate crack regions based on the directional field analysis of the blade texture. Since the surface texture of the wind turbine blade has an inherent dominant direction, the real crack usually shows a significant deviation from the texture direction. Therefore, by analyzing the directional field deviation, the interference of inherent features such as blade structural lines can be effectively eliminated. The surface damage screening unit 133 screens candidate surface damage regions based on the consistency analysis of the surface texture. The texture entropy value of the real surface damage region is naturally higher than that of the normal region due to the texture disorder caused by material defects. However, interference such as stains often maintain local texture consistency. By setting an entropy threshold, material defect damage and stain coverage damage can be accurately distinguished. The damage determination unit 134 determines the true damage through stability verification over N consecutive frames. True damage has a relatively stable position and shape in the video sequence, while noise or temporary interference changes rapidly with each frame. By monitoring the persistence of position offset and shape change rate within a threshold range, transient noise can be effectively filtered out, ensuring the temporal reliability of the identification results. This embodiment significantly improves the accuracy and reliability of wind turbine blade damage identification by constructing a multi-level screening and verification mechanism.

[0082] In one embodiment of this specification, the crack screening unit 132 is further configured to:

[0083] Step 1: Determine the dominant local texture direction in the leaf region of the standardized image;

[0084] Among them, the local dominant texture direction refers to the texture arrangement direction that has statistical significance in a certain area of ​​the blade surface, which can be realized by the orientation field estimation algorithm based on the gradient structure tensor.

[0085] Step 2: Extract the skeleton line of the strip-shaped abnormal region and determine the main direction of the skeleton line;

[0086] Among them, the skeleton line refers to the geometric representation of the central axis of the strip-shaped abnormal region. It can be extracted by morphological refinement operation and the main direction can be determined by statistical analysis of the tangent direction of key points.

[0087] Step 3: Determine the angle between the main direction of the skeleton line and the local dominant texture direction of the leaf area where the skeleton line is located;

[0088] Step 4: When the included angle is greater than the first angle threshold, or when the strip-shaped abnormal region crosses the macroscopic feature line on the blade structure, the strip-shaped abnormal region is identified as a candidate crack region.

[0089] Among them, macroscopic feature lines include the blade's cap line or leading-edge bonding line. Macroscopic feature lines are the boundary lines of stress concentration areas in the blade's inherent structure, such as the cap line or leading-edge bonding line, which can be accurately located by projecting the three-dimensional structural model and aligning it with the image coordinate system.

[0090] In this embodiment, a sliding window is set within the blade region, and a gradient magnitude weighted voting mechanism is used to dynamically capture the local dominant texture direction. This process can adapt to the local variation characteristics of the blade surface texture and avoid the rigidity of global orientation field analysis. Skeleton lines are extracted by binarizing and morphologically refining the strip-shaped anomaly region, and their principal direction is quantified using principal component analysis, thus providing a reliable basis for subsequent orientation deviation calculation. By calculating the angle between the principal direction of the skeleton line and the local dominant texture direction, the true crack and naturally varied texture regions are effectively distinguished. A macroscopic feature line check is introduced as a supplementary judgment condition. When the strip-shaped anomaly region crosses stress concentration areas such as the beam cap line or leading edge bonding line, candidate judgment can be triggered even if the orientation deviation does not exceed the threshold. This dual verification mechanism ensures the reliability of the screening results.

[0091] In one embodiment of this specification, the crack screening unit 132 is further configured to:

[0092] Step 1: Combine the 3D structural model of the wind turbine blade with the UAV pose information and project it onto the image coordinate system of the current video frame to obtain the projection range of the beam cap area;

[0093] The 3D structural model of the wind turbine blade refers to the digital representation of the blade's geometry generated through precise modeling technology. This can be achieved using point cloud-based 3D reconstruction methods or parametric modeling methods. The UAV pose information can be understood as the UAV's position and attitude parameters in space, including longitude, latitude, altitude, pitch angle, yaw angle, and roll angle. Its purpose is to provide real-time spatial positioning data to support accurate mapping. The projection range of the beam cap region refers to the pixel area covered after mapping the 3D beam cap region onto a 2D image plane through geometric transformation. Its purpose is to provide a reliable reference area for subsequent crack screening.

