Unmanned aerial vehicle based wind turbine blade inspection method, system, device and medium

Abstract

The present application relates to the technical field of fan defect detection, and particularly relates to a fan blade inspection method based on a UAV, which comprises: obtaining fan static basic data; constructing a blade reference model based on the fan static basic data, and collecting dynamic vibration response data according to a preset short-time dynamic vibration excitation rule; identifying a blade high-risk area according to the dynamic vibration response data, and generating a layered circumferential flight path in combination with the blade reference model; controlling the UAV to collect blade multi-angle images according to the layered circumferential flight path, processing the blade multi-angle images to identify surface defects and generate visual confidence; mapping the dynamic vibration response data to the blade reference model by using a space-time alignment model, extracting power spectral density characteristics of corresponding positions, fusing the visual confidence to calculate a damage index, and outputting an inspection report. The purpose is to solve the technical problem that internal damage of a blade is difficult to identify in a static inspection mode.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fan defect detection, in particular to a fan blade inspection method, system, device and medium based on a UAV. BACKGROUND

[0002] At present, the fan blade inspection technology based on a UAV mainly realizes the state monitoring of the blade through two modes of dynamic inspection and static inspection. In the dynamic inspection mode, the UAV is equipped with an on-board computing unit and a posture sensing device, and the real-time analysis of the fan running posture is realized by means of an artificial intelligence algorithm, the flight route around the machine is dynamically planned, and the image of the blade is quickly collected without stopping the machine under the condition that the fan keeps normal power generation. The static inspection needs to stop the machine and lock the blade, the UAV approaches the blade surface to obtain high-resolution images, and computer vision technology is used to identify subtle defects.

[0003] In the above-mentioned prior art, although the dynamic inspection can obtain the running state information of the blade without affecting the power generation, it is difficult to realize the accurate identification of subtle defects due to the limitation of flight distance and motion blur interference; although the static inspection has high-precision detection capability for surface defects, it cannot associate and identify internal structural damage, and the maintenance personnel may ignore the internal risk of the blade according to the detection results under the static inspection mode, thereby leading to the lack of damage evolution trend prediction capability in the maintenance decision. SUMMARY

[0004] In order to solve the technical problem that it is difficult to identify the internal damage of the blade in the static inspection mode, the present application provides a fan blade inspection method, system, device and medium based on a UAV, and the technical solution adopted is as follows:

[0005] The technical solution of the first aspect of the present application provides a fan blade inspection method based on a UAV, which comprises:

[0006] Obtaining fan static basic data, including spatial positioning data, structural size data and blade material characteristic data;

[0007] Constructing a blade reference model based on the fan static basic data, and collecting dynamic vibration response data according to a preset short-time dynamic vibration excitation rule;

[0008] Identifying a high-risk area of the blade according to the dynamic vibration response data, and generating a layered circumnavigation route containing a basic route and high-risk area encryption waypoints in combination with the blade reference model;

[0009] In the static inspection mode, controlling the UAV to collect multi-angle images of the blade according to the layered circumnavigation route, processing and identifying surface defects of the multi-angle images of the blade, and generating visual confidence;

[0010] The dynamic vibration response data is mapped to the blade reference model by using a space-time alignment model, power spectrum density features of corresponding positions are extracted, and an injury index is calculated by fusing visual confidence;

[0011] The damage level is divided based on the injury index, and an inspection report is output.

[0012] Further, the blade reference model is constructed based on the static basic data of the fan, and the dynamic vibration response data is collected according to a preset short-time dynamic vibration excitation rule, including:

[0013] Based on the structural size data, a layered surrounding modeling flight path is planned, and the unmanned aerial vehicle is controlled to collect multi-angle images of the blade along the layered surrounding modeling flight path;

[0014] The multi-angle images are fused by a motion recovery structure algorithm to generate the blade reference model;

[0015] The fan is controlled to execute the preset short-time dynamic vibration excitation rule, and is started from a shutdown state to a preset rotating speed and stably runs for a preset time length, and then is gradually stopped, forming a vibration excitation process including start, steady state and stop;

[0016] During the vibration excitation process, vibration signals of the blade start-stop transient process are collected, and a dynamic vibration response data is formed by synchronizing the SCADA system timestamp.

[0017] Further, according to the dynamic vibration response data, a high-risk area of the blade is identified, and a layered surrounding flight path including a basic flight path and an encrypted flight point of the high-risk area is generated in combination with the blade reference model, including:

[0018] The dynamic vibration response data is subjected to frequency spectrum analysis, and the characteristic frequency of different regions of the blade and the power spectrum density under the corresponding frequency are extracted;

[0019] The power spectrum density of each region is compared with the preset health region power spectrum density threshold of the blade, and the region exceeding the power spectrum density threshold is marked as a high-risk area of the blade;

[0020] Based on the surface curvature and size characteristics of the blade reference model, a basic flight path covering the windward surface, the leeward surface and the leading edge of the blade is generated;

[0021] For the high-risk area of the blade, the density of flight points is increased on the basis of the basic flight path, forming a layered surrounding flight path with encrypted flight points.

