Fan blade fault detection method based on unmanned aerial vehicle and X-ray imaging

Abstract

The invention relates to a fan blade fault detection method based on unmanned aerial vehicles and X-ray imaging, and the method comprises the steps: employing two unmanned aerial vehicles to carry an X-ray generator and a detector respectively, enabling the two unmanned aerial vehicles to synchronously fly at the front and rear sides of a fan blade, and maintaining a preset distance to form a perspective light path; accurate flight planning and image acquisition calculation are carried out, an acquisition point matrix and the number of images required for covering the whole blade are determined, and exposure parameters are dynamically adjusted to carry out image acquisition according to the blade linear velocity detected in real time under the state that the fan is not stopped; the collected images are subjected to unsupervised image deblurring processing and a splicing algorithm based on collection position information, and a clear blade internal structure panorama is finally reconstructed and used for accurately recognizing internal damage such as layering, debonding and cracks; according to the invention, high-efficiency, accurate and non-contact online nondestructive testing is realized in a non-stop state, and the safety and economical efficiency of fan operation and maintenance are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power generation equipment detection, and in particular to a wind turbine blade fault detection method based on a UAV and X-ray imaging. BACKGROUND

[0002] As an important part of clean energy, the stable operation of wind power generation is crucial; longer blades enable the wind turbine to sweep more areas and capture more wind, thereby generating more electricity; longer blades bear more stress and weight, requiring stronger structural support. At the same time, wind turbines are directly exposed to the external environment, and as they rotate throughout the year, their initial performance may decline, or their structure, material, and connection layer may be damaged. As the core component for capturing wind energy, wind turbine blades are prone to internal damage such as delamination, debonding, cracking, and web cracking due to their large size, complex structure, long-term exposure to alternating loads, and external impacts such as extreme weather. Data shows that wind turbine blade breakage accidents caused by fatigue and external damage account for 50% of the failure rate of key components of wind turbine units, which is a major operational pain point and safety hazard for wind farm owners and manufacturers. Regular and effective detection and diagnosis of the internal structure of wind turbine blades are crucial for their operation and maintenance.

[0003] To ensure the safe operation of wind turbines, various non-destructive testing (NDT) and online detection technologies have emerged, including the following: External visual detection: using a UAV to carry a high-definition camera or laser radar to take pictures or scans of the blade surface. This method can effectively identify cracks, corrosion, and lightning damage on the surface of the wind turbine blade, but it cannot detect internal structural defects and damage. Internal damage is often a precursor to structural failure. Internal detection based on a robot platform: using a crawling robot or tracked vehicle to carry an ultrasonic probe into the blade interior for detection. This method can access the interior of the wind turbine blade, but it must be performed when the wind turbine is shut down, requiring more than a day of downtime for three blades, resulting in significant loss of power generation. In addition, this method relies heavily on manual operation and experience-based judgment, resulting in low detection efficiency and inconsistent results. Furthermore, due to the difficulty of robot movement in the interior of large or curved wind turbine blades, there is a risk of jamming. Online monitoring and inference technology, including acoustic emission monitoring, vibration analysis and sound detection, etc., which collects data through sensors installed on the tower or nacelle, combined with machine learning, artificial intelligence algorithm to infer the structural health status of the blade; However, these methods are indirect methods, the accuracy depends heavily on the accumulation of a large amount of historical data, accurate feature extraction and complex model training, it is difficult to accurately locate the micro damage, resulting in high false positive rate and false negative rate, and ultimately still need human intervention to verify; Traditional X-ray detection: X-ray transmission imaging technology can intuitively present the internal structure of the object, and is maturely applied in the field of industrial flaw detection; However, applying X-ray transmission technology to the detection of running fan blades faces many problems. Due to the large volume of the blade, fixed X-ray equipment cannot achieve full coverage detection, if mobile X-ray equipment is used, a huge lifting platform needs to be built, which not only has high cost and long construction period, but also needs to detect the fan in the shutdown state.

[0004] The prior art can detect the internal structure of the fan blade, but it must be shut down for a long time and is inefficient. If the fan is not shut down or detected online, the internal damage of the blade cannot be diagnosed intuitively and accurately. And with the acceleration of the large-scale trend of wind turbine, the length of the blade is growing from 65 meters to more than 150 meters. Longer blades mean greater stress and more complex internal structure, which puts higher challenges on the efficiency, accuracy and coverage of the detection technology. SUMMARY

[0005] Therefore, the present application provides a fan blade fault detection method based on unmanned aerial vehicle and X-ray imaging, which effectively solves the problem that the prior art cannot realize high-precision and high-efficiency detection of the internal structure of the blade in the shutdown state of the fan blade, and provides a new detection method that can complete the internal imaging of the whole blade of the fan in the grid-connected operation state of the fan unit.