[0094] Step 2: If the candidate crack region is located within the projection range and the angle between the main direction of its skeleton line and the direction of the beam cap is less than the second angle threshold, then the processing priority of the candidate crack region is increased.

[0095] Among them, candidate crack regions refer to strip-shaped anomaly regions that are marked as potentially containing cracks after preliminary screening, and can be extracted through multi-scale morphological filtering. The main direction of the skeleton line refers to the main extension direction of the skeleton line in the strip-shaped anomaly region, which can be calculated using principal component analysis. The second angle threshold can be understood as a preset angle range used to measure the degree of deviation between the main direction of the skeleton line and the direction of the beam cap, and its purpose is to improve the accuracy of crack identification through directional consistency verification.

[0096] In this embodiment, the 3D structural model of the wind turbine blades and the UAV pose information are used to project the beam cap region from 3D space onto the 2D image of the current video frame, thereby generating a stable and reliable projection range for the beam cap region. This process avoids the blurring or missing problems caused by perspective distortion or environmental interference when relying solely on image texture to extract feature lines. Through dual verification of the candidate crack region's position matching and orientation consistency, it is ensured that the candidate region is both located near the critical beam cap structure and conforms to the physical law of crack propagation along the structural direction. When a candidate crack region meets the above conditions, the system will increase its processing priority, making resource allocation more efficient and optimizing the reliability of damage determination decisions.

[0097] In one embodiment of this specification, determining the local dominant texture direction of a leaf region in a standardized image includes:

[0098] Step 1: Use the orientation field estimation algorithm based on gradient structure tensor to calculate the gradient direction and gradient magnitude of each pixel in the image;

[0099] The orientation field estimation algorithm based on gradient structure tensor is a method for extracting local geometric features by constructing the gradient covariance matrix within the pixel neighborhood. It can be implemented using gradient calculation tools such as the Sobel operator and the Scharr operator. The purpose of introducing this algorithm is to suppress random noise and enhance the geometric characteristics of continuous textures, thereby avoiding orientation drift problems caused by uneven lighting or small perturbations.

[0100] Step 2: Set up a sliding window within the leaf area, and perform weighted voting on all pixels within the window based on their gradient magnitude; where pixels with larger gradient magnitudes have higher voting weights in their gradient direction.

[0101] The sliding window refers to a rectangular area that moves across the image in fixed steps. Its size can be dynamically adjusted according to the complexity of the blade surface texture, such as 16×16 or 32×32. The purpose of the sliding window design is to focus on the texture characteristics of a local area while reducing global noise interference.

[0102] Step 3: Statistically analyze the gradient direction distribution after all weighted votes within the sliding window, and determine the gradient direction that receives the highest weighted votes as the dominant texture direction for the local region of that window.

[0103] The weighted voting mechanism refers to a statistical method that assigns different weights based on the gradient magnitude of pixels. It can be implemented through linear or non-linear weighting, with the aim of increasing the contribution of pixels with high gradient magnitudes to the direction estimation, thereby enhancing the reliability of the direction judgment.

[0104] Step 4: Traverse the entire blade area through a sliding window to establish the mapping relationship between each window position and its local dominant texture direction, forming a complete blade surface orientation field distribution map.

[0105] In this embodiment, the orientation field estimation algorithm based on gradient structure tensor calculates the gradient direction and magnitude of each pixel, providing a high-precision orientation data foundation for subsequent processing. A sliding window is set within the blade region, and gradient magnitude weighted voting is applied to the pixels within the window. This fully utilizes the characteristic that texture edge regions typically have significant gradient magnitudes, ensuring that orientation estimation focuses on reliable feature points while effectively suppressing interference from low-magnitude noise pixels. Statistical analysis of the gradient direction distribution after weighted voting is performed, and the direction with the highest weighted votes is selected as the dominant texture direction of the window. This avoids the distortion of results by outliers in the average orientation calculation, ensuring the consistency of orientation in local regions. By traversing the entire blade region through the sliding window to construct an orientation field distribution map, a texture orientation mapping from local to global is achieved, overcoming the shortcomings of global methods that cannot adapt to gradual changes in blade surface texture or abrupt structural changes.