[0022] Further, in the static inspection mode, the unmanned aerial vehicle collects multi-angle images of the blade according to the layered surrounding flight path, processes the multi-angle images of the blade to identify surface defects, and generates visual confidence, including:

[0023] In the static inspection mode, the unmanned aerial vehicle is controlled to collect multi-angle images of the windward surface, leeward surface and leading edge of the blade along a layered circumferential flight path, and spatial position information corresponding to each image is recorded synchronously;

[0024] Based on the inherent texture period and reflectivity characteristics in the blade material characteristic data, the area consistent with the inherent texture of the blade is filtered out, and the abnormal texture area is retained;

[0025] The abnormal texture area is processed by using multi-scale filtering, the defect contour is extracted, and the defect geometric feature is quantified;

[0026] According to the edge sharpness of the defect area and the contrast of the background area, the edge sharpness of the defect contour, and the defect geometric feature, a visual confidence is calculated.

[0027] Further, a space-time alignment model is used to map the dynamic vibration response data to the blade reference model, including:

[0028] According to the blade attitude parameters during the dynamic vibration response data acquisition, a blade space rotation matrix is constructed;

[0029] According to the spatial coordinates of the blade reference model and the installation position of the vibration sensor, the dynamic vibration response data is mapped to the spatial coordinates of the blade reference model through the blade space rotation matrix;

[0030] Based on the timestamp, the dynamic vibration response data and the position information of the blade reference model are synchronized to realize the correlation and alignment of the dynamic vibration response data and the blade space position.

[0031] Further, the power spectral density features of the corresponding positions are extracted, and the damage index is calculated by fusing the visual confidence, including:

[0032] In the blade characteristic frequency interval, the power spectral density integral values of the damage area and the preset healthy area corresponding to the blade surface defects at the corresponding positions are calculated respectively to obtain a vibration energy ratio;

[0033] According to the type of the surface defect, the weight coefficient of the visual confidence is adjusted;

[0034] The visual confidence and the vibration energy ratio are weighted and fused according to the adjusted weight coefficient to generate a damage index reflecting the damage degree of the surface and the internal damage.

[0035] Further, the damage index can be expressed as:

[0036]

[0037] In the formula, The damage index is represented by D; The visual confidence weight factor is represented by C; The visual confidence correction term based on the defect area is represented by A. Indicates the area of ​​the defect; This represents the preset correction coefficient, used to adjust the sensitivity of the impact of defect area on visual confidence. Indicates the characteristic frequency of the blade; Indicates the characteristic frequency bandwidth; Represents the power spectral density function of the damaged region; This represents the power spectral density function of the healthy region.

[0038] The second aspect of the present invention provides a wind turbine blade inspection system based on unmanned aerial vehicles (UAVs), employing the UAV-based wind turbine blade inspection method described in the first aspect of the present invention. The system includes:

[0039] The data acquisition module is configured to acquire static basic data of the wind turbine, including spatial positioning data, structural dimension data, and blade material property data.

[0040] The short-time dynamic vibration excitation module is configured to build a blade baseline model based on the static basic data of the wind turbine, and collect dynamic vibration response data according to the preset short-time dynamic vibration excitation rules;

[0041] The route planning module is configured to identify high-risk areas of the blade based on dynamic vibration response data, and generate a layered loop route including a basic route and encrypted waypoints in high-risk areas in combination with the blade reference model.

[0042] The visual inspection module is configured to control the UAV to collect multi-angle images of the blades according to the layered circling route in static inspection mode, process the multi-angle images of the blades to identify surface defects and generate visual confidence scores.

[0043] The fusion diagnostic module is configured to use a spatiotemporal alignment model to map dynamic vibration response data to the blade reference model, extract the power spectral density features at the corresponding locations, and fuse visual confidence to calculate the damage index.

[0044] The decision output module is configured to classify damage levels based on damage indices and output inspection reports.

[0045] The third aspect of the present invention provides an electronic device, the electronic device comprising: a processor and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the steps of the unmanned aerial vehicle-based wind turbine blade inspection method described in the first aspect of the present invention.

[0046] The technical scheme of the fourth aspect of the present application provides a computer readable storage medium, and the computer readable storage medium stores a program for implementing the unmanned aerial vehicle based fan blade inspection method.

[0047] The present application has the following advantages:

[0048] The unmanned aerial vehicle based fan blade inspection method provided by the present application integrates short-time dynamic vibration excitation and static visual inspection in time sequence, thereby breaking through the technical bottleneck of internal damage identification while retaining the high precision advantage of static inspection. First, a blade reference model is constructed based on fan static basic data, dynamic vibration response data is collected through a preset short-time dynamic vibration excitation rule, and high-risk areas of the blade are accurately located. Then, a layered and surrounding flight path is generated in combination with the reference model, multi-angle images are collected by encrypting flight points in the high-risk areas in the static inspection mode, surface defect identification and visual confidence quantification are realized, and finally, the dynamic vibration data are mapped to the reference model by using a space-time alignment model, power spectral density features and visual confidence are fused, and a damage index is calculated. The method establishes a correlation diagnosis mechanism for surface defects and internal damage, so that the static inspection has the perception ability of internal structural damage without the need to modify the fan equipment and without the increase of significant power generation loss, which helps to improve the comprehensiveness of operation and maintenance decisions. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical schemes and advantages of the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0050] Figure 1 The method flowchart of the unmanned aerial vehicle based fan blade inspection method provided by an embodiment of the present application;

[0051] Figure 2 The structural schematic diagram of the unmanned aerial vehicle based fan blade inspection system provided by an embodiment of the present application;

[0052] Figure 3 The panoramic schematic diagram of the leading edge of the blade provided by an embodiment of the present application;

[0053] Figure 4 The panoramic schematic diagram of the windward surface of the blade provided by an embodiment of the present application;

[0054] Figure 5A panoramic view of the back surface of a blade according to an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined inventive purpose, the specific embodiments, structures, features and effects of a wind turbine blade inspection method, system, device and medium based on a UAV according to the present application are described in detail as follows in combination with the drawings and preferred embodiments. Different "one embodiment" or "another embodiment" in the following description do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0057] The specific scheme of a wind turbine blade inspection method, system, device and medium based on a UAV provided by the present application is described in detail below in combination with the drawings.