[0006] To achieve the above purpose, the present application provides a fan blade fault detection method based on unmanned aerial vehicle and X-ray imaging, comprising the following steps: S1, configuring an unmanned aerial vehicle system, including a first unmanned aerial vehicle and a second unmanned aerial vehicle, the first unmanned aerial vehicle carrying an X-ray generator, and the second unmanned aerial vehicle carrying an X-ray detector; S2, according to the size parameters of the fan blade to be detected and the field of view size of the X-ray detector, calculating the image acquisition point matrix required to cover the surface of the fan blade to be detected and the number of image acquisitions required for each acquisition point; S3, generating a UAV flight plan according to the type of wind turbine to be detected, controlling the first UAV and the second UAV to fly synchronously on the front and back sides of the wind turbine blade to be detected and maintain a relative position, while controlling the exposure position of the X-ray generator and the time interval of the transverse movement of the wind turbine blade to be detected, so as to ensure uniform overlap rate and consistent view angle, so that the rays emitted by the X-ray generator penetrate the blade and are received by the detector, forming a perspective light path; S4, in the running state of the wind turbine to be detected without stopping, controlling the first UAV to move to each collection point according to the flight plan to collect images, and the second UAV to reach each collection point to collect images on the other side of the wind turbine blade to be detected; S401, the first UAV carries the X-ray emitter to the collection point according to the absolute route of the flight plan, determines the blade sequence of the wind turbine to be detected reaching the collection point in the rotation process through the position sensor, and calculates the linear speed of the wind turbine blade to be detected , dynamically adjusts the exposure interval, exposure speed and image accuracy of the X-ray generator according to the linear speed of the wind turbine blade to be detected; S402, the UAV system intermittently takes pictures and exposes at each collection point position along the linear speed direction of the wind turbine blade to be detected, sequentially collects three blade images of the wind turbine to be detected, and calculates and updates the wind turbine blade speed of the next collection point, when the image collection of the third blade of the current collection point is completed, the UAV system automatically moves to the next collection point along the length direction of the wind turbine blade and repeats step S401 until the image collection of all pre-calculated collection points is completed. S5, preprocessing all collected X-ray images, using an unsupervised image deblurring method to improve clarity, and based on the position information of the collection point and the image collection parameters, splicing the images and reconstructing the complete internal structure image of the wind turbine blade to be detected; S6, judging the internal fault of the wind turbine blade to be detected according to the complete internal structure image of the wind turbine blade to be detected and the original internal structure image or the adjacent wind turbine blade image.

[0007] Preferably, step S2 further comprises the following steps: S201, establishing a two-dimensional coordinate system with the length of the wind turbine blade to be detected as the first direction and the chord length as the second direction; S202, calculating the required image collection point matrix in the first direction and the second direction and the required image collection quantity of each collection point according to the field of view size of the X-ray detector and the preset adjacent image overlap rate.

[0008] Preferably, generating a UAV flight plan comprises the following steps: S301. The coordinate position of the center of the wind turbine blade shaft is calculated based on the coordinate position of the wind turbine tower base to be tested and the wind turbine structure type. S302. Preset the attenuation level of X-rays, and calculate the distance between the X-ray generator and the X-ray detector according to Beer-Lambert's law, the expression is: ; in, Indicates the linear attenuation coefficient. This indicates the thickness of the wind turbine blade material to be tested. Indicates the initial X-ray intensity of the X-ray generator. This indicates the intensity of X-rays received by the X-ray detector. This indicates the degree of X-ray attenuation; the distance between the X-ray generator and the X-ray detector is maintained between 10 and 12 meters. The first and second UAVs automatically maintain a distance from the wind turbine blades to be tested based on position sensors.

[0009] Preferably, after the position sensor of the first UAV detects that the wind turbine blade to be tested has entered the field of view of the X-ray generator, the X-ray generator calculates the lateral acquisition frequency of the wind turbine blade to be tested based on the linear velocity of the wind turbine blade to be tested, including the image exposure time and image interval. Calculate the linear velocity of the wind turbine blades under test at the current data acquisition point. Includes the following steps: The position sensor of the first UAV starts timing from the moment the first blade of the wind turbine under test passes through the X-ray generator's field of view at the current acquisition point, and waits to record the time points when the second and third blades enter the X-ray generator's field of view. , That is, the time it takes for the three blades of the wind turbine under test to traverse the current data acquisition point in a 240° arc, and the linear velocity of the wind turbine blades at the current data acquisition point. for: ; in, This indicates the distance from the center of the impeller of the fan under test to the current data collection point.

[0010] Preferably, the unsupervised image deblurring method employs an optimization approach combining the Latent Space Model (DIP) with the Unsharp Masking Image Sharpening Method (UMFS). The DIP iterative convergence expression is: ; in, The loss function represents the convergence of the iteration. , These represent the weight coefficients of the original image and the target image, respectively. This represents the initial reconstructed image generated by the encoder-decoder process, representing random noise variables. This represents the original "blurred" image as input. This represents the sharpened target image obtained after unsharpening masking, achieved by adjusting the loss weights. and Controlled image reconstruction The balance between fidelity and sharpening intensity; Image reconstruction is initialized using forward propagation with random noise. The difference between the original image and the target image is calculated based on the loss function, and the reconstructed image is updated via backpropagation. This process is repeated to seek a clear image; the loss function serves as a priori to guide the optimization process toward convergence and reconstruction of the image space.