[0106] In one embodiment of this specification, extracting the skeleton line of the strip-shaped abnormal region and determining the main direction of the skeleton line includes:

[0107] Step 1: Binarize the strip-shaped abnormal regions in the standardized image;

[0108] Binarization, in this context, refers to converting a grayscale image into an image format containing only black and white pixel values. Its purpose is to enhance the contrast between banded anomaly regions and the background, thereby highlighting the boundary features of the anomaly regions. In practical applications, global thresholding, adaptive thresholding, or edge detection-based segmentation algorithms can be used to achieve binarization.

[0109] Step 2: Use morphological operations (such as erosion and dilation) to refine the binarized strip-shaped anomaly region and extract its skeleton lines;

[0110] Morphological operations are an image processing method based on mathematical morphology. Their core lies in using erosion and dilation operations on images through structuring elements to adjust the shape and topological properties of target regions. Specifically, erosion gradually strips away edge pixels to reduce the region width, while dilation compensates for potential breaks caused by erosion. The combination of these two operations effectively preserves the central topological structure of strip-shaped abnormal regions.

[0111] Step 3: Select multiple key points on the skeleton line and calculate the local tangent direction of each key point;

[0112] The local tangent direction refers to the instantaneous direction at a point on the skeleton line, which can be calculated using the gradient information of neighboring pixels. The purpose is to capture the directional changes of the skeleton line at the microscale.

[0113] Step 4: Perform statistical analysis on the local tangent directions of all key points, and determine the principal direction of the skeleton line using principal component analysis (PCA).

[0114] Principal Component Analysis (PCA) is a dimensionality reduction and data analysis method that projects discrete data onto the principal component space and extracts the direction of maximum variance to determine the dominant direction of the skeleton line. This method can effectively suppress the influence of noise interference and local outliers.

[0115] In this embodiment, binarization provides a clear input basis for subsequent morphological operations, significantly enhancing the contrast between the abnormal region and the background. Morphological operations refine the strip-shaped abnormal region, ensuring the realism and continuity of the skeleton line and avoiding breakage or adhesion. By densely sampling key points on the skeleton line and calculating their local tangent directions, the directional characteristics of the skeleton line at the microscale can be accurately described, avoiding the problem of ignoring local fluctuations in a single global estimate. Principal component analysis (PCA) statistically analyzes the local tangent directions of all key points, extracting the direction with the maximum variance as the principal direction of the skeleton line, thus providing a stable and robust direction estimation method.

[0116] In one embodiment of this specification, the surface damage screening unit 133 is further configured to:

[0117] Step 1: Within the patchy abnormal area, calculate the similarity of its texture features with the preset surrounding normal leaf area at multiple scales;

[0118] Texture feature similarity refers to assessing the similarity between patchy abnormal regions and surrounding normal leaf regions by comparing the consistency of texture patterns at different scales. It can be achieved using methods such as gray-level co-occurrence matrix, local binary pattern, or wavelet transform.

[0119] Step 2: Determine the average brightness difference between the patchy abnormal area and the preset surrounding normal leaf area;

[0120] The average brightness difference can be understood as the average difference in brightness between patchy abnormal areas and normal leaf areas. It can be calculated by measuring the mean difference of pixel values ​​or the difference in histogram distribution between the two areas. The purpose of introducing texture feature similarity and average brightness difference is to improve the ability to distinguish between material defects and stain coverage through a dual verification mechanism, thereby avoiding misjudgment caused by a single indicator.