[0058] Please refer to Figure 1 which shows a method flowchart of a wind turbine blade inspection method based on a UAV according to an embodiment of the present application, the method comprising:

[0059] Step S100: acquiring wind turbine static basic data, including spatial positioning data, structural size data and blade material characteristic data; wherein the spatial positioning data includes wind turbine center longitude and latitude coordinates and wind turbine hub center elevation; the structural size data includes blade length, curvature radius, nacelle size and tower diameter; the blade material characteristic data includes blade surface reflectivity and inherent texture period.

[0060] In some embodiments, the wind turbine hub center elevation may be obtained by RTK elevation measurement combined with tower height conversion, wherein is the ground elevation measured by RTK, is the tower height, i.e. the vertical distance from the ground to the hub center; the structural size data is obtained by layered and surrounding scanning of the UAV, the blade length is the straight-line distance from the blade root to the blade tip; the curvature radius is calculated by surface fitting of blade surface point cloud, the nacelle size and the tower diameter are obtained by point cloud boundary extraction, providing geometric parameters for blade reference model construction; in the blade material characteristic data, the surface reflectivity can be obtained by waveband collection and radiation calibration processing; the inherent texture period is obtained by Fourier transform on the high-definition image of the blade defect-free area , extract the spatial frequency corresponding to the maximum energy value in the Fourier spectrum , for subsequent distinguishing intrinsic texture and defect features, which can be expressed as:

[0061]

[0062] wherein, is the intrinsic texture period, that is, the minimum physical size of the repeated texture unit on the blade surface; is a two-dimensional Fourier transform operator, used to convert a spatial domain image into a frequency domain; is a gray-scale image of a defect-free area;

[0063] Step S200: constructing a blade reference model based on static basic data of the fan, and collecting dynamic vibration response data according to a preset short-time dynamic vibration excitation rule;

[0064] Step S200 specifically includes:

[0065] Step S210: planning a layered surrounding modeling flight path based on structural size data, and controlling the UAV to collect multi-angle images of the blade along the layered surrounding modeling flight path;

[0066] In some embodiments, the blade length direction is divided into a blade root section, a middle blade section and a blade tip section in a ratio of 3:4:3, and an independent surrounding layer is arranged in each section; the surrounding radius is dynamically adjusted according to the blade curvature to ensure that the distance between the UAV and the blade surface is constant; the flight path heading overlap rate is ≥80%, and the lateral overlap rate is ≥70%, to ensure the continuity of image stitching;

[0067] In some embodiments, as shown in Figures 3 to 5 , a camera is carried by the UAV to fly along the planned flight path, collect multi-angle images of the windward surface, the leeward surface and the leading edge of the blade, and record the GPS coordinates and the camera attitude of each image synchronously;

[0068] Step S220: generating a blade reference model by fusing multi-angle images through a motion recovery structure algorithm; specifically, the ORB algorithm is used to extract feature points such as blade edge inflection points, bolt holes and leading edge edges in each image, and the RANSAC algorithm is used to eliminate mis-matching point pairs to improve the matching degree of the same name points; then, the feature point matching results are taken as input, the rotation matrix and the translation vector of the camera are optimized through the bundle adjustment method, and the objective function can be expressed as:

[0069]

[0070] In the formula, is the rotation matrix of the camera; is the translation vector, describing the camera position; is a perspective projection function, used to project a three-dimensional point projected to a two-dimensional image point ; is the Laplacian of the image, i.e., the second-order differential of the image at the feature point , which reflects the degree of change in texture; is a regularization coefficient;

[0071] In some embodiments, based on the optimized camera pose, a three-dimensional point cloud of the leaf is generated by triangulation calculation; a UV parameterization method is used to fit the texture information of the multi-angle image to the surface of the triangular mesh model to generate a leaf reference model with high-definition texture;

[0072] Step S230: Control the fan to execute a preset short-time dynamic vibration excitation rule, start from a shutdown state to a preset rotating speed, and gradually stop after stable operation for a preset time length, to form a vibration excitation process including a start-up stage, a steady state stage, and a shutdown stage;

[0073] In some embodiments, the preset short-time dynamic vibration excitation rule includes:

[0074] In the start-up stage, the fan is controlled to start from a shutdown state at a preset acceleration, gradually increase the rotating speed to the rated rotating speed, and cover the transient response from the low rotating speed to the rated rotating speed;

[0075] In the steady state stage, the rated rotating speed is maintained for a preset time length to capture the steady state vibration characteristics of the blade under the design load;

[0076] In the shutdown stage, the rotating speed is gradually reduced to the shutdown state at a preset deceleration, covering the transition process from the rated rotating speed to the static state.