[0011] Preferably, image stitching based on the location information of the acquisition points and image acquisition parameters includes the following steps: The initial distance between the two images is calculated based on the acquisition point location, image acquisition time, drone attitude angle, and the current wind turbine blade linear velocity at the acquisition point when the first drone acquires two adjacent X-ray images. Homography matrix; With the initial Starting with the homography matrix, the Powell optimization method combined with cross-correlation is used to maximize the cross-correlation value between two images. A one-dimensional search is performed in the homography matrix space along each direction, and the resulting displacement is used to correct and optimize the obtained homography matrix. The optimized final homography matrix is ​​used to project the image onto the final stitching plane, thus completing the image stitching.

[0012] Compared with the prior art, the beneficial effects of the present invention are: This invention combines UAV swarm technology with X-ray imaging to construct a complete online non-destructive testing system for wind turbine blades, bringing multiple disruptive benefits. For the first time, this invention achieves high-precision, high-efficiency, and fully automated non-destructive imaging testing of the internal structure of wind turbine blades under normal operating conditions without shutting down the turbine. This fundamentally solves the long-standing paradox in the wind power industry that "testing requires shutdown." Compared with existing technologies, this invention can clearly present minute defects such as delamination, debonding, and cracks inside the blades in image form, elevating the testing mode from traditional "data analysis," "expert diagnosis," "post-maintenance," or "periodic maintenance" to intuitive "predictive maintenance," thus improving the operational philosophy. Meanwhile, this invention ensures the efficiency and reliability of the detection process through precise flight planning and adaptive image acquisition strategies. Based on accurate calculations of blade size and detector field of view, combined with synchronous control of the UAV swarm, it achieves seamless full-coverage scanning of blades ranging from tens to hundreds of meters in length, completely eliminating detection blind spots. By sensing the blade linear velocity in real time and dynamically adjusting exposure parameters, it intelligently compensates for the impact of high-speed blade rotation on image quality, ensuring the basic clarity of every image from root to tip. Furthermore, the unsupervised image deblurring and position-based stitching algorithms employed in this invention effectively overcome image quality problems caused by UAV jitter, low-dose X-ray noise, and blade surface distortion. The resulting panoramic internal structure map, with high signal-to-noise ratio and rich details, provides a reliable basis for accurate diagnosis. Attached Figure Description

[0013] Figure 1 is a flowchart of the present invention for detecting internal faults in wind turbine blades based on UAV and X-ray imaging. Figure 2 This invention relates to an unmanned aerial vehicle (UAV) system; Figure 2 (a) is a schematic diagram of the first UAV of the present invention carrying an X-ray generator. Figure 2 (b) is a schematic diagram of the second UAV of the present invention equipped with an X-ray detector; Figure 3 This is a basic schematic diagram of the imaging principle of the UAV system of the present invention, which carries an X-ray generator and a detector. In the diagram, ① represents a cross-sectional view of the wind turbine blade being inspected, ② represents the UAV carrying the X-ray generator, ③ represents the UAV carrying the X-ray detector, ④ represents the optical path between the X-ray generator and the detector, and ⑤ represents the field of view (FOV) obtained by the X-ray detector from the optical path. Figure 4 This invention provides an X-ray image set of a 65-meter-long wind turbine blade with a chord length of 2.8 meters, wherein the coordinates in the x-direction can be used as the relative acquisition point positions; Figure 4 (a) This invention generates a cut image of a 65-meter-long wind turbine blade with a chord length of 2.8 meters. To maintain a proportional display, the image is halved in the x-direction. Figure 4 (b) is a coordinate system for a 65-meter-long wind turbine blade X-ray image group with a chord length of 2.8 meters according to the present invention, wherein the dark color represents the image group covering the blade, and the coordinate in the x-direction is multiplied by 2; Figure 5 This is a schematic diagram of the wind turbine blade rotation for the flight planning of the UAV X-ray equipment of this invention. ① represents the main view of the wind turbine blade and rotation axis; ② represents a blade of the wind turbine during operation; ③ represents the rotation direction of the wind turbine; ④ represents the starting point where the UAV carrying the X-ray generator "photographs" the blade, which can also be called... Figure 4The (0,0) relative starting point; ⑤ The flight plan of the UAV carrying X-ray equipment according to the blade shape (considering the influence of motion) (matching the streamline shape of the blade when in motion) and the "photographing" exposure time (collection) starting point, the straight line indicates that the UAV flight collection trajectory does not move with the blade rotation; ⑥ The linear velocity of the wind turbine blade in operation at the planned collection point (location). Detailed Implementation

[0014] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0015] This embodiment employs two drones, each carrying an X-ray generator and a detector, to conduct X-ray transmission imaging to reveal the internal structure of wind turbine blades. The drones are simultaneously positioned and hovered before and after the wind turbine blades for transmission imaging. The imaging principle is similar to taking a chest X-ray in a hospital radiology department to diagnose lung conditions such as tuberculosis, inflammation, effusion, or even tumors. The image accuracy can reach 0.5 mm. Image processing clearly reveals the internal structure of the wind turbine blades. The number of images required for each blade is calculated based on the diagnostic accuracy. The method addresses the impact of environmental factors on drone acquisition, image stitching, and deblurring, among other key aspects. Non-stop testing can complete image acquisition for three wind turbine blades within 43 minutes. The differences in X-ray energy absorption and refraction of the blade structural materials, as well as the X-ray image reconstruction method, are not the primary protection content of this application.

