Unmanned aerial vehicle based offshore wind turbine blade damage detection system

Abstract

The application belongs to the technical field of wind turbine damage detection, and specifically discloses an offshore wind turbine blade damage detection system based on a UAV, which comprises a detection subarea determination module, a UAV acquisition control module, a damage image acquisition module, a wind turbine damage analysis module and a wind turbine damage feedback terminal. The application preliminarily locates damage points and blade areas by using a UAV and divides the wind turbine array into subareas, changes the difficulty of traditional detection methods in the high wind speed environment at sea, can flexibly cope with the complex offshore environment, adjusts the shooting interval and shooting time point of the UAV by combining the spatial coordinates of the damage points and the real-time blade speed and real-time offshore wind conditions to generate shooting parameter instructions, breaks the limitations of traditional fixed angle monitoring and monitoring dead angles and restrictions, and finally can more comprehensively and accurately evaluate the damage degree by analyzing cracks at different depths of optical images and ray images.

Description

Technical Field

[0001] This invention belongs to the field of wind turbine damage detection technology, and more specifically, relates to a damage detection system for offshore wind turbine blades based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Under the dual-carbon development goals, offshore wind power, as an important clean energy source, has developed rapidly. However, the offshore environment is complex, and wind turbine blades are constantly affected by various factors such as high humidity, salt spray, strong ultraviolet radiation, rain, snow, and typhoons, which can cause varying degrees of corrosion and wear, thus affecting the power generation efficiency of wind turbine units. Therefore, damage detection of wind turbine blades is necessary.

[0003] Existing technologies, such as the wind turbine blade damage detection method, system, storage medium, and electronic device disclosed in Chinese invention patent application number 202410421837.7, use a non-contact acoustic signal acquisition array to collect the sound emitted by the blades during wind turbine operation, and then analyze these sound signals to determine the damage condition. This solves the problem of traditional detection requiring shutdown to deploy sensors, improves operation and maintenance efficiency, and reduces costs. It enables rapid detection of damage type and location, and allows for overall inspection of large blades without requiring shutdown.

[0004] Existing technologies, such as the online damage detection method for wind turbine blades based on a two-axis gimbal disclosed in Chinese invention patent application number 202210671479.6, utilize a two-axis gimbal equipped with an improved camera to simultaneously output video streams and images in continuous shooting mode. It then precisely stitches the blade images to form a high-resolution panoramic image (1 billion pixels), which is then used in conjunction with the gimbal to track blade rotation. This solves the problems of image stitching and tracking speed in traditional visual inspection, achieving high-precision damage detection, while also being compact, lightweight, and low-cost.

[0005] The first method employs a non-contact acoustic signal acquisition array and signal analysis technology, while the second method utilizes an improved high-speed visual acquisition system, two-axis gimbal precision motion control, and ultra-high resolution image stitching and angular velocity fusion analysis technology. Both aim to address the shortcomings of traditional wind turbine blade damage detection methods, such as the need for downtime, low efficiency, high cost, and difficulty in comprehensive and rapid detection. However, it is clear that both current detection methods have certain limitations, specifically in the following aspects: 1. Neither method considers the specific environmental interference caused by certain environments. For example, high wind speeds at sea can severely interfere with acoustic signal acquisition, and high wind speeds at sea can cause significant shaking of wind turbine blades, making accurate positioning difficult and resulting in insufficient environmental adaptability.

[0006] 2. Accessibility and coverage are still somewhat lacking. Currently, monitoring and damage determination are based on a fixed perspective. However, the degree of damage varies in different parts, making it difficult to cover areas such as the leaf tip and leaf root. This creates blind spots in the detection, and damage is easily missed.

[0007] 3. The current damage localization accuracy depends on the array density, which leads to a certain deviation between the final result and the actual situation. Summary of the Invention

[0008] In view of this, in order to solve the above problems, a damage detection system for offshore wind turbine blades based on unmanned aerial vehicles is proposed.

[0009] The objective of this invention can be achieved through the following technical solution: This invention provides a damage detection system for offshore wind turbine blades based on unmanned aerial vehicles (UAVs). The system includes: a detection zone determination module, which uses UAVs to initially locate damage points and damaged blade areas during initial cruising, combines data from acoustic sensors and infrared cameras to calibrate the spatial coordinates of damage points and damage risk coefficients, and divides the wind turbine array into zones.

[0010] The UAV acquisition and control module dispatches corresponding UAV groups based on the zoning results. At the same time, it adjusts the shooting spacing and shooting time of the UAVs by combining the spatial coordinates of the damage point, real-time blade rotation speed and real-time sea wind conditions, and generates shooting parameter instructions.

[0011] The damage image acquisition module acquires optical and X-ray images of the wind turbine based on the shooting parameter commands.

[0012] The wind turbine damage analysis module is used to analyze the degree of damage based on the optical and X-ray images and match the corresponding damage levels.

[0013] The wind turbine damage feedback terminal generates a wind turbine damage detection report based on the damage level and degree of damage in each area, and then provides feedback.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention uses drones to initially cruise and locate the damage point and blade area, and marks the spatial coordinates of the damage point and the damage risk coefficient before dividing the wind turbine array into areas. This changes the predicament of traditional detection methods being easily interfered with in the high wind speed environment at sea. It can flexibly cope with the complex marine environment, is no longer severely interfered with by high wind speed on the acquisition of acoustic signals, and also avoids the problem of difficulty in accurate positioning due to large blade sway.

[0015] (2) This invention generates shooting parameter instructions by combining the spatial coordinates of the damage point with the real-time blade rotation speed and real-time sea wind conditions to adjust the shooting distance and shooting time of the UAV. This breaks the limitations and blind spots of traditional fixed-viewpoint monitoring, and enables comprehensive detection. It greatly improves the accessibility and coverage of detection, thereby effectively avoiding the occurrence of damage omissions and ensuring that damage to all parts of the wind turbine blade can be detected in a timely manner.

[0016] (3) This invention acquires optical and X-ray images of the wind turbine and analyzes the degree of damage. It no longer relies solely on array density for damage location. Furthermore, through the mutual supplementation and precise analysis of optical and X-ray images, the location and degree of damage can be determined more accurately, effectively reducing the deviation between the final result and the actual situation and significantly improving the accuracy of wind turbine damage location.