[0077] The entire process needs to ensure that the complete stages of start-up, steady state, and shutdown are included to avoid loss of power generation caused by long-time operation.

[0078] In some embodiments, the rotating speed change rate satisfies a preset acceleration constant ;

[0079] Step S240: In the vibration excitation process, the vibration signal of the blade including the start-stop transient process is collected, and the dynamic vibration response data is formed by synchronizing the SCADA system timestamp; specifically, the vibration sensor pre-installed in the fan cabin, such as the IEPE acceleration sensor installed on the main shaft bearing seat, is used to directly collect the structural vibration signal of the blade transmitted to the cabin;

[0080] In some embodiments, the real-time rotating speed and the timestamp are obtained through the SCADA system, the vibration signal is bound with the timestamp, and the vibration time domain signal and the start-up, steady state, and shutdown state labels are stored to form the dynamic vibration response data;

[0081] In some embodiments, the vibration data is synchronized with the timestamp of the SCADA system Real-time rotation speed Blade rotation angle Binding, forming dynamic vibration response data:

[0082]

[0083] In the formula, This refers to the moment when the vibration sensor data was acquired. , , The acceleration is the three-cycle vibration. Yaw angle, typically the azimuth angle of the wind turbine nacelle relative to true north; The blade pitch angle is the angle of rotation of the blade around its own axis.

[0084] This step generates a blade baseline model through layered surround modeling, providing a precise three-dimensional spatial and textural baseline for static inspection, ensuring the accuracy of surface defect location and feature extraction; by pre-setting short-time dynamic vibration excitation and response data acquisition, the structural vibration characteristics of the blade under all operating conditions are captured without affecting the normal power generation of the wind turbine, providing a dynamic response basis for internal damage identification.

[0085] Step S300: Identify high-risk areas of the blade based on dynamic vibration response data, and generate a layered circumnavigation route including a basic route and encrypted waypoints in high-risk areas by combining the blade reference model.

[0086] Step S300 specifically includes:

[0087] Step S310: Perform spectral analysis on the dynamic vibration response data to extract the characteristic frequencies of different regions of the blade and the power spectral density at the corresponding frequencies;

[0088] In some embodiments, based on the time-domain vibration acceleration signal acquired in step S240 By using cubic spline interpolation, non-uniformly sampled signals are unified to a fixed sampling rate. Furthermore, the Hanning window function is used to suppress spectral leakage;

[0089] In some embodiments, the leaf roots of the leaf , leaves leaf tips leaf area Calculate PSD region by region:

[0090]

[0091] In the formula, For the blade area In frequency The power spectral density; For the region inner vibration acceleration sample value of the sampling point, is the sampling time; is the Hanning window function; is the analysis frequency;

[0092] In some embodiments, the blade root area : locate the maximum peak in the 80~120HZ frequency band , corresponding to the characteristic frequency ;

[0093] blade tip area : locate the maximum peak in the 30~50HZ frequency band , corresponding to the characteristic frequency ;

[0094] blade middle area : according to the modal data of the same type blade, calibrate the characteristic frequency band, and extract the peak frequency ;

[0095] Step S320: compare the power spectrum density of each area with the preset health area power spectrum density threshold of the blade, and mark the area exceeding the power spectrum density threshold as the high-risk area of the blade; call the historical health database of the same type blade to extract the baseline PSD of the corresponding area: blade root baseline , is the number of healthy samples, and the median is taken to avoid the interference of abnormal values; for the blade tip and blade middle baseline, the same calculation is performed 、 .

[0096] In some embodiments, the area damage index quantifies the degree of vibration energy anomaly, which can be expressed as:

[0097]

[0098] In the formula, is the area damage index; is the characteristic frequency band boundary; is the maximum PSD value extracted in the characteristic frequency band; based on the threshold , if , mark the area high-risk area, and output its spatial coordinate set;

[0099] Step S330: Based on the surface curvature and size characteristics of the blade reference model, a basic flight path covering the windward surface, leeward surface and leading edge of the blade is generated; specifically, based on the three-dimensional reference model of the blade constructed in step S220, in combination with the structural dimensions such as the length and curvature of the blade obtained in step S100, the blade is first divided into a root section, a middle section and a tip section along the length direction. Then, the safety distance between the unmanned aerial vehicle and the blade is dynamically set according to the maximum curvature of the blade surface. It should be noted that in this embodiment, when planning the flight points along the length direction of the blade, the spacing between adjacent flight points needs to be controlled to ensure that the image stitching overlap rate is ≥80%. Finally, by adjusting the pitch angle and yaw angle of the unmanned aerial vehicle, a surrounding path around the blade is generated, and the surface direction information of the reference model is used to ensure that the flight path covers the windward surface, leeward surface and leading edge of the blade.

[0100] Step S340: For the high-risk areas of the blade, the density of flight points is increased based on the basic flight path to form a layered surrounding flight path with encrypted flight points.