[0016] The main components of commonly used wind turbine blade materials include: polyester gel coat, fiberglass, polyester resin, wood, PVC foam, vinyl ester adhesive, and alloy materials (metals). Among them, metal materials are mainly used for fixing the blade root, lightning protection wires, and necessary internal conductive materials (for lighting). The blade structure is mainly non-metallic, and the blade surface is made of polyester gel coat material. Taking a 65-meter long blade as an example, the blade root diameter is 1.9 meters, the chord length is less than 2.8 meters, and the maximum laminate thickness is 110 millimeters.

[0017] This embodiment provides a method for detecting wind turbine blade faults based on UAVs and X-ray imaging, specifically including the following steps: S1. Configure an unmanned aerial vehicle system, including a first unmanned aerial vehicle and a second unmanned aerial vehicle, wherein the first unmanned aerial vehicle carries an X-ray generator and the second unmanned aerial vehicle carries an X-ray detector; This embodiment selects existing X-ray equipment in the lightweight market so that it can be equipped on a drone. The technical specifications of the X-ray generator are shown in Table 1. Table 1 Technical Specifications of X-ray Generator:

[0018] In this embodiment, the Teledyne ICM CP series SITEX CP 300DM is used; The technical specifications of the X-ray detector are shown in Table 2: Table 2 Technical Specifications of X-ray Detectors:

[0019] In this embodiment, a Varex PaxScan 4343W (430mm x 430mm) flat panel detector is used; like Figure 2 As shown in (a) and (b), the first UAV carries an X-ray generator and the second UAV carries an X-ray detector to detect and diagnose internal structural damage (delamination, debonding, cracks) of wind turbine blades using radiographic imaging.

[0020] S2. Based on the size parameters of the wind turbine blade to be inspected and the field of view (FOV) of the X-ray detector, calculate the matrix of image acquisition points required to cover the surface of the wind turbine blade to be inspected and the number of images required for each acquisition point. S201. Establish a two-dimensional coordinate system with the length of the wind turbine blade to be tested as the first direction and the chord length as the second direction. S202. Based on the field of view of the X-ray detector and the preset overlap rate of adjacent images, calculate the required image acquisition point matrix in the first direction and the second direction, as well as the required number of images to be acquired at each acquisition point. like Figure 3 As shown in Figure ⑤, assuming the field of view (FOV) of the X-ray detector is 400mm × 400mm, and the required image pixel resolution is 0.5mm, each image should be no smaller than a 2048 × 2048 pixel array. For a 65-meter-long wind turbine blade with a chord length of 2.8 meters, a two-dimensional image matrix of 7 × 163 FOVs is needed to cover it. Figure 4 As shown in (a), to display proportionally, the X-axis direction actually needs to be doubled (multiplied by 2); considering that image preprocessing before detection and diagnosis requires image stitching, it is required that at least 30% overlap area be acquired between adjacent images, that is, adjacent images have 30% overlap in both horizontal and vertical directions, so that the coverage of the image group matrix is ​​expanded to 10×212 FOVs; according to Figure 4 (b) The results of the calculation and stitching show that approximately 952 images are needed to reconstruct the blades, while approximately 1237 images are actually needed. For one task (inspecting one wind turbine), three blades are collected, requiring a total of 3711 images.

[0021] S3. Generate a UAV flight plan based on the type of wind turbine to be tested, control the first UAV and the second UAV to fly synchronously on the front and rear sides of the wind turbine blade to be tested and maintain relative positions, and control the exposure position of the X-ray generator and the time interval of the lateral movement of the wind turbine blade to be tested to ensure uniform overlap and consistent viewing angle, so that the rays emitted by the X-ray generator penetrate the blade and are received by the detector to form a transparent light path. S301. The coordinate position of the center of the wind turbine blade shaft is calculated based on the coordinate position of the wind turbine tower base to be tested and the wind turbine structure type. S302. Preset the attenuation level of X-rays, and calculate the distance between the X-ray generator and the X-ray detector according to Beer-Lambert's law, the expression is: ; in, This represents the linear attenuation coefficient, which is related to the material density. This indicates the thickness of the wind turbine blade material to be tested. Indicates the initial X-ray intensity of the X-ray generator. This indicates the intensity of X-rays received by the X-ray detector. Indicates the degree of X-ray attenuation, if (Air), ,but: ; That is, if the X-ray generator and detector are 10m apart, the intensity will decrease by 9.5%; if they are 12m apart, the intensity will decrease by 11.3%. Therefore, the distance between the X-ray generator and the X-ray detector should be kept between 10 and 12 meters. If the blade thickness is 2 meters (the thickest part is at the root of the wind turbine blade), the drone and the blade can maintain a distance of 4 meters. Because the rotating plane of the wind turbine blade is subjected to wind load, it tilts towards the inner corner of the rear end of the wind turbine (from the blade root to the blade tip). Due to the length of the blade, even with a small angle of tilt, the blade tip will have a relatively large positional difference or spatial range (the right-angled side opposite the inner corner) relative to the blade root. This range affects the position and spacing of the lead drone carrying the X-ray generator and the slave drone carrying the detector as they fly synchronously in front of and behind the wind turbine blade to collect images. Therefore, the drone needs to automatically maintain the distance between the drone and the blade based on its position sensor. The first and second UAVs automatically maintain a distance from the wind turbine blades to be tested based on position sensors.