[0121] Step 3: If the texture feature similarity is lower than the first preset threshold, then the patchy abnormal area is determined as a candidate surface damage area.

[0122] Step 4: For candidate surface damage areas, if the absolute value of their brightness difference is higher than the second preset threshold, they are classified as material defects.

[0123] If the absolute value of the difference in brightness is lower than the second preset threshold, it is classified as stain coverage damage.

[0124] In this embodiment, the surface damage screening unit 133 performs multi-scale analysis on patchy abnormal areas to capture texture details at different levels, thereby reducing the impact of local noise or lighting interference. By combining cascaded judgment logic of texture feature similarity and average brightness difference, a complete chain from anomaly detection to damage classification is constructed. For example, when calculating texture feature similarity, the system can establish a reference benchmark based on the texture pattern of healthy areas to ensure the reliability of anomaly detection. By analyzing the magnitude of brightness differences, the physical nature of the damage can be effectively correlated; for example, material defects are usually accompanied by significant brightness changes, while stain coverage only causes slight brightness fluctuations.

[0125] In one embodiment of this specification, the damage determination unit 134 is further configured to:

[0126] Step 1: In N consecutive frames of images, perform spatial alignment and deformation registration on the same candidate crack region or candidate surface damage region to obtain a temporal damage region sequence.

[0127] The temporal damage region sequence refers to an ordered data set generated by aligning and registering the same candidate region in multiple consecutive frames of images. Its purpose is to eliminate coordinate drift caused by external interference (such as changes in UAV pose or blade vibration), thereby providing a stable data foundation for subsequent analysis. In practical applications, spatial alignment and deformation registration can be achieved using algorithms based on feature point matching or methods based on optical flow fields to ensure accurate correspondence between candidate regions in different frames.

[0128] Step 2: Based on the temporal damage region sequence, construct time series data of its location, length, width, area, and shape factor;

[0129] Time series data refers to records of multi-dimensional geometric features extracted from a time-series sequence of damaged regions, showing how these features change over time. This data can include information such as location, length, width, area, and shape factors. In practical applications, time series data can be constructed by extracting the geometric features of candidate regions frame by frame and arranging them in chronological order.

[0130] Step 3: Calculate the slope and variance of the time series data within the current window using the sliding window statistical method;

[0131] The sliding window statistical method is a technique used to analyze local data characteristics. It can be implemented using a fixed window size or an adaptive window size. By calculating the slope of the trend and the variance of the fluctuation, it can identify the damage propagation trend and quantify data stability, thereby effectively distinguishing between real damage and noise interference.

[0132] Step 4: When the slope of the change trend exceeds the preset growth threshold, or the fluctuation variance continues to be lower than the preset noise level threshold, dynamically adjust the confidence weight of the candidate region.

[0133] The mechanism for dynamically adjusting confidence weights can be flexibly designed according to actual scenario requirements, such as through weighting functions or rule engines to achieve real-time updates of weights.

[0134] Step 5: If the overall confidence level of the candidate region is higher than the judgment threshold in N consecutive frames, it is confirmed as a real damage.

[0135] In this embodiment, by spatially aligning and deformably registering candidate regions in N consecutive frames of images, the inter-frame displacement caused by external factors is compensated, ensuring that the temporal damage region sequence is established on a unified coordinate system. The time-series data constructed based on the temporal damage region sequence integrates geometric features such as position, length, width, area, and shape factor, comprehensively reflecting the gradual change pattern of damage. The sliding window statistical method achieves a dual evaluation of damage propagation trend and data stability by calculating the slope of the change trend and the variance of fluctuation. Low variance of fluctuation usually indicates noise interference, while a high slope may correspond to the actual damage propagation process. This dual index design effectively overcomes the defect of the fixed threshold method being sensitive to noise. Finally, by dynamically adjusting the confidence weight and combining it with the comprehensive confidence judgment of N consecutive frames, the system can optimize the judgment process according to the actual scenario, significantly enhancing its adaptability to complex environments and fundamentally reducing the occurrence of false alarms and false negatives.