[0017] (4) This invention can accurately identify crack information at different depths by analyzing optical and X-ray images of cracks at different depths. It effectively overcomes the limitation of traditional detection methods that can only detect surface damage. It can not only determine the surface characteristics of the damage, but also analyze the extension of the crack inside the blade, thereby enabling a more comprehensive and accurate assessment of the degree of damage. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the system module connections of the present invention.

[0020] Figure 2 This is a schematic diagram of the overall implementation process of the present invention.

[0021] Figure 3 This is a schematic diagram of the shooting distance adjustment process of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 and Figure 2As shown, the present invention provides a damage detection system for offshore wind turbine blades based on unmanned aerial vehicles (UAVs). The system includes: a detection zone determination module, a UAV acquisition and control module, a damage image acquisition module, a wind turbine damage analysis module, and a wind turbine damage feedback terminal.

[0024] In the above, the UAV acquisition and control module is connected to the detection zone determination module and the damage image acquisition module, respectively, and the wind turbine damage analysis module is connected to the damage image acquisition module and the wind turbine damage feedback terminal, respectively.

[0025] The detection zone determination module uses a drone to initially locate the damage point and the damaged blade area, combines data from acoustic sensors and infrared cameras to calibrate the spatial coordinates of the damage point and the damage risk coefficient, and divides the wind turbine array into zones.

[0026] This invention utilizes drones for initial cruise to locate damage points and blade areas, and after marking the spatial coordinates of the damage points and the damage risk coefficient, divides the wind turbine array into zones. This overcomes the predicament of traditional detection methods being easily interfered with in high-wind-speed environments at sea, and can flexibly cope with complex marine environments. It is no longer severely interfered with by high wind speeds on acoustic signal acquisition, and also avoids the problem of difficulty in accurate positioning due to large blade swaying.

[0027] Specifically, the specific calibration process of the spatial coordinates of the damage point includes: A1, performing noise suppression and enhancement processing on the acoustic signal recorded in the initial cruise, and calculating the signal phase shift and measuring the arrival time difference after receiving the time difference of the same damage sound source through the acoustic sensor array.

[0028] A2. Based on phase shift and time difference of arrival, the azimuth angle of the damage in the circumferential direction of the blade is determined by geometric positioning.

[0029] A3. Deblur and segment temperature anomaly regions in the infrared image to identify the location of damage.

[0030] A4. Combining the real-time positioning and attitude angle of the UAV, the damaged location is converted into three-dimensional spatial coordinates of the blade.

[0031] A5. Dynamically compare the azimuth angle with the three-dimensional spatial coordinates, and output the final spatial coordinates of the damage point through error correction.

[0032] It should be added that, based on the phase difference of the signals collected by each microphone channel of the acoustic array, and combined with the actual measured time difference, the arrival time delay of each channel signal can be calculated through the cross-correlation function. Then, according to the spatial topology of the acoustic array, such as the radius r of the ring array, triangulation or polygonal positioning geometric algorithms are used to solve the relationship between time delay and sound speed, that is, the distance difference is the product of sound speed and time delay, and an azimuth angle calculation model is constructed. Finally, the azimuth angle of the damage in the circumferential direction of the blade can be determined. The cross-correlation function, geometric algorithm, etc. are existing functions, and their specific formulas will not be shown.

[0033] It should be added that blind deconvolution algorithms, such as the Richardson-Lucy algorithm, can be used to address the motion blur problem in infrared images caused by drone flight. This algorithm estimates the blur kernel function and iteratively inverts the sharp image. First, an initial sharp image estimate is set. Then, a theoretical blurred image is calculated based on the blur model and compared with the actual blurred image to obtain the error. The error is then back-projected to update the sharp image estimate. This process is repeated until the image edge sharpness reaches a preset threshold, such as a 30% increase in gradient magnitude.

[0034] It should be noted that the blind deconvolution algorithm is an existing algorithm, and its execution principle and the estimation of fuzzy kernel functions and fuzzy models involved in the execution process all adopt existing technologies, so they will not be described in detail here.

[0035] Understandably, temperature anomaly region segmentation can employ adaptive thresholding algorithms such as the Otsu method. This method first statistically analyzes the image's gray-level histogram and automatically calculates the optimal segmentation threshold by maximizing the inter-class variance, dividing the image into background and temperature anomaly regions. For infrared images under complex sea conditions, combining prior knowledge of blade temperature (i.e., the temperature distribution range during normal operation), a dynamic threshold adjustment mechanism is introduced. A Gaussian mixture model is used to fit the temperature distribution curve, marking regions deviating from the mean by more than 2.5 standard deviations as temperature anomalies. Simultaneously, morphological operations, such as corrosion and dilation, remove noise interference, ultimately segmenting the temperature anomaly regions on the blade surface. The Gaussian mixture model is an existing model; its specific execution process will not be elaborated further.

[0036] Understandably, the real-time positioning of a drone is recorded by the GPS positioning device on board the drone, which consists of longitude, latitude, and altitude.

[0037] Preferably, the temperature gradient difference at different depths of the crack can be utilized based on the theory of heat conduction. For example, the temperature decreases by 0.3-0.5℃ for every 1mm increase in depth, and the crack depth can be inverted by the temperature field distribution of infrared thermal imaging. For the corrosion area, the temperature anomaly region is first segmented using a Gaussian mixture model, then the number of pixels in the corrosion region in the binarized image is counted, and combined with the image resolution, it is converted into the actual corrosion area. Finally, image features such as crack length, depth, and corrosion area are obtained. The theory of heat conduction and the Gaussian mixture model are existing knowledge theories and processing methods, and will not be elaborated further.

[0038] In one specific embodiment, the final output process of the damage spatial coordinates includes: obtaining the azimuth of the damage sound source through acoustic array beamforming and time-of-flight positioning technology; simultaneously constructing a three-dimensional spatial coordinate system for the blade by combining the accuracy of the UAV's longitude, latitude, altitude, pitch angle, roll angle, and yaw angle; and mapping the azimuth data and the two-dimensional damage location detected by vision to the three-dimensional space through a coordinate transformation algorithm. A dynamic error correction model is established using a Kalman filter algorithm, with the wind turbine rotation phase and sea wind speed disturbance as state variables, to iteratively correct the deviation between the acoustic signal azimuth and the visual three-dimensional coordinates. The iteration period can be set to 100ms. Finally, the error-compensated damage spatial coordinates are output. The time-of-flight positioning technology and the Kalman filter algorithm are existing technologies and will not be described in detail.