[0101] In some embodiments, based on the high-risk areas marked in step S320, the nearest flight point of each high-risk area is first located on the basic flight path, which can be matched through a spatial search algorithm to ensure that the encrypted area is connected to the basic flight path;

[0102] In some embodiments, the density of flight points is increased in the local area around the flight point, and the flight points are arranged more densely near the high-risk areas on the blade surface, so that the image resolution is improved to a higher precision to adapt to the detection needs of fine defects such as micro-cracks;

[0103] In some embodiments, a path optimization algorithm is used, such as the Traveling Salesman Problem solution, to plan the order of flight points, with constraints on the total number of flight points and the turning angle between adjacent flight points;

[0104] In summary, step S300 quantifies the vibration abnormality area of the blade in the whole span direction through the spectral feature The layered basic flight path is generated based on the geometric parameters of the blade reference model, and the flight points are encrypted in the high-risk areas, which improves the micro-crack recognition rate without significantly increasing the inspection time, and optimizes the flight path through the TSP algorithm to ensure the safety and smoothness of the unmanned aerial vehicle flight, providing high-confidence images and spatial correlation data for subsequent defect quantification analysis, and finally realizing the detection logic of dynamic vibration early warning and static visual verification.

[0105] Step S400: In the static inspection mode, the unmanned aerial vehicle is controlled to collect multi-angle images of the blade according to the layered surrounding flight path, and the surface defects are identified and the visual confidence is generated by processing the multi-angle images of the blade.

[0106] Step S400 specifically includes:

[0107] Step S410: In the static inspection mode, the unmanned aerial vehicle is controlled to collect multi-angle images of the windward surface, leeward surface and leading edge of the blade along the layered circumnavigation route, and the spatial position information corresponding to each image is recorded synchronously;

[0108] In some embodiments, the unmanned aerial vehicle is controlled to fly along the layered circumnavigation route, and a high-resolution camera is carried to dynamically adjust exposure parameters to adapt to the change of the reflectivity of the blade surface. The collection position and attitude data of each image are recorded synchronously, including the rotation matrix of the unmanned aerial vehicle and the world coordinates of the camera optical center, to ensure that the image is strictly bound with the spatial position.

[0109] In some embodiments, according to the blade surface reflectivity data of step S100, the camera exposure time, gain and other parameters are dynamically adjusted to avoid overexposure or underexposure, so as to ensure that the image details are clear.

[0110] Step S420: Based on the inherent texture period and reflectivity characteristics in the blade material characteristic data, the regions consistent with the inherent texture of the blade are filtered out, and the abnormal texture regions are retained;

[0111] In some embodiments, the Gabor filter is constructed using the blade inherent texture period and direction of step S100. Then, the background regions matching the texture period are suppressed through convolution operation. The abnormal regions with reflectivity difference exceeding the set threshold are extracted to generate a binary mask to separate potential defects;

[0112] Step S430: The abnormal texture regions are processed by using multi-scale filtering to extract the defect contour and quantify the defect geometric features;

[0113] In some embodiments, a Gaussian pyramid is constructed to realize multi-scale analysis, and a Laplace edge enhancement operator is used to strengthen the weak edges of the defects. A direction-sensitive Canny algorithm is used to extract continuous contours, and the physical size of the defects is quantified: the length is calculated by the contour polygon perimeter, the width is calculated by the projection boundary difference, and the actual damage area is calculated by the pixel area.

[0114] Step S440: According to the edge sharpness of the defect region, the contrast of the background region and the edge sharpness of the defect contour, the visual confidence is calculated combined with the geometric features of the defect;

[0115] In some embodiments, the gradient mean of the defect edge and the gradient mean of the background region are calculated, and the edge difference between the defect and the background is quantified by the ratio of the two. The local light variation coefficient is calculated to compensate for the influence of uneven light on the gradient. Combined with the area influence coefficient, the visual confidence is calculated by the formula:

[0116]

[0117] In the formula, is the visual confidence; is the gradient mean of the defect edge pixel; is the average gradient of the background region pixels; is the area influence coefficient; is the defect projection area; is the illumination uniformity compensation factor; the ratio of the numerator and denominator of the formula reflects the clarity of the defect edge; the exponential term reduces the confidence of large-size false defects through area penalty; the illumination factor compensates for environmental interference, and finally quantifies the reliability of defect recognition.

[0118] In summary, step S400 filters out inherent texture interference according to the material characteristics of S100, realizes accurate image acquisition of the whole surface by combining the hierarchical flight path of S300, extracts defect details through multi-scale analysis and geometric quantization, and quantifies defect authenticity by a confidence model. This method breaks through the interference of inherent texture of the blade on defect recognition, provides high-credibility surface defect characterization for subsequent dynamic and static data fusion, and supports the grading diagnosis and decision of the fan blade defects.