[0022] In this embodiment, the UAV flight planning is completed through flight planning software services provided by the UAV manufacturer, which improves the efficiency of UAV operations and ensures the quality of subsequent image processing. This is especially important for obtaining and imaging parameters during wind turbine operation. Since each wind turbine has its precise coordinates, the coordinates of the turbine blade shaft center can be accurately calculated based on the turbine's structure (type). This allows for the establishment of an absolute flight path for the UAV targeting each wind turbine. If the wind turbine is shut down, the starting point for blade detection, also known as the zero point, can be fixed, which can be understood as... Figure 4 In (a), the coordinates of position (0, 0) and the center of the blade shaft are relative to a similar type of wind turbine; therefore, according to Figure 4 (a) The coordinate positions of 212 “rigid” acquisition points (considering 30% overlap) in the longitudinal direction (standard axis X direction) were determined and input into the flight planning program; Figure 4 (a) The transverse (standard axis Y direction) image is obtained by fixing the FOV of the longitudinal acquisition point relative to the longitudinal coordinate of the blade, using the rotation of the wind turbine blade, combined with the feedback information from the position sensor on the UAV (detecting that the blade enters the FOV of the X-ray equipment), calculating the linear velocity of the blade from the acquisition point to the center of the impeller shaft, and determining the exposure time and interval for image acquisition.

[0023] S4. While the wind turbine under test is running without stopping, control the first UAV to move to each collection point according to the flight plan to collect images, and the second UAV to reach the image collection point on the other side of the wind turbine blades corresponding to each collection point. S401. The first UAV, carrying an X-ray emitter, travels to the collection point along the planned absolute route. Using a position sensor, it determines the order in which the blades of the wind turbine under test arrive at the collection point during rotation and calculates the linear velocity of the wind turbine blades. The exposure interval, exposure speed, and image accuracy of the X-ray generator are dynamically adjusted according to the linear velocity of the wind turbine blades to be tested. After the position sensor of the first UAV detects that the wind turbine blade under test has entered the field of view of the X-ray generator, the X-ray generator calculates the lateral acquisition frequency of the wind turbine blade under test based on the linear velocity of the wind turbine blade under test, including the image exposure time and image interval. Calculate the linear velocity of the wind turbine blades under test at the current data acquisition point. Includes the following steps: The position sensor of the first UAV starts timing from the moment the first blade of the wind turbine under test passes through the X-ray generator's field of view at the current acquisition point, and waits to record the time points when the second and third blades enter the X-ray generator's field of view. , That is, the time it takes for the three blades of the wind turbine under test to traverse the current data acquisition point in a 240° arc, and the linear velocity of the wind turbine blades at the current data acquisition point. for: ; in, This indicates the distance from the center of the impeller of the fan under test to the current data collection point; Because the linear velocity at the root of the wind turbine blades is slow, the X-ray generator can improve the sensitivity of the acquired images by increasing the exposure time and radiation dose (15 FPS). As the wind turbine blades approach the tip, their linear velocity increases, and the blade thickness also tends to decrease (to its thinnest point), while the chord length decreases (to its minimum). For a 65-meter blade, at a normal rotational speed of 15 revolutions per minute, the speed at the blade tip exceeds 100 meters per second. By reducing the exposure time (100+ FPS), the impact of the blade rotation speed on the acquired images can be reduced, and the number of lateral images acquired can be reduced to compensate for (increase) the exposure time. Simultaneously, using the drone's position sensor, as the acquisition point gradually approaches the blade tip, the drone can appropriately increase its distance from the blade, maintaining a distance of 5-6 meters. This avoids collisions caused by sudden airflow.