[0039] Specifically, the calibration process of the damage risk coefficient includes: B1, calculating the straight-line distance between the damage location point and the root of the wind turbine blade based on the spatial coordinates of the damage point, and using the ratio of the straight-line distance to the straight-line distance between the tip of the wind turbine blade and the root of the wind turbine blade as the location risk factor.

[0040] B2. The center frequency and impulse factor of the acoustic signal are extracted as sound features by short-time Fourier transform, and the crack length, crack depth and corrosion area of ​​the damage in the infrared image are identified as image features.

[0041] B3. After normalizing the sound features and image features respectively, the sound representation anomaly and visual representation anomaly are obtained by linear weighted summation.

[0042] B5. Import the location risk factor, sound representation anomaly degree, and visual representation anomaly degree into the multi-factor coupling model to output the damage risk coefficient.

[0043] In one specific embodiment, the multi-factor coupling model is specifically represented by the following formula: , Indicates the damage risk coefficient. , and These represent the location risk factor, the degree of abnormality in sound representation, and the degree of abnormality in visual representation, respectively. This represents the location factor coefficient, with a default value of 0.5. This is the non-linear correction exponent, with a default value of 1.2. The value reflects the leaf tip effect. This represents the anomaly gain coefficient, with a default value of 0.8. and These represent the contribution weights of sound representation anomaly degree and visual representation anomaly degree to the anomaly gain, respectively.

[0044] It is important to note that when hour Increased to 0.7 to reinforce positional risk, when hour, Upgraded to version 1.0, at this point, visual confirmation of severe surface defects is achieved. Furthermore, the contribution weights of acoustic and visual representation anomalies to the anomaly gain in the multi-factor coupling model can be determined by referring to the weight determination method of the probe fit function. For example, using historical wind turbine blade damage data as training samples, covering acoustic and visual representation anomalies under different damage conditions and corresponding actual damage risk cases, regression analysis or machine learning algorithms are used in conjunction with data such as damage risk coefficients and location risk factors. After iterative optimization through simulation and actual verification, the output determines the weight values ​​that accurately reflect the contribution of the two to the anomaly gain.

[0045] Understandably, normalization can be performed using the Min-Max standardization method.

[0046] In another specific instance, the process of dividing the wind turbine array into zones includes: C1, matching the damaged blade area and risk coefficient of each wind turbine with a preset risk category mapping table to obtain an initial risk category.

[0047] C2. If the initial risk category of a certain wind turbine is inconsistent with that of the wind turbines on both sides, it shall be adjusted according to the following rules, and the wind turbines with the same risk category after adjustment shall be classified into high-risk area, medium-risk area and normal area. The rule is: when the categories of both sides are higher than that of the wind turbine, the category of the wind turbine shall be adjusted to the lower one of the two sides.

[0048] When both categories are lower than or one of the categories is lower than that of the fan, the category of the fan on the lower side shall be adjusted to that of the fan.

[0049] In one specific embodiment, the data in the risk category mapping table can be set by integrating records of damage frequency, maintenance cost, and downtime of multiple blades in various regions from the offshore wind turbine operation and maintenance database over the past 5 years, as well as material strength parameters and load limits for blade tips, blade roots, and other parts specified in the blade structure design manual. It can also be combined with stress distribution data under different damage conditions generated by ANSYS finite element simulation and threshold standards developed by multiple industry experts based on practical experience in multiple offshore wind farms.

[0050] The table stores a cross-mapping relationship with the damaged area including leaf tip, leaf root, leaf middle, leading edge and trailing edge as the row dimension and the risk coefficient threshold range as the column dimension. The risk category mapping table sets different risk coefficient levels according to the structural characteristics and load differences of leaf tip, leaf root, leaf middle, leading edge and trailing edge.

[0051] For example, based on historical data and risk coefficients, when the risk coefficient is greater than or equal to 0.7, 85% of tip damage will cause failure within 6 months. 0.7 is set as the dividing line for high risk of tip damage. That is, when the risk coefficient is greater than or equal to 0.7, the high-risk category is output according to the table; when the risk coefficient is between 0.5 and 0.7, the medium-risk category is output according to the table; and when the risk coefficient is less than 0.5, the low-risk category is output.

[0052] According to finite element simulation, when the blade root stress exceeds the design limit by 30%, the corresponding risk coefficient is greater than or equal to 0.8. 0.8 is set as the dividing line for high risk of blade root stress. That is, when the risk coefficient is greater than or equal to 0.8, the high risk category is output according to the table. When the risk coefficient is between 0.5 and 0.8, the medium risk category is output according to the table. When the risk coefficient is less than 0.5, the low risk category is output.

[0053] The leaf mid-area is located between the leaf tip and leaf root. Based on empirical data analysis, 0.75 is set as the dividing line for high risk in the leaf mid-area. That is, when the risk coefficient is greater than or equal to 0.75, the high-risk category is output according to the table; when the risk coefficient is between 0.5 and 0.75, the medium-risk category is output according to the table; and when the risk coefficient is less than 0.5, the low-risk category is output.

[0054] The leading edge is sensitive to airflow impact. A risk coefficient greater than or equal to 7 is matched to the table to output a high-risk category; a risk coefficient between 0.5 and 0.7 is matched to the table to output a medium-risk category; and a risk coefficient less than 0.5 is matched to the table to output a low-risk category. The trailing edge, located at the blade tip, is affected by vortices generated by airflow separation during rotation, easily forming alternating bending stress and vibration loads. However, the trailing edge structure is usually wider and thicker than the leading edge, and the high-risk risk coefficient classification boundary can be slightly exceeded by the leading edge. For example, a trailing edge risk coefficient greater than or equal to 7.2 is matched to the table to output a high-risk category; a risk coefficient between 0.5 and 0.72 is matched to the table to output a medium-risk category; and a risk coefficient less than 0.5 is matched to the table to output a low-risk category.