[0119] Step S500: mapping the dynamic vibration response data to the blade reference model using a space-time alignment model, extracting the power spectral density features of the corresponding positions, and fusing the visual confidence to calculate the damage index;

[0120] Step S500 specifically includes:

[0121] Step S510: constructing a blade space rotation matrix according to the blade posture parameters during dynamic vibration response data acquisition; specifically, acquiring the blade rotation angle acquired by the fan SCADA system, which is used to reflect the spatial posture of the blade during vibration acquisition and determine the spatial mapping relationship of the vibration signal; constructing a space rotation matrix for the plane rotation of the blade around the root point, which realizes the conversion of the relative position of the vibration sensor to the coordinates in the local coordinate system of the blade, and can be expressed as:

[0122]

[0123] Step S520: mapping the dynamic vibration response data to the spatial coordinates of the blade reference model through the blade space rotation matrix according to the spatial coordinates of the blade reference model and the installation position of the vibration sensor;

[0124] In some embodiments, according to the conversion of the fan center longitude and latitude to the root point coordinates in the world coordinate system and the installation position of the vibration sensor , the spatial mapping can be expressed as:

[0125]

[0126] In the formula, The formula maps the vibration signal from the sensor coordinate system to the blade local coordinate system corresponding to the dynamic vibration response of the blade spatial coordinate;

[0127] Step S530: Synchronize the dynamic vibration response data and the position information of the blade reference model based on the time stamp, and realize the correlation alignment of the dynamic vibration response data and the spatial position of the blade;

[0128] In some embodiments, by acquiring the time stamp of the vibration data in step S240 and the time stamp of the image in step S410, aligning the non-synchronous time stamps by linear interpolation, ensuring that each vibration data point corresponds to a unique spatial position of the blade reference model, such as the basic waypoint in S330 or the encrypted waypoint in S340, the problem of space-time separation of dynamic and static data is solved.

[0129] Step S540: Calculate the power spectral density integral value of the damage area corresponding to the position of the blade surface defect and the preset healthy area in the blade characteristic frequency interval, respectively, to obtain the vibration energy ratio; The integral of the damage area is represented as: , wherein is the dynamic vibration power spectral density of the defect position after mapping in step S530; The integral of the healthy area is represented as: , wherein is the reference power spectral density of the healthy area in S320; The ratio of the integral of the vibration abnormality damage area and the integral of the healthy area is used to reflect the change of the vibration energy of the damage area, and the greater the ratio, the more serious the internal damage.

[0130] Step S550: Adjust the weight coefficient of visual confidence according to the type of surface defect;

[0131] In some embodiments, for the visual confidence , the defect area , i.e. the actual area quantified in step S430; is calculated based on the unmanned aerial vehicle binocular camera; The corrected visual confidence can be represented as:

[0132]

[0133] In the formula, represents a preset correction coefficient, which is used to adjust the influence sensitivity of the defect area on the visual confidence;

[0134] In some embodiments, the visual confidence weight factor can be preset according to the defect type in S430, such as crack , corrosion , and balance the contribution of visual and vibration characteristics.

[0135] Step S560: The visual confidence and the vibration energy ratio are weighted and fused according to the adjusted weight coefficient to generate a damage index reflecting the surface and internal damage degree. The damage index can be expressed as:

[0136]

[0137] In the formula, DI represents the damage index; Wvis represents the visual confidence weight factor; Wvis represents the visual confidence correction term based on the defect area; A represents the defect area; K represents a preset correction coefficient for adjusting the sensitivity of the defect area to the visual confidence; f represents the blade characteristic frequency; B represents the characteristic frequency bandwidth; P represents the power spectral density function of the damage area; P represents the power spectral density function of the healthy area. This method realizes accurate binding of the vibration signal and the spatial position of the blade through a space-time alignment model; fuses the visual confidence of step S400 and the vibration frequency spectrum feature of step S310 to construct a dual-mode damage index, which quantifies the morphology and credibility of the surface defect on one hand, and also represents the internal structural damage degree on the other hand, thereby providing a full-dimensional and quantitative damage diagnosis basis for the fan blade and improving the accuracy and reliability of defect diagnosis.

[0138] Step S600: Damage levels are divided based on the damage index, and an inspection report is output;

[0139] In some embodiments, the mapping relationship between the damage index and the damage level can be established through data calibration and threshold derivation. Specifically, a historical failure database of the same type of fan blade is collected, including the damage index (DI) calculated in step S500, the actual damage degree, and the maintenance record. The correlation between DI and damage risk is fitted through statistical regression analysis to determine the grading threshold;

[0140] In some embodiments, low risk corresponds to surface micro-defects and internal vibration energy change ≤ 30%, which is controllable; medium risk corresponds to deeper surface defects or internal vibration energy change of 30% to 100%, which needs to be handled within a limited period of time; high risk corresponds to serious surface and internal collaborative damage, which has a risk of fracture and needs emergency intervention; the specific threshold can be dynamically adjusted according to the fan operating environment, such as offshore or land, and the blade aging degree;

[0141] In some embodiments, the triggering condition of periodic re-inspection can be configured as: the DI reflects that the surface defect is a micro crack or slight corrosion, and the internal vibration energy ratio is less than or equal to 1.3, at which time the monthly re-inspection plan is started, and the layered circumferential route of step S300 is reused to focus on monitoring the DI change of the area; the triggering condition of limited maintenance can be configured as: the DI reflects that the surface defect is a penetrating crack or large area corrosion, or the internal vibration energy ratio is in the interval [1.3, 2], then the 72-hour maintenance work order is issued, combined with the defect space coordinates of step S410 and the blade structure size of step S100, a maintenance plan is planned; the triggering condition of emergency shutdown maintenance can be configured as: the DI is close to 1, the surface defect is a penetrating crack or the internal vibration energy ratio is greater than 2, then the fan is immediately triggered to stop, a maintenance plan is generated at the same time, and based on the defect geometric features of step S430, the feasibility of blade replacement or overall maintenance is evaluated, and the safety of the equipment is prioritized.