[0024] Unmanned aerial vehicle (UAV) flight planning primarily targets UAVs carrying X-ray generators, using the navigator (also known as the pilot aircraft) as the central control point to make flight decisions. Based on the flight plan, the navigator generates and distributes detailed real-time flight path commands to the slave UAVs carrying X-ray detectors. This ensures that the two UAVs maintain synchronized distance, angle, attitude, and status. S402. The UAV system intermittently takes and exposes images along the transverse linear velocity direction of the wind turbine blades to be tested at each acquisition point, sequentially acquiring images of the three blades of the wind turbine to be tested, while simultaneously calculating and updating the wind turbine blade velocity at the next acquisition point. When the image acquisition of the third blade at the current acquisition point is completed, the UAV system automatically moves along the length direction of the wind turbine blades to the next acquisition point, with a straight-line distance of 290mm, and then waits for the first blade to move over to repeat the image acquisition process until the image acquisition of all pre-calculated acquisition points is completed. Determine the flight plan based on the wind turbine type, such as Figure 2 The two drones shown carry X-ray emitters and detectors. A pre-calculated flight plan is input into both drones, and the drone carrying the X-ray emitter is designated as the navigator, and the drone carrying the X-ray detector as the slave. Figure 3 The diagram shows the X-ray path that ensures X-ray stability during flight. In this embodiment, the discussion assumes a wind turbine blade length of 65 meters, a blade root diameter of 1.9 meters, a chord length of 2.8 meters, and a wind turbine rotation speed of 5 revolutions per minute. According to the flight plan, such as Figure 5 As shown in ④, once the navigator reaches the wind turbine blade data collection starting point, the UAV's onboard position sensor detects the first blade passing through the X-ray emitter's field of view (FOV). This marks the first blade as the first one, and simultaneously begins timing, waiting, and recording the times when the second and third blades pass over the wind turbine. , The time required for the third blade to trace a 240-degree rotating arc. , Calculate the wind turbine blade speed (5 revolutions / minute), assuming the first data collection point is... Figure 4 At position (0, 0) in (a), 3 meters from the center of the blade shaft, the linear velocity of the wind turbine blade at the first detection point is: ; This generates a list of image acquisition parameters for the acquisition point, as shown in Table 3-1; where "location" corresponds to... Figure 4 (b) In the X-axis direction, the transverse sampling area of ​​the leaf is 1.9 meters (refer to the diameter of the leaf root being 1.9 meters). Table 3-1: ; Assumption Figure 5 -⑥ The sampling point is located at 40 (based on 212 "rigid" sampling points across the entire blade field; the calculated radius needs to include the fixed distance from the blade's rotation axis center to the first sampling point), with a chord length of 2.8 meters. The image sampling parameter list for this sampling point is shown in Table 3-2: Table 3-2: ; The number of leaf images is based on Figure 4 (b) The y-axis direction is determined by the image spacing (considering 30% overlap); the image spacing is equivalent to the image dimension, 290mm; the blade linear velocity is obtained in real time; the number of images acquired depends on the blade linear velocity, and the exposure interval is determined by the exposure speed; the exposure speed should be shorter than the exposure interval; the image accuracy (0.5mm-2mm) can be adjusted according to the exposure speed; the contents of Tables 3-1 and 3-2 will be bound to each acquired image to assist in subsequent image stitching.

[0025] Since the root of the wind turbine blade is the thickest and slowest part with the slowest linear velocity, the X-ray generator can improve the image acquisition sensitivity by increasing the exposure time and radiation dose (15 FPS). As the wind turbine blade approaches the tip, its linear velocity increases, and the blade thickness also tends to become thinner (thinnest), and the chord length becomes smaller (minimum). For a 65-meter blade, if it rotates at a normal speed of 15 revolutions per minute, the speed at the tip of the wind turbine blade exceeds 100 meters per second. By reducing the exposure time (100+ FPS), the impact of the blade rotation speed on the acquired image can be reduced, and the number of lateral acquired images can be reduced to compensate for (increase) the exposure time. At the same time, using the drone's position sensor, as the acquisition point gradually approaches the blade tip, the drone can appropriately increase its distance from the blade, maintaining a distance of 5-6 meters. This avoids collisions caused by sudden airflow. Unmanned aerial vehicle (UAV) flight planning is mainly for UAVs carrying X-ray generators, and the navigator is used as the central control point to make flight decisions. Based on the flight planning task, the navigator generates and distributes detailed real-time flight path commands to the slave UAVs carrying X-ray detectors, thereby ensuring that the two UAVs maintain synchronized distance, angle, attitude and status. Since the data is collected without stopping, assuming that the wind turbine rotates for 12 seconds (5 rotations / minute) and there are 212 data collection points, it would take 43 minutes (12×212 / 60=42.4) to complete the image collection for one wind turbine (three blades). This parameter affects the selection of the drone's endurance.