[0055] Understandably, the risk categories consist of high risk, medium risk, and low risk, with the high-risk category corresponding to the high-risk area, the medium-risk category corresponding to the medium-risk area, and the low-risk category corresponding to the normal area.

[0056] The UAV acquisition and control module dispatches corresponding UAV groups based on the zoning results, and adjusts the shooting spacing and shooting time of the UAVs by combining the spatial coordinates of the damage point, real-time blade rotation speed and real-time sea wind conditions, and generates shooting parameter instructions.

[0057] This invention combines the spatial coordinates of the damage point with real-time blade rotation speed and real-time sea wind conditions to adjust the shooting distance and shooting time of the UAV to generate shooting parameter instructions. This breaks through the limitations and blind spots of traditional fixed-viewpoint monitoring, enabling comprehensive detection and greatly improving the accessibility and coverage of detection. This effectively avoids the occurrence of missed damage detection and ensures that damage to all parts of the wind turbine blade can be detected in a timely manner.

[0058] Understandably, for units in high-risk areas, dedicated drones equipped with X-ray detectors can be dispatched; for units in medium-risk areas, high-precision optical inspection drones can be dispatched; and for units in regular areas, standard inspection drones can be dispatched.

[0059] It is also understood that the system of the present invention can be connected to an external wind turbine SCADA system, and can read the real-time blade rotation speed from the wind turbine SCADA system. At the same time, it can collect real-time sea wind condition data through the high-precision wind speed sensor and wind direction instrument installed on the wind turbine. The sea wind condition consists of wind direction and wind speed.

[0060] Specifically, please refer to Figure 3 As shown, the adjustment of the drone's shooting distance includes: D1, statistical average and maximum sea wind speed, with the real-time wind direction standard deviation as the wind direction fluctuation degree.

[0061] D2. Aggregate historical sea wind speed data by hourly granularity to generate daily wind speed variation curves, and simultaneously construct the current sea wind speed variation curve based on the current real-time wind speed.

[0062] D3. Calculate the sea state disturbance coefficient by combining the average sea wind speed, the maximum sea wind speed, the wind direction fluctuation, and the wind speed change curve.

[0063] D4. Based on the spatial coordinates of the damage point, the distance from the center of the hub to the damage point is calculated using a specific formula between the two points. This distance is used as the radius of the damage location. The linear velocity of the damage point is obtained by multiplying the current speed of the wind turbine by the radius of the damage location.

[0064] D5. Based on the basic safety time margin and basic safety distance of the damaged blade area, the product of the basic safety time margin and the linear velocity of the damaged point is used as the safety compensation distance.

[0065] D6. If the sea state interference coefficient is less than the preset threshold, adjust the shooting distance to the sum of the basic safety distance and the compensation distance; otherwise, adjust the shooting distance by correcting the sum using the sea state coefficient.

[0066] Understandably, the formula for the distance between two points is a fairly common formula, and will not be shown here.

[0067] Understandably, for different damaged blade regions, different basic safety time margins and basic safety distances can be pre-set through comprehensive analysis of the blade's structural design drawings, material mechanical property testing, and risk-based operational safety standards. For example, the basic safety time margin for the blade tip region is 2 seconds, and the basic safety distance is 5 meters. The basic safety time margin for the blade root region is 1 second, and the basic safety distance is 3 meters. Multiplying the basic safety time margin by the linear velocity of the damaged point yields the safety compensation distance, which is used to compensate for the safety risks caused by the movement of the damaged point.

[0068] It is important to note that in actual operation, it is also necessary to observe the inspection situation of the UAV under different sea conditions and different blade operating conditions, analyze whether the safety distance and time margin between the UAV and the blade are reasonable, and conduct simulation tests to set different working conditions in the laboratory or on site to test and verify the obstacle avoidance performance of the UAV and the motion characteristics of the blade, so as to correct and improve the basic safety distance and time margin.

[0069] Understandably, the specific correction method for adjusting the sum of the basic safety distance and the safety compensation distance using the sea state coefficient is to multiply the sum of the basic safety distance and the safety compensation distance by 1 and then by the sum of the sea state coefficient. Furthermore, to ensure image quality and avoid motion blur or image overlap issues, the anti-blur distance can be derived based on camera parameters and set as the upper limit for adjusting the shooting distance; that is, the adjusted shooting distance must not exceed this upper limit.

[0070] Furthermore, the calculation of the sea state interference coefficient in step D3 includes: D31, aligning the daily wind speed change curve with the current sea wind speed change curve on the time axis, and calculating the ratio of the overlap length of the two curves to the total length of the current curve to obtain the overlap ratio of the current wind speed change.

[0071] D32. If the overlap ratio exceeds the corresponding preset threshold, the portion after the current time point is extracted from the daily wind speed change curve, and the proportion of the curve segment length that is higher than the reference interference wind speed in this portion is used as the predicted wind speed over-limit ratio; otherwise, the predicted wind speed over-limit ratio is assigned a value of 0.

[0072] D33. Locate the average sea wind speed and the maximum sea wind speed from the current sea wind speed variation curve, and mark them as the average point and the threshold point, respectively.

[0073] D34. Based on the location of the average point and the threshold point and the predicted wind speed over-limit ratio, confirm the target sea wind speed and calculate the degree of deviation of the target sea wind speed from the set reference disturbance wind speed.

[0074] D35. After normalizing the deviation and wind direction fluctuation, import them into the Sigmoid function to output the sea state disturbance coefficient.

[0075] It should be added that the degree of deviation can be calculated by the difference between the target sea wind speed and the set reference interference wind speed. If the difference is less than or equal to 0, the degree of deviation is directly assigned to 0. If the difference is greater than 0, the ratio of the difference to the set reference interference wind speed is used as the specific value of the degree of deviation.

[0076] Furthermore, confirming the target sea wind speed includes: calculating the slope of the wind speed change curve as the wind speed change rate.

[0077] If the rate of change of wind speed is greater than 0, the highest wind speed will be selected as the candidate wind speed.

[0078] The wind speed development smoothness ratio is obtained by comparing the distance between the statistical average point location and the threshold point location with the distance between the start and end points of the curve.

[0079] The proportion of the length of the curve segment where the wind speed is greater than the average wind speed after the average point is calculated to the total length of the subsequent curves.

