Wind power blade multi-mode nondestructive testing system and method based on unmanned aerial vehicle

Abstract

The invention relates to the technical field of blade nondestructive testing, in particular to a wind power blade multi-mode nondestructive testing system and method based on an unmanned aerial vehicle, and the system comprises an identification number screening module which obtains unit and blade numbers, lists the number of occurrence of the blade numbers, sorts frequent numbers and sets flight task options; the image acquisition module sets waypoints, acquires images and marks angles and positions, the channel image processing module aligns and fuses image layers, the defect registration module observes cracks and spalling and writes current situation descriptions one by one, and the result summarization module outputs leaf areas and damage types, completes a table and obtains a defect summary table. According to the method, the focus risk area is inspected by arranging the numbering frequency. Recording angles, channel sources and positions and images have traceability. And brightness calibration, edge comparison, layer processing and multi-view content coherence are realized. By writing crack paths, stripping boundaries and damage feature presentation, connection of inspection and damage arrangement is ensured.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for wind turbine blades, and more particularly to a multi-modal non-destructive testing system and method for wind turbine blades based on unmanned aerial vehicles (UAVs). Background Technology

[0002] The field of non-destructive testing technology for wind turbine blades involves the non-destructive testing and monitoring of wind turbine blades, aiming to evaluate the structure and performance of blades through various testing methods. Core technologies include defect identification, vibration analysis, and stress monitoring. Commonly used methods include ultrasonic testing, infrared imaging, and vibration analysis, which are widely applied in the wind power industry, improving safety and reducing maintenance costs.

[0003] Traditional UAV-based multimodal nondestructive testing systems and methods for wind turbine blades involve using UAVs equipped with sensors such as infrared imaging, high-definition photography, and LiDAR to perform multi-angle, multi-mode inspections of wind turbine blades. UAV systems are flexible in flight, capable of covering all parts of the blade, collecting data, and performing real-time analysis. Compared to traditional manual inspection methods, UAV-based methods avoid the risks of high-altitude operations, improve inspection efficiency and accuracy, and enable a comprehensive assessment of the blade's health status through various sensor technologies.

[0004] Current technologies lack historical data-based numbering and screening methods in mission planning, resulting in homogeneous allocation of inspection targets and hindering timely attention to high-risk blades. Image acquisition and recording methods are relatively simplistic, with inconsistent expression of angle and channel attributes, making judgment errors prone to occur during retrospective analysis. The lack of brightness relationship calibration and edge trend matching between multi-angle footage leads to insufficient