[0142] In some embodiments, the data integration and visualization includes a surface defect module for integrating the defect type, spatial position, size, visual confidence of step S400; a vibration feature module for extracting the characteristic frequency and power spectral density ratio of step S540; a damage assessment module for displaying the DI value, damage level and risk description of step S560; a processing suggestion module for generating maintenance strategies corresponding to the level; and finally generating a structured PDF report embedded with a three-dimensional view of the blade reference model and labeled with defect positions, vibration spectrum graphs and highlighted feature frequency intervals, to facilitate the intuitive understanding of the damage state by the operation and maintenance personnel.

[0143] In summary, the fan blade inspection method based on the unmanned aerial vehicle provided by the present application collects dynamic vibration response data by pre-setting short-term dynamic vibration excitation rules, captures the vibration characteristics of the blade in the whole phase of starting and steady state shutdown, and provides dynamic structure response basis for identifying internal damage of the blade, while avoiding long-term impact on fan power generation; based on the dynamic vibration data, high-risk areas are identified and layered circumferential routes containing basic routes and encrypted waypoints are generated, so that static inspection can not only cover the whole surface of the blade, but also focus on detecting high-risk areas, improving the pertinence and efficiency of the inspection; at the same time, in the static inspection, the surface defects are identified in combination with the blade material characteristics and the visual confidence is generated, realizing accurate identification and reliability quantification of surface defects; the damage index is calculated by using the spatio-temporal alignment model to fuse dynamic vibration response data and visual confidence, associating surface defect features with internal vibration characteristics, breaking through the limitation of static inspection that can only detect surface defects, and realizing the cooperative evaluation of surface and internal damage of the blade; finally, the damage index is divided into levels and a report is output, providing a comprehensive and accurate state basis for blade operation and maintenance, effectively solving the technical problem that internal damage cannot be identified by static inspection, and improving the integrity and accuracy of fan blade inspection.

[0144] Please refer toFigure 2 Fig. 1 shows a structural schematic diagram of a UAV-based fan blade inspection system according to an embodiment of the present application, which comprises:

[0145] a data acquisition module configured to acquire fan static basic data, including spatial positioning data, structural size data and blade material characteristic data;

[0146] a short-time dynamic vibration excitation module configured to construct a blade reference model based on the fan static basic data, and collect dynamic vibration response data according to a preset short-time dynamic vibration excitation rule;

[0147] a flight path planning module configured to identify a high-risk area of the blade according to the dynamic vibration response data, and generate a hierarchical surrounding flight path containing a basic flight path and high-risk area encrypted waypoints in combination with the blade reference model;

[0148] a visual detection module configured to, in a static inspection mode, control the UAV to collect multi-angle images of the blade according to the hierarchical surrounding flight path, process and identify surface defects of the multi-angle images of the blade and generate a visual confidence;

[0149] a fusion diagnosis module configured to map the dynamic vibration response data to the blade reference model by using a space-time alignment model, extract power spectral density features of corresponding positions, and fuse the visual confidence to calculate a damage index;

[0150] a decision output module configured to divide damage levels based on the damage index and output an inspection report.

[0151] The technical solution of the third aspect of the present application provides an electronic device, which comprises a processor and a memory connected in communication with the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor to enable the processor to execute the steps of the UAV-based fan blade inspection method according to the technical solution of the first aspect of the present application.

[0152] The technical solution of the fourth aspect of the present application provides a computer readable storage medium, which stores a program for implementing a UAV-based fan blade inspection method, and the program for implementing the UAV-based fan blade inspection method is executed by a processor to implement the steps of the UAV-based fan blade inspection method according to the technical solution of the first aspect of the present application.

[0153] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0154] The various embodiments described in this specification are presented by way of example, and each embodiment is not necessarily composed of all features described with respect to other embodiments.

Claims

1. A method for inspecting wind turbine blades based on unmanned aerial vehicles (UAVs), characterized in that, The method includes: Obtain static basic data of the wind turbine, including spatial positioning data, structural dimension data, and blade material property data; A blade baseline model is constructed based on the static basic data of the wind turbine, and dynamic vibration response data is collected according to the preset short-time dynamic vibration excitation rules. Based on dynamic vibration response data, high-risk areas of the blade are identified, and a layered circumnavigation route containing a basic route and encrypted waypoints in high-risk areas is generated by combining the blade baseline model. In static inspection mode, the drone is controlled to collect multi-angle images of the blades according to the layered circling route, and the multi-angle images of the blades are processed to identify surface defects and generate visual confidence scores. Dynamic vibration response data is mapped to a blade baseline model using a spatiotemporal alignment model. Power spectral density features at corresponding locations are extracted, and visual confidence scores are fused to calculate the damage index, including: Within the characteristic frequency range of the blade, the power spectral density integral values ​​of the damaged area and the preset healthy area at the corresponding positions of the blade surface defects are calculated to obtain the vibration energy ratio. The weighting coefficient of visual confidence is adjusted according to the type of surface defect; The ratio of visual confidence to vibration energy is weighted and fused according to the adjusted weighting coefficients to generate a damage index that reflects the degree of surface and internal damage. Damage levels are classified based on damage indices, and inspection reports are generated.