[0080] The wind speed setting compensation factor is obtained by combining the wind speed development smoothness ratio and the wind speed exceeding the average concentration ratio, and the target sea wind speed is obtained after compensating the candidate wind speed.

[0081] If the rate of change of wind speed is less than or equal to 0, the average wind speed will be used as the target sea wind speed.

[0082] Understandably, when the rate of change of wind speed is greater than 0, the overall curve shows an increasing trend. Under normal circumstances, it is unlikely that the average point position will be located before the threshold point position. Other instantaneous special cases are not considered by default in this invention.

[0083] It should be explained that when the rate of change of wind speed is greater than 0, it indicates that the wind speed is in an upward phase. To guard against the risk of potential extreme weather, the highest wind speed is used as the candidate wind speed. When the rate of change of wind speed is less than or equal to 0, it indicates that the wind speed is stable or decreasing. In this case, the average wind speed can represent the current and short-term normal, and it is directly used as the target sea wind speed.

[0084] Understandably, the ratio of the time span from the average state to the extreme state of wind speed development is used to measure the temporal characteristics of the wind speed increase process. The smaller the ratio, the faster the wind speed rises from the average to the maximum, indicating a more concentrated risk. The larger the ratio, the slower the wind speed increase process, and the longer the peak duration may be, meaning that high wind speeds last longer. This ratio directly affects the adjustment direction of the selected wind speed. A small ratio requires aggressive compensation to cope with short-term strong winds, while a large ratio allows for a more moderate adjustment to avoid overestimating the risk, ultimately making the target wind speed more closely match the actual risk scenario.

[0085] Understandably, by calculating the ratio of the length of the curve segment where the wind speed is greater than the average wind speed after the mean point to the total length after the mean point, focusing on the curve after the mean point, the persistence of high wind speed in subsequent periods can be assessed, providing a reference for adjusting the candidate wind speed.

[0086] In one specific embodiment, the formula for setting the wind speed compensation factor is as follows: , This indicates that a compensation factor is set for wind speed. and These represent the ratio of wind speeds with gradual development and the ratio of wind speeds exceeding the average concentration, respectively. and These represent the weighted proportions corresponding to the wind speed development level ratio and the wind speed exceeding the average concentration ratio, respectively.

[0087] Understandably, the weights corresponding to the wind speed development smoothness ratio and the wind speed exceeding the average concentration ratio need to be determined specifically based on the actual application scenario, such as offshore operation safety requirements and wind energy utilization goals, through historical data simulation verification, expert experience calibration, or machine learning optimization.

[0088] In another specific instance, the specific adjustment process of the shooting time includes the following steps: E1, matching the base time window according to the damaged blade area, and setting the wind condition compensation duration based on real-time wind speed and wind direction.

[0089] E2. Add the base time window to the wind compensation duration, and compare it with the corresponding preset minimum trigger threshold of the shooting device. Take the larger value as the final shooting time window.

[0090] E3. Obtain the initial synchronization time and real-time angular velocity of the blade rotation, calculate the blade rotation period, and extract the matching observation phase window of the corresponding region from the observation phase window and the blade region mapping table based on the damaged blade region.

[0091] E4. Determine the starting and ending angles based on the matched observation phase window, and calculate the first frame trigger time by combining the starting angle and the blade rotation period.

[0092] E5. Calculate the time interval of subsequent frames based on the trigger time of the first frame. The time interval is the ratio of the difference between the ending angle and the starting angle to twice the angular velocity.

[0093] E6. The adjusted shooting time is composed of the first frame trigger time, the time interval between subsequent frames, and the final shooting time window.

[0094] Understandably, the blade rotation period is The ratio of the blade rotation speed to the blade rotation speed.

[0095] It should be added that the observation phase window and blade region mapping table are determined based on the optimal observation phase window according to the motion characteristics and susceptibility to interference of different regions of the blade. For example, the tip region, due to its high linear velocity, is prone to image blurring, and its optimal observation phase window... Set as The mid-leaf region structure is relatively stable, and the optimal observation phase window is... The leaf root region, being close to the tower, is easily obstructed; the optimal observation phase window is... This provides a phase basis for calculating the shooting time point.

[0096] It should be added that different areas of the leaf have different requirements for the minimum effective shooting time due to differences in their motion characteristics. The leaf tip area has a high linear velocity and requires a very short time to freeze motion blur. The basic window formula is: , The distance from the damage point to the leaf root. This represents the total length of the wind turbine blades. The minimum clear imaging distance for leaf tip photography can be set to 0.5m. The mid-leaf region exhibits moderate motion characteristics, and the basic window formula is as follows: ,in, The minimum distance for capturing a clear image of the leaf area can be set to 1m. The set shooting duration for compensating for attitude changes can be 0.01s. The leaf root region has low linear velocity and stable motion. The basic window formula is... ,in, This is the minimum clear imaging distance for the leaf root region, which can be set to 2m.

[0097] Understandably, the first frame trigger time is the sum of the initial synchronization time, the integer number of complete cycles, and the time offset required from the zero position to the starting angle. The integer number of complete cycles is the product of the integer number of cycles and the blade rotation cycle. When the integer number of cycles is 0, it represents the current cycle, and when it is 1, it represents the next cycle. The time offset required from the zero position to the starting angle refers to the ratio of the starting angle to the blade angular velocity.

[0098] Understandably, the initial synchronization moment refers to the time marker point when the blade rotates to the preset zero angle, typically the vertically downward position. This moment serves as the alignment reference between the blade angle and the system time, and is used for calculating all trigger times.

[0099] In one specific embodiment, during image acquisition, three frames are captured consecutively for optical images. The first frame is triggered at the first frame trigger time, and the second frame is triggered after waiting for the calculated time interval. The third frame is triggered after waiting for the calculated time interval again. The three frames correspond to the starting angle, intermediate angle, and ending angle positions, respectively, and can completely cover the phase window. During X-ray imaging, the midpoint angle of the optimal observation phase window is used. The X-ray trigger time is the sum of the initial synchronization time, an integer number of complete cycles, and the time offset required from the zero position to the midpoint angle. The time offset required from the zero position to the midpoint angle refers to the ratio of the midpoint angle to the blade angular velocity. A single X-ray frame is triggered when the X-ray trigger time is reached.