[0136] In one embodiment of this specification, the human-computer interaction module 140 is further configured to:

[0137] It provides a multi-mode visual interface and supports real-time inspection mode and historical playback mode;

[0138] In real-time inspection mode, the current video frame, the marked damage location box, the damage type label, and the detection timestamp are displayed simultaneously and overlaid; different types of damage are distinguished by different colored or graphic style label boxes.

[0139] In the historical playback mode, it supports searching historical inspection records by wind turbine number, blade serial number, inspection time or multi-dimensional filtering conditions, and displays the evolution process of the same damaged area in multiple inspections in the form of a timeline.

[0140] In this embodiment, the multi-mode visualization interface refers to an interactive graphical interface that can switch display modes according to user needs. It can be implemented using software development kits (such as Qt, Electron, etc.) to meet the operational needs of maintenance personnel in different scenarios. The real-time inspection mode can be understood as a functional module focused on providing immediate feedback on damage information. It ensures the real-time nature and accuracy of information by synchronously processing video streams and analysis results. The historical playback mode supports the tracing and analysis of historical data. It can be implemented using database query technologies (such as SQL, NoSQL, etc.) combined with time series analysis algorithms, aiming to help maintenance personnel grasp the long-term evolution trend of damage. Furthermore, the design of the label's color or legend style aims to quickly convey information through visual differences. It can be implemented using graphics rendering engines (such as OpenGL, DirectX, etc.) to improve the operator's recognition efficiency.

[0141] In real-time inspection mode, the system synchronously overlays the current video frame with damage annotation information based on the continuity of the video stream. This mechanism ensures a strict correspondence between the damage location frame, type label, and timestamp and the original image, avoiding positioning deviations caused by processing delays. Different colors or icon styles are used to distinguish damage types, significantly improving the intuitiveness and accuracy of on-site assessment. In historical playback mode, the system supports flexible query methods based on multi-dimensional filtering conditions such as turbine number, blade sequence number, and inspection time, adapting to the complex needs of large-scale equipment management in wind farms. By displaying the damage evolution process in a timeline format, the system organizes multiple inspection results according to a time sequence, intuitively presenting the morphological change trajectory of the damaged area. This helps technicians analyze the damage expansion rate and pattern, distinguishing between natural environmental interference and actual structural degradation.

[0142] In one embodiment of this specification, the image processing module 120 includes an illumination enhancement unit 121, a size normalization unit 122, and a pixel normalization unit 123.

[0143] The illumination enhancement unit 121 is used to improve the image contrast and sharpness of the video stream through an adaptive histogram equalization algorithm to obtain a processed image;

[0144] Among them, the illumination enhancement unit 121 refers to a technical component specifically used to improve the local brightness distribution of an image. It can be implemented using a dynamic contrast adjustment method based on the statistical characteristics of local regions, with the aim of eliminating uneven illumination in complex natural environments.

[0145] The size normalization unit 122 is used to uniformly scale the processed image to a fixed resolution;

[0146] Among them, the size normalization unit 122 can be understood as a functional module that performs unified processing of image spatial scale. It can be implemented through image scaling algorithms such as bilinear interpolation and nearest neighbor interpolation, aiming to eliminate the resolution difference between different acquisition devices or video frames.

[0147] The pixel normalization unit 123 is used to normalize the pixel values ​​of the image to a preset numerical range to generate a standardized image;

[0148] Among them, the pixel normalization unit 123 is a mechanism for standardizing the intensity of image pixels. It can be achieved through linear mapping, nonlinear transformation, etc., with the aim of suppressing brightness fluctuations caused by changes in ambient light or sensor noise.

[0149] In this embodiment, the illumination enhancement unit 121 dynamically adjusts the contrast based on the local brightness characteristics of each frame in the video stream, thereby highlighting the subtle damage features on the leaf surface in shadow or bright light areas. The size normalization unit 122 uniformly scales the processed image to ensure that all input data remain consistent in geometric scale, avoiding positioning deviations caused by resolution differences. The pixel normalization unit 123 maps the image pixel values ​​to a preset range, further stabilizing the pixel intensity distribution and providing a high-quality standardized image for subsequent damage analysis.