2. The wind turbine blade inspection method based on unmanned aerial vehicles as described in claim 1, characterized in that, A blade baseline model is constructed based on the static foundation data of the wind turbine, and dynamic vibration response data is collected according to a preset short-time dynamic vibration excitation rule, including: Based on structural dimension data, a layered circumferential modeling flight path is planned, and the UAV is controlled to collect multi-angle images of the blades along the layered circumferential modeling flight path. A blade baseline model is generated by fusing multi-angle images using a motion reconstruction structure algorithm. The control fan executes a preset short-term dynamic vibration excitation rule, starting from the shutdown state, reaching the preset speed, and running stably for a preset time before gradually shutting down, forming a vibration excitation process that includes the startup, steady state, and shutdown stages; During the vibration excitation process, vibration signals during the blade start-up and shutdown transient processes are collected, and the SCADA system timestamps are synchronized to form dynamic vibration response data.

3. The wind turbine blade inspection method based on unmanned aerial vehicles as described in claim 1, characterized in that, Based on dynamic vibration response data, high-risk areas of the blade are identified. Combined with the blade baseline model, a layered circumnavigation route is generated, including a basic route and additional waypoints for high-risk areas. Spectral analysis was performed on the dynamic vibration response data to extract the characteristic frequencies of different regions of the blade and the power spectral density at the corresponding frequencies. The power spectral density of each region is compared with the preset power spectral density threshold of the healthy region of the leaf, and the regions that exceed the power spectral density threshold are marked as high-risk regions of the leaf. Based on the surface curvature and dimensional characteristics of the blade reference model, a basic flight path covering the windward side, leeward side and leading edge of the blade is generated. For high-risk areas of the blades, the density of waypoints is increased on the basis of the basic route to form a layered circular route with denser waypoints.

4. The wind turbine blade inspection method based on unmanned aerial vehicles as described in claim 1, characterized in that, In static inspection mode, the drone is controlled to collect multi-angle images of the blades according to a layered circling route. These multi-angle images are then processed to identify surface defects and generate visual confidence scores, including: In static inspection mode, the drone is controlled to collect multi-angle images of the windward side, leeward side and leading edge of the blade along the layered circling route, and the spatial position information corresponding to each image is recorded simultaneously. Based on the inherent texture period and reflectivity characteristics in the blade material property data, areas consistent with the inherent texture of the blade are filtered out, while abnormal texture areas are retained. Multi-scale filtering is used to process the abnormal texture region, extract the defect contour, and quantify the defect geometric features; Visual confidence is calculated based on the edge sharpness of the defect area, the contrast of the background area, and the edge sharpness of the defect outline, combined with the geometric features of the defect.

5. The wind turbine blade inspection method based on unmanned aerial vehicles (UAVs) as described in any one of claims 1 to 4, characterized in that, The dynamic vibration response data is mapped to the blade reference model using a spatiotemporal alignment model, including: A blade spatial rotation matrix is ​​constructed based on the blade attitude parameters acquired during dynamic vibration response data collection. Based on the spatial coordinates of the blade reference model and the installation position of the vibration sensor, the dynamic vibration response data is mapped to the spatial coordinates of the blade reference model through the blade spatial rotation matrix. By synchronizing the dynamic vibration response data with the position information of the blade reference model based on the timestamp, the association and alignment between the dynamic vibration response data and the spatial position of the blade are achieved.

6. The wind turbine blade inspection method based on unmanned aerial vehicles as described in claim 5, characterized in that, The damage index can be expressed as: In the formula, Indicates the damage index; Indicates the visual confidence weighting factor; This represents the visual confidence correction term based on the defect area; Indicates the area of ​​the defect; This represents the preset correction coefficient, used to adjust the sensitivity of the impact of defect area on visual confidence. Indicates the characteristic frequency of the blade; Indicates the characteristic frequency bandwidth; Represents the power spectral density function of the damaged region; This represents the power spectral density function of the healthy region.

7. A wind turbine blade inspection system based on unmanned aerial vehicles (UAVs), characterized in that, The wind turbine blade inspection method based on unmanned aerial vehicles (UAVs) according to any one of claims 1 to 6, wherein the system comprises: The data acquisition module is configured to acquire static basic data of the wind turbine, including spatial positioning data, structural dimension data, and blade material property data. The short-time dynamic vibration excitation module is configured to build a blade baseline model based on the static basic data of the wind turbine, and collect dynamic vibration response data according to the preset short-time dynamic vibration excitation rules; The route planning module is configured to identify high-risk areas of the blade based on dynamic vibration response data, and generate a layered loop route including a basic route and encrypted waypoints in high-risk areas in combination with the blade reference model. The visual inspection module is configured to control the UAV to collect multi-angle images of the blades according to the layered circling route in static inspection mode, process the multi-angle images of the blades to identify surface defects and generate visual confidence scores. The fusion diagnostic module is configured to use a spatiotemporal alignment model to map dynamic vibration response data to the blade reference model, extract the power spectral density features at the corresponding locations, and fuse visual confidence to calculate the damage index. The decision output module is configured to classify damage levels based on damage indices and output inspection reports.

8. An electronic device, characterized in that, The electronic device includes: a processor and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the steps of the unmanned aerial vehicle-based wind turbine blade inspection method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program for implementing a UAV-based wind turbine blade inspection method, which is executed by a processor to implement the steps of the UAV-based wind turbine blade inspection method according to any one of claims 1 to 6.

Images