[0100] Furthermore, the specific settings for setting the wind condition compensation duration in step E1 are as follows: If the current wind speed exceeds the preset disturbance wind speed, calculate the difference between the current wind speed and the preset disturbance wind speed to obtain the excess wind speed difference. Otherwise, the excess wind speed difference will be assigned a value of 0.

[0101] The wind direction fluctuation is denoted as Statistical wind condition compensation duration , , The compensation duration corresponding to the set unit over-limit wind speed difference. The compensation duration corresponding to the set unit wind direction fluctuation difference. To set the permissible wind direction fluctuation.

[0102] Understandably, the normal operating wind speed range for offshore wind turbines is generally 3-25 m / s. However, during blade inspections, wind speeds exceeding 10 m / s can cause increased blade vibration, necessitating compensation. Wind tunnel experiments and operational statistics indicate that 10 m / s is a critical value for significantly increased vibration in engineering practice. Therefore, 10 m / s is used as the preset disturbance wind speed. While offshore wind direction fluctuates randomly, historical data from the wind turbine's SCADA system shows a standard deviation of less than [a certain value]. At that time, the blade pitch system can adaptively compensate, when it exceeds... At this time, pitch control cannot completely offset attitude changes, requiring a longer shooting window to stabilize the image. Set to allowable wind direction fluctuation.

[0103] In one specific embodiment, high-speed camera testing shows that for every 1 m / s increase in wind speed, the blade vibration frequency increases by 0.5 Hz, resulting in freezing vibration blur. Experimental calibration shows that a 0.002 s window needs to be added for every 1 m / s increase, thereby... Assigned value Furthermore, based on engineering experience, it can be known that wind direction fluctuations exceed... At that time, the blade attitude change period is about 10 seconds, and for every 10° increase, a window of 0.001 seconds needs to be added, that is... Can be set to .

[0104] It should be noted that in one specific embodiment, the focal length of the optical camera decreases proportionally as the shooting distance increases; for example, if the distance increases by 10%, the focal length decreases by 15%.

[0105] The damage image acquisition module acquires optical and X-ray images of the wind turbine based on shooting parameter commands.

[0106] The wind turbine damage analysis module is used to analyze the degree of damage based on the optical image and the X-ray image, and match the corresponding damage level.

[0107] This invention, through optical and X-ray image acquisition and damage analysis of wind turbines, no longer relies solely on array density for damage localization. Furthermore, by complementing and precisely analyzing optical and X-ray images, the location and extent of damage can be determined more accurately, effectively reducing the deviation between the final result and the actual situation, and significantly improving the accuracy of wind turbine damage localization.

[0108] Specifically, the specific analysis process of the damage degree analysis includes: F1. For high-risk areas, optical and X-ray images of the blade are acquired simultaneously, and the geometric feature point set of the blade edge in the two images is extracted. The geometric feature point set consists of the curvature extreme points, chord length ratio division points and manufacturing process line intersections in the aerodynamic profile of the blade.

[0109] F2. Based on the feature point set, the affine transformation matrix is ​​fitted using the RANSAC algorithm to map the ray image to the optical image coordinate system.

[0110] F3. Enhance the optical and X-ray images respectively, and identify surface cracks, subsurface cracks, and penetrating cracks.

[0111] F4. Extract the corresponding crack assessment index for each type of crack and calculate the difference between it and the reference value of the corresponding crack assessment index in the crack type to which it belongs.

[0112] F5. The ratio of the number of crack assessment indicators with a statistical difference greater than 0 to the total number of crack assessment indicators is used as the crack risk ratio.

[0113] F6. For crack assessment indicators with a difference greater than 0, calculate the relative deviation value between the index and the corresponding reference value to obtain the crack risk rate. The product of the crack risk ratio and the crack risk rate is used as the damage coefficient of the corresponding crack.

[0114] F7. Based on the preset damage influence weight for each type of crack, the damage coefficient of each type of crack is weighted and summed to obtain a comprehensive damage coefficient, and the comprehensive damage coefficient is corrected by the damage risk coefficient to obtain the damage degree value.

[0115] F8. For the medium-risk area, acquire optical images of the blade and use the product of the preset influence weight of the surface crack and its damage coefficient as the damage degree value. For the normal area, assign the damage degree value to 0.

[0116] This invention, through analysis of cracks at different depths using optical and X-ray images, can accurately identify crack information at various depths. This effectively overcomes the limitations of traditional detection methods that can only detect surface damage. It can not only determine the surface characteristics of the damage but also analyze the extension of cracks within the blade, thus enabling a more comprehensive and accurate assessment of the damage extent.

[0117] It should be added that the specific execution process of the enhancement processing in step F3 is as follows: guided filtering and adaptive histogram equalization are performed on the optical image, and wavelet threshold denoising and pseudo-color mapping are performed on the ray image. The pseudo-color mapping refers to converting grayscale differences into color level differences, which can highlight internal defects. The enhancement processing methods all adopt existing methods, and their specific execution process will not be described in detail.

[0118] It should be added that in step F3, surface cracks are identified by extracting the crack contour with linear gray-level abrupt changes from the optical image, calculating the crack width using a skeletonization algorithm, and calculating the crack length using chain code tracing. If no corresponding gray-level anomaly is found in the ray image, it is determined to be a surface crack. Subsurface cracks are identified by using the blurred features at the crack ends in the optical image and the local gray-level reduction band in the ray image, and applying the ray attenuation formula. The depth d is calculated by reverse calculation, where The material attenuation coefficient is obtained through calibration tests. For material density, It is a natural constant. and These are the grayscale values ​​of the damaged and undamaged areas, respectively. If the depth is within a preset range, it is determined to be a subsurface crack; for example, the preset range is between 0.5mm and 3mm. Penetrating cracks are identified by recognizing low-density channels penetrating the X-ray image and cracks with a width greater than or equal to a preset depth threshold in the optical image. If such cracks exist, they are determined to be penetrating cracks, and the preset depth threshold can be, for example, a crack of 0.5mm.

[0119] Understandably, optical images primarily reflect the surface condition of the blade, with cracks appearing as linear gaps on the surface. Image processing techniques, such as edge detection and contour extraction, can be used to measure the width of the crack on the surface. When the crack width detected in the optical image is greater than or equal to 0.5 mm, it indicates that the crack opening on the surface has reached a certain extent. Combined with the characteristics of penetrating low-density channels in the X-ray image, it is possible to more accurately determine that the crack is a penetrating crack, meaning that the crack depth is large and has a serious impact on the structural strength of the blade.