[0150] Based on the same general inventive concept, this invention also protects an intelligent method for judging wind turbine blade damage, such as... Figure 3 As shown, Figure 3 This is a flowchart illustrating the intelligent judgment method for wind turbine blade damage provided in an embodiment of the present invention. The intelligent judgment method for wind turbine blade damage provided by the present invention will be described below. The intelligent judgment method for wind turbine blade damage described below can be referred to in correspondence with the intelligent judgment system for wind turbine blade damage described above.

[0151] Intelligent methods for judging wind turbine blade damage include:

[0152] Step 301: Collect image data of the wind turbine blades.

[0153] Step 302: Preprocess the image data to generate a standardized image.

[0154] Step 303: Extract the defect features of the wind turbine blades from the standardized image, and locate and classify the blade damage based on the defect features to generate a judgment result containing the damage location and damage type.

[0155] Step 304: Display the image data and judgment results on the display interface; wherein, the judgment results are marked in the video screen with the damage location and the damage type in the form of a label box.

[0156] Figure 4 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention.

[0157] like Figure 4 As shown, the electronic device may include a processor 410, a communication interface 420, a memory 430, and a communication bus 440. The processor 410, communication interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions from the memory 430 to execute an intelligent method for judging wind turbine blade damage.

[0158] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the intelligent judgment method for wind turbine blade damage provided by the above methods.

[0160] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent judgment method for wind turbine blade damage provided by the methods described above.

[0161] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent judgment system for wind turbine blade damage, characterized in that, include: The image acquisition module is used to acquire image data of the wind turbine blades; The image processing module is used to preprocess the image data to generate a standardized image; The damage analysis module is used to extract the defect features of the wind turbine blades from the standardized image, and to locate and classify the blade damage based on the defect features, generating a judgment result that includes the damage location and damage type. The human-computer interaction module is used to display the image data and the judgment result on the display interface; wherein, the judgment result marks the damage location in the video screen in the form of a label box and displays the damage type in association.

2. The intelligent judgment system for wind turbine blade damage according to claim 1, characterized in that, The damage analysis module includes: The feature extraction unit is used to perform multi-scale morphological filtering on the standardized image to extract strip-shaped and patch-shaped abnormal regions respectively. The crack screening unit is used to screen out strip-shaped abnormal regions whose directions deviate from the blade texture direction as candidate crack regions based on the directional field analysis of the blade texture. The surface damage screening unit is used for consistency analysis based on surface texture to screen out patchy abnormal regions with texture entropy exceeding a preset threshold as candidate surface damage regions. The damage determination unit is used to determine the candidate crack region and the candidate surface damage region as real crack damage and real surface damage respectively if the position offset and shape change rate of the candidate crack region and the candidate surface damage region remain within their respective preset threshold ranges within N consecutive frames; N is an integer greater than 1.

3. The intelligent judgment system for wind turbine blade damage according to claim 2, characterized in that, The crack screening unit is also used for: Determine the dominant local texture direction in the leaf region of the standardized image; Extract the skeleton line of the strip-shaped abnormal region and determine the main direction of the skeleton line; Determine the angle between the main direction of the skeleton line and the local dominant texture direction of the blade region where the skeleton line is located; When the included angle is greater than the first angle threshold, or when the strip-shaped abnormal region crosses the macroscopic feature line on the blade structure, the strip-shaped abnormal region is determined to be a candidate crack region. The macroscopic feature lines include the blade cap line or the leading edge bonding line.

4. The intelligent judgment system for wind turbine blade damage according to claim 3, characterized in that, The crack screening unit is also used for: The three-dimensional structural model of the wind turbine blade is combined with the UAV pose information and projected onto the image coordinate system of the current video frame to obtain the projection range of the beam cap area; If a candidate crack region is located within the projection range, and the angle between the main direction of its skeleton line and the direction of the beam cap is less than the second angle threshold, then the processing priority of the candidate crack region is increased.