[0026] S5. Preprocess all acquired X-ray images, use unsupervised image deblurring method to improve clarity, and stitch images based on the location information of acquisition points and image acquisition parameters to reconstruct the complete internal structure image of the wind turbine blade to be detected. X-ray imaging is used to detect internal structural damage in wind turbine blades (early characteristics of wind turbine blade failure), including delamination, debonding, and cracks (fractures). This is a comprehensive solution from image acquisition to image processing. Image acquisition is performed during the rotation of the wind turbine blades, and the slight shaking caused by airflow when the drone hovers at the acquisition point can lead to image blurring. Considering the drone's load limitations, using low-dose X-ray equipment can increase image noise and reduce the signal-to-noise ratio, further degrading image quality. Delamination, debonding, and cracks are usually small, low-contrast features that are extremely sensitive to image sharpness. Therefore, image preprocessing is needed to compensate for and resolve the blurring caused by X-ray images acquired during drone flight and blade rotation. Considering the limitations of detection accuracy and the FOV of the X-ray detector, hundreds or even thousands of images need to be stitched together for a single blade. Effective image stitching methods also fall under the scope of image preprocessing. In this embodiment, the unsupervised image deblurring method adopts an optimization method that combines the latent space model DIP (Deep Image Prior) with the unsharp masking image sharpening method UMFS (Unsharp Masking Filter Sharping), thereby accelerating convergence and improving the clarity of the final output image. Using a DIP image with a potential spatial structure The input is a randomly initialized noise vector. This yields a corresponding initial evaluation image. During initialization, we can assume... Given a uniform image, subsequent iterations will accumulate the feature differences. At the same time, the acquired "blurred" images are directly... The sharpening mask generates the sharpening target function (Unsharp Mask Filter Sharpening), and the specific expression is as follows: ; ; ; in, Represents a convolution (blurring) function (filter). This represents the kernel size of the convolution function. Indicates the sharpening intensity coefficient; The convergence expression for the DIP iteration is: ; in, The loss function represents the convergence of the iteration. , These represent the weight coefficients of the original image and the target image, respectively. This represents the initial reconstructed image generated by the encoder-decoder process, representing random noise variables. This represents the original "blurred" image as input. This represents the sharpened target image obtained after unsharpening masking, achieved by adjusting the loss weights. and Controlled image reconstruction The balance between fidelity and sharpening intensity; Image reconstruction is initialized using forward propagation with random noise. The difference between the original image and the target image is calculated based on the loss function, and the reconstructed image is updated via backpropagation. This process is repeated to seek a clear image; the loss function serves as a priori to guide the optimization process toward convergence and reconstruction of the image space. The first part of the formula is the reconstruction loss function, which aims to ensure that the DIP output is consistent with the original input image in overall structure, forming the basis of reconstruction. The second part of the formula is the sharpening guide loss function, which encourages the output of the DIP. As close as possible to its own sharpened image. This means network parameters The optimization direction is forced towards producing sharper and clearer results. Each iteration involves overlaying the difference images (features) corresponding to the above differences onto... After each iteration, The difference should be less than the value of the previous iteration (convergence), guiding the optimization process toward a clearer and less distorted reconstruction of the image space; This image sharpening method combines DIP with UMFS, allowing DIP to internally learn how to generate sharp images, rather than forcibly sharpening (filtering) the image during the raw spatial processing stage. This effectively avoids motion blur noise, producing more natural results with fewer artifacts, as DIP attempts to find a balance between sharpening and more random noise suppression. The entire process does not require any sharp reference image (Ground Truth), making it suitable for applications where paired data cannot be obtained as a reference during UAV X-ray acquisition. This is achieved by adjusting the loss weights. and Controlling the balance between the fidelity of the reconstructed image and the sharpening intensity, The higher the value, the stronger the sharpening effect, which can effectively compensate for the potential reduction in image signal-to-noise ratio caused by using the lowest possible dose of X-ray equipment, and is more sensitive to imaging damage detection of the internal structure of wind turbine blades.

[0027] The characteristics of X-ray images of the internal structure of wind turbine blades are that local structural features are not obvious, and the composite materials of the blades exhibit repetitive textures. Using perspective-based methods, the shooting angle of the UAV carrying the X-ray equipment and the curved surface of the blades can lead to perspective and projection distortions, rendering traditional feature-point-based stitching methods (such as SIFT and ORB) ineffective. Therefore, a method suitable for the characteristics of wind turbine blades is needed, a stitching strategy that does not rely on obvious image features. Due to the strict control of UAV flight planning and image acquisition, image stitching can rely on positional parameters during image acquisition, as shown in Tables 3-1 and 3-2. This is called a stitching method based on the UAV acquisition position. Image stitching based on the positional information of the acquisition points and image acquisition parameters includes the following steps: The initial homography matrix (3×3) between the two images is calculated based on the acquisition point position, image acquisition time, UAV attitude angle, and wind turbine blade linear velocity at the current acquisition point when the first UAV acquires two adjacent X-ray images. Starting from the initial homography matrix, the Powell optimization method is used in combination with cross-correlation. With the goal of maximizing the cross-correlation value of the two images, a one-dimensional search is performed along each direction in the initial homography matrix space. The resulting displacement correction optimizes the obtained homography matrix. The optimized final homography matrix is ​​used to project the image onto the final stitching plane, completing the image stitching process. Figure 4 As shown in (b).

[0028] S6. Determine the internal faults of the wind turbine blade under test based on the reconstructed complete internal structure image of the wind turbine blade under test and the original internal structure image (obtained at the factory) or the image of an adjacent wind turbine blade, for example, by image subtraction.

[0029] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for detecting wind turbine blade faults based on unmanned aerial vehicles (UAVs) and X-ray imaging, characterized in that, Includes the following steps: S1. Configure an unmanned aerial vehicle system, including a first unmanned aerial vehicle and a second unmanned aerial vehicle, wherein the first unmanned aerial vehicle carries an X-ray generator and the second unmanned aerial vehicle carries an X-ray detector; S2. Based on the size parameters of the wind turbine blade to be inspected and the field of view of the X-ray detector, calculate the matrix of image acquisition points required to cover the surface of the wind turbine blade to be inspected and the number of images required for each acquisition point. S3. Generate a UAV flight plan based on the type of wind turbine to be tested, control the first UAV and the second UAV to fly synchronously on the front and rear sides of the wind turbine blade to be tested and maintain relative positions, and control the exposure position of the X-ray generator and the time interval of the lateral movement of the wind turbine blade to be tested to ensure uniform overlap and consistent viewing angle, so that the rays emitted by the X-ray generator penetrate the blade and are received by the detector to form a transparent light path. S4. While the wind turbine under test is running without stopping, control the first UAV to move to each collection point according to the flight plan to collect images, and the second UAV to reach the image collection point on the other side of the blade of the wind turbine under test corresponding to each collection point. S401. The first UAV, carrying an X-ray emitter, travels to the collection point along the planned absolute route. Using a position sensor, it determines the order in which the blades of the wind turbine under test arrive at the collection point during rotation and calculates the linear velocity of the wind turbine blades. The exposure interval, exposure speed, and image accuracy of the X-ray generator are dynamically adjusted according to the linear velocity of the wind turbine blades to be tested. S402. The UAV system takes intermittent photos and exposes them along the transverse linear velocity direction of the wind turbine blades to be tested at each collection point, and sequentially collects images of the three blades of the wind turbine to be tested. At the same time, it calculates and updates the wind turbine blade velocity at the next collection point. When the image of the third blade at the current collection point is completed, the UAV system automatically moves to the next collection point along the length direction of the wind turbine blades and repeats step S401 until the image collection of all pre-calculated collection points is completed. S5. Preprocess all acquired X-ray images, use unsupervised image deblurring method to improve clarity, and stitch images based on the location information of acquisition points and image acquisition parameters to reconstruct the complete internal structure image of the wind turbine blade to be detected. S6. Determine the internal faults of the wind turbine blade under test based on the reconstructed complete internal structure image of the wind turbine blade under test and the original internal structure image or adjacent wind turbine blade images.