[0120] In one specific embodiment, the crack width, total crack length, and the angle between the crack principal axis and the local chord line of the blade are used as crack evaluation indicators for surface cracks. This is achieved by identifying linear gray-scale abrupt change regions through optical image recognition, extracting the crack centerline using a skeletonization algorithm, and calculating the minimum width by dynamically scanning the edge distance along the centerline normal. The crack width is then determined by obtaining the continuous crack direction using chain code tracing technology, combining this with B-spline smoothing to fit the total centerline length, which is denoted as the total crack length. Principal component analysis is used to determine the crack principal axis direction and calculate its angle with the local chord line of the blade. Furthermore, the skeletonization algorithm, chain code tracing technology, B-spline smoothing, and principal component analysis all employ existing methods, and their specific execution processes will not be demonstrated or explained further.

[0121] Crack depth and horizontal diffusion area are used as crack evaluation indicators for subsurface cracks. Crack depth is calculated using the ray attenuation formula, and horizontal diffusion area is obtained by enhancing crack contrast through morphological top-hat transformation of X-ray images and by watershed segmentation and region growing extraction. Morphological top-hat transformation and watershed segmentation are existing techniques, and the specific execution process will not be described further.

[0122] Crack depth, crack volume, and the width of the opening on the opposite side are used as crack assessment indicators for penetrating cracks. Specifically, two X-ray images from different angles are input. Based on multi-angle X-ray projection, the crack depth is calculated using the geometric relationship between the two views. Projected images with a difference of 30 degrees or more are acquired, registered, and the crack projection length is measured. The true depth is then inverted using triangulation. Combining the orthogonal three views (projected areas at 0° / 45° / 90°), the crack volume is estimated using an ellipsoidal equivalent model. Simultaneously, the blade rotation sequence is analyzed. In the critical angle image where the opening on the back side is exposed, the width of the opening on the opposite side is measured by the width of the rising edge of the grayscale profile. The triangulation method and the ellipsoidal equivalent model for estimating crack volume are existing techniques and will not be explained in detail here.

[0123] In one specific embodiment, the damage severity value is obtained by modifying the comprehensive damage coefficient by multiplying the sum of 1 and the damage risk coefficient by the comprehensive damage coefficient.

[0124] In another specific embodiment, the preset damage impact weights are typically set based on the degree of harm that cracks pose to the structural safety and operational performance of the wind turbine blades. For example, penetrating cracks, because they directly pierce the blade structure and severely damage mechanical integrity, have the highest weight. Subsurface cracks, although not penetrating the surface, may expand into structural damage under alternating loads, and therefore have the next lowest weight. Surface cracks are mostly confined to the surface layer and have a relatively small impact on structural strength, so they have the lowest weight. In addition, the weights are also determined comprehensively based on factors such as the location of the crack, such as the blade tip, main beam, and other critical parts, and the size expansion trend, through engineering experience, failure mode analysis, or wind turbine design specifications, to quantify the maintenance priority of different crack types.

[0125] The wind turbine damage feedback terminal generates a wind turbine damage detection report based on the damage level and degree of damage in each area, and then provides feedback on the report.

[0126] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A damage detection system for offshore wind turbine blades based on unmanned aerial vehicles (UAVs), characterized in that, The system includes: The detection zoning module uses drones for initial cruise to locate damage points and damaged blade areas. It combines data from acoustic sensors and infrared cameras to calibrate the spatial coordinates of damage points and damage risk coefficients, and divides the wind turbine array into high-risk, medium-risk, and normal zones. The UAV acquisition and control module dispatches corresponding UAV groups based on the zoning results. At the same time, it adjusts the shooting distance and shooting time of the UAVs by combining the spatial coordinates of the damage point, real-time blade rotation speed and real-time sea wind conditions, and generates shooting parameter instructions. The damage image acquisition module acquires optical and X-ray images of the wind turbine based on the shooting parameter commands. The wind turbine damage analysis module is used to analyze the degree of damage based on the optical and X-ray images and match the corresponding damage levels. The wind turbine damage feedback terminal generates a wind turbine damage detection report based on the damage level and degree of damage in each area, and then provides feedback.

2. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 1, characterized in that: The specific calibration process for the spatial coordinates of the damage point includes: The acoustic signals recorded during the initial cruise are subjected to noise suppression and enhancement processing. The signal phase shift is calculated and the arrival time difference is measured after receiving the time difference of the same damaged sound source through the acoustic sensor array. Based on phase shift and time of arrival difference, the azimuth of the damage in the circumferential direction of the blade is determined by geometric positioning; Infrared images are deblurred and temperature anomaly regions are segmented to identify damage locations; By combining the real-time positioning and attitude angle of the UAV, the damaged location is converted into three-dimensional spatial coordinates of the blade; The azimuth angle is dynamically compared with the three-dimensional spatial coordinates, and the final spatial coordinates of the damage point are output through error correction.

3. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 1, characterized in that: The calibration process for the damage risk coefficient includes: The straight-line distance between the damaged point and the root of the wind turbine blade is calculated based on the spatial coordinates of the damaged point. The ratio of the straight-line distance to the straight-line distance between the tip of the wind turbine blade and the root of the wind turbine blade is used as the location risk factor. The center frequency and impulse factor of the acoustic signal were extracted by short-time Fourier transform as acoustic features, and the crack length, crack depth and corrosion area of ​​the damage in the infrared image were identified as image features. After normalizing the sound features and image features respectively, the sound representation anomaly and the visual representation anomaly are obtained by linear weighted summation respectively. The location risk factor, sound representation anomaly, and visual representation anomaly are imported into the multi-factor coupling model to output the damage risk coefficient.

4. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 1, characterized in that: The specific process of dividing the wind turbine array into zones includes: The damaged blade area and risk coefficient of each wind turbine are matched with a preset risk category mapping table to obtain the initial risk category; If a wind turbine's initial risk category differs from that of the wind turbines on either side of it, the following rules shall be applied for adjustment, and wind turbines with the same risk category after adjustment shall be classified into high-risk, medium-risk, and normal zones. The rules are as follows: If both categories are higher than the fan's category, the fan's category is adjusted to the lower of the two categories. When both categories are lower than or one of the categories is lower than that of the fan, the category of the fan on the lower side shall be adjusted to that of the fan.

5. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 1, characterized in that: The adjustment of the drone's shooting distance includes: The average and maximum sea wind speeds were statistically analyzed, and the wind direction fluctuation was expressed as the real-time standard deviation of wind direction. Historical sea wind speed data is aggregated at the hourly granularity to generate daily wind speed variation curves, while the current sea wind speed variation curve is constructed based on the current real-time wind speed. The sea state disturbance coefficient is calculated by combining the average sea wind speed, the maximum sea wind speed, the wind direction fluctuation, and the wind speed change curve. The distance from the damage point to the center of the hub is obtained based on the spatial coordinates of the damage point, and used as the radius of the damage location. The linear velocity of the damage point is obtained by combining the current speed of the wind turbine and the radius of the damage location. Based on the basic safety time margin and basic safety distance for matching the damaged blade region, the product of the basic safety time margin and the linear velocity of the damaged point is used as the safety compensation distance. If the sea state interference coefficient is less than the preset threshold, the shooting distance is adjusted to be the sum of the basic safety distance and the compensation distance; otherwise, the shooting distance is adjusted by correcting the sum using the sea state coefficient.

6. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 5, characterized in that: The calculation of the sea state disturbance coefficient includes: Align the daily wind speed variation curve with the current sea wind speed variation curve on the time axis, and calculate the ratio of the overlap length of the two curves to the total length of the current curve to obtain the current wind speed variation overlap ratio. If the overlap ratio exceeds the corresponding preset threshold, the portion after the current time point is extracted from the daily wind speed change curve, and the proportion of the curve segment length that is higher than the reference interference wind speed in this portion is counted as the predicted wind speed exceedance ratio; otherwise, the predicted wind speed exceedance ratio is assigned a value of 0. Locate the average sea wind speed and the maximum sea wind speed from the current sea wind speed variation curve, and mark them as the average point and the threshold point, respectively. The target sea wind speed is determined based on the location of the average point and the threshold point and the predicted wind speed exceedance ratio. The deviation of the target sea wind speed from the set reference disturbance wind speed is calculated. After normalizing the deviation and wind direction fluctuation, the values ​​are imported into the Sigmoid function to output the sea state disturbance coefficient.

7. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 6, characterized in that: The confirmed target sea wind speed includes: The slope of the wind speed change curve is calculated as the rate of wind speed change. If the rate of change of wind speed is greater than 0, the highest wind speed will be selected as the candidate wind speed. The wind speed development smoothness ratio is obtained by comparing the distance between the statistical average point location and the threshold point location with the distance between the start and end points of the curve. The proportion of the length of the curve segment with wind speed greater than the average wind speed after the average point is calculated to the total length of the subsequent curve. The wind speed setting compensation factor is obtained by combining the wind speed development smoothness ratio and the wind speed exceeding the average concentration ratio, and the target sea wind speed is obtained after compensating the candidate wind speed. If the rate of change of wind speed is less than or equal to 0, the average wind speed will be used as the target sea wind speed.

8. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 1, characterized in that: The specific process for adjusting the shooting time includes: The base time window is matched according to the damaged blade area, and the wind condition compensation duration is set based on real-time wind speed and wind direction. The base time window is added to the wind condition compensation duration and compared with the corresponding preset minimum trigger threshold of the shooting device. The larger value is taken as the final shooting time window. The initial synchronization time and real-time angular velocity of the blade rotation are obtained to calculate the blade rotation period, and the matching observation phase window of the corresponding region is extracted from the observation phase window and the blade region mapping table based on the damaged blade region. Determining the starting angle based on the matched observation phase window and ending angle The trigger time of the first frame is calculated by combining the starting angle and the blade rotation period. The time interval for subsequent frames is calculated based on the trigger time of the first frame, and the time interval is the ratio of the difference between the ending angle and the starting angle to twice the angular velocity. The adjusted shooting time is composed of the first frame trigger time, the time interval between subsequent frames, and the final shooting time window.

9. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 8, characterized in that: The specific settings for the wind condition compensation duration are as follows: If the current wind speed exceeds the preset disturbance wind speed, calculate the difference between the current wind speed and the preset disturbance wind speed to obtain the excess wind speed difference. Otherwise, the excess wind speed difference will be assigned a value of 0; The wind direction fluctuation is denoted as Statistical wind condition compensation duration , , The compensation duration corresponding to the set unit over-limit wind speed difference. The compensation duration corresponding to the set unit wind direction fluctuation difference. To set the permissible wind direction fluctuation.

10. The offshore wind turbine blade damage detection system based on unmanned aerial vehicles as described in claim 1, characterized in that: The specific analysis process for the damage severity analysis includes: For high-risk areas, optical and ray images of the blades are acquired simultaneously, and the set of geometric feature points of the blade edges in the two images is extracted; the geometric feature points are composed of the curvature extrema points, chord length ratio division points and manufacturing process line intersections in the aerodynamic profile of the blade; Based on the feature point set, the affine transformation matrix is ​​fitted using the RANSAC algorithm to map the ray image to the optical image coordinate system; The optical and X-ray images are enhanced separately, and surface cracks, subsurface cracks, and penetrating cracks are identified. For each type of crack, extract the corresponding crack assessment index and calculate the difference between it and the reference value of the corresponding crack assessment index in the crack type to which it belongs; The ratio of the number of crack assessment indicators with a statistical difference greater than 0 to the total number of crack assessment indicators is used as the crack risk ratio. For crack assessment indicators with a difference greater than 0, the relative deviation value between the index and the corresponding reference value is calculated to obtain the crack risk rate. The product of the crack risk ratio and the crack risk rate is used as the damage coefficient of the corresponding crack. The damage coefficient of each type of crack is obtained by weighting and summing the damage coefficients of each type of crack based on the preset damage influence weights. The damage degree value is then obtained by correcting the comprehensive damage coefficient with the damage risk coefficient. For the medium-risk area, optical images of the blades are collected, and the product of the preset influence weight of the surface cracks and its damage coefficient is used as the damage degree value. For the normal area, the damage degree value is assigned to 0.

Images