5. The intelligent judgment system for wind turbine blade damage according to claim 3, characterized in that, Determine the dominant local texture orientation of the leaf region in the normalized image, including: An orientation field estimation algorithm based on gradient structure tensor is used to calculate the gradient direction and gradient magnitude of each pixel in the image. A sliding window is set within the leaf area, and all pixels within the window are weighted and voted on based on their gradient magnitude; where pixels with larger gradient magnitudes have higher voting weights in their gradient direction. The gradient direction distribution after all weighted votes within the sliding window is statistically analyzed, and the gradient direction that receives the highest number of weighted votes is determined as the dominant texture direction in the local region of that window. By traversing the entire blade region through a sliding window, a mapping relationship is established between each window position and its local dominant texture direction, forming a complete blade surface orientation field distribution map.

6. The intelligent judgment system for wind turbine blade damage according to claim 3, characterized in that, Extracting the skeleton line of the strip-shaped abnormal region and determining the main direction of the skeleton line includes: Binarize the strip-shaped abnormal regions in the standardized image; Morphological operations were used to refine the binarized strip-shaped abnormal regions and extract their skeleton lines. Select multiple key points on the skeleton line and calculate the local tangent direction of each key point; Statistical analysis was performed on the local tangent directions of all key points, and the principal component analysis method was used to determine the main direction of the skeleton line.

7. The intelligent judgment system for wind turbine blade damage according to claim 2, characterized in that, The surface damage screening unit is also used for: Within the patchy abnormal region, the similarity of its texture features with the preset surrounding normal leaf region at multiple scales is calculated. Determine the average brightness difference between the patchy abnormal area and the preset surrounding normal leaf area; If the texture feature similarity is lower than the first preset threshold, the patchy abnormal region is determined as a candidate surface damage region. For the candidate surface damage area, if the absolute value of its brightness difference is higher than the second preset threshold, it is classified as material defect damage. If the absolute value of the difference in brightness is lower than the second preset threshold, it is classified as stain coverage damage.

8. The intelligent judgment system for wind turbine blade damage according to claim 7, characterized in that, The damage determination unit is also used for: In N consecutive frames of images, spatial alignment and deformation registration are performed on the same candidate crack region or candidate surface damage region to obtain a temporal damage region sequence. Based on the temporal damage region sequence, time series data of its location, length, width, area and shape factor are constructed; The slope and variance of the time series data within the current window are calculated using a sliding window statistical method. When the slope of the change trend exceeds a preset growth threshold, or the fluctuation variance remains below a preset noise level threshold, the confidence weight of the candidate region is dynamically adjusted. If the confidence level of the candidate region is higher than the judgment threshold in N consecutive frames, it is confirmed as a real damage.

9. The intelligent judgment system for wind turbine blade damage according to claim 1, characterized in that, The human-computer interaction module is also used for: It provides a multi-mode visual interface and supports real-time inspection mode and historical playback mode; In real-time inspection mode, the current video frame, the marked damage location box, the damage type label, and the detection timestamp are displayed simultaneously and overlaid; different types of damage are distinguished by different colored or graphic style label boxes. In the historical playback mode, it supports searching historical inspection records by wind turbine number, blade serial number, inspection time or multi-dimensional filtering conditions, and displays the evolution process of the same damaged area in multiple inspections in the form of a timeline.

10. The intelligent judgment system for wind turbine blade damage according to claim 1, characterized in that, The image processing module includes: The illumination enhancement unit is used to improve the image contrast and sharpness of image data through an adaptive histogram equalization algorithm to obtain the processed image. Size normalization unit is used to uniformly scale the processed image to a fixed resolution; The pixel normalization unit is used to normalize the pixel values ​​of an image to a preset range to generate a standardized image.

Images