2. The wind turbine blade fault detection method based on UAV and X-ray imaging according to claim 1, characterized in that, Step S2 also includes the following steps: S201. Establish a two-dimensional coordinate system with the length of the wind turbine blade to be tested as the first direction and the chord length as the second direction. S202. Based on the field of view of the X-ray detector and the preset overlap rate of adjacent images, calculate the required image acquisition point matrix in the first direction and the second direction, as well as the required number of images to be acquired at each acquisition point.

3. The method for wind turbine blade fault detection based on UAV and X-ray imaging according to claim 1, characterized in that, Generating a drone flight plan includes the following steps: S301. The coordinate position of the center of the wind turbine blade shaft is calculated based on the coordinate position of the wind turbine tower base to be tested and the wind turbine structure type. S302. Preset the attenuation level of X-rays, and calculate the distance between the X-ray generator and the X-ray detector according to Beer-Lambert's law, the expression is: ； in, Indicates the linear attenuation coefficient. This indicates the thickness of the wind turbine blade material to be tested. Indicates the initial X-ray intensity of the X-ray generator. This indicates the intensity of X-rays received by the X-ray detector. This indicates the degree of X-ray attenuation; the distance between the X-ray generator and the X-ray detector is maintained between 10 and 12 meters. The first and second drones automatically maintain a distance from the wind turbine blades to be tested based on position sensors.

4. The method for wind turbine blade fault detection based on UAV and X-ray imaging according to claim 1, characterized in that, After the position sensor of the first UAV detects that the wind turbine blade under test has entered the field of view of the X-ray generator, the X-ray generator calculates the lateral acquisition frequency of the wind turbine blade under test based on the linear velocity of the wind turbine blade under test, including the image exposure time and image interval. Calculate the linear velocity of the wind turbine blades under test at the current data acquisition point. Includes the following steps: The position sensor of the first UAV starts timing from the moment the first blade of the wind turbine under test passes through the X-ray generator's field of view at the current acquisition point, and waits to record the time points when the second and third blades enter the X-ray generator's field of view. , That is, the time it takes for the three blades of the wind turbine under test to traverse the current data acquisition point in a 240° arc, and the linear velocity of the wind turbine blades at the current data acquisition point. for: ； in, This indicates the distance from the center of the impeller of the fan under test to the current data collection point.

5. The method for wind turbine blade fault detection based on UAV and X-ray imaging according to claim 1, characterized in that, The unsupervised image deblurring method employs an optimization approach that combines the Latent Space Model (DIP) with the Unsharp Masking Image Sharpening Method (UMFS). The expression for the DIP iterative convergence loss function is as follows: ； in, The loss function represents the convergence of the iteration. , These represent the weight coefficients of the original image and the target image, respectively. This represents the initial reconstructed image generated by the encoder-decoder process, representing random noise variables. This represents the original "blurred" image as input. This represents the sharpened target image obtained after unsharpening masking, achieved by adjusting the loss weights. and Controlled image reconstruction The balance between fidelity and sharpening intensity; Image reconstruction is initialized using forward propagation with random noise. The difference between the original image and the target image is calculated based on the loss function, and the reconstructed image is updated via backpropagation. This process is repeated to seek a clear image; the loss function serves as a priori to guide the optimization process toward convergence and reconstruction of the image space.

6. The method for wind turbine blade fault detection based on UAV and X-ray imaging according to claim 5, characterized in that, Image stitching based on the location information of the acquisition points and image acquisition parameters includes the following steps: The initial distance between the two images is calculated based on the acquisition point location, image acquisition time, drone attitude angle, and the current wind turbine blade linear velocity at the acquisition point when the first drone acquires two adjacent X-ray images. Homography matrix; With the initial Starting with the homography matrix, the Powell optimization method combined with cross-correlation is used to maximize the cross-correlation value between two images. A one-dimensional search is performed in the homography matrix space along each direction, and the resulting displacement is used to correct and optimize the obtained homography matrix. The optimized final homography matrix is ​​used to project the image onto the final stitching plane, thus completing the image stitching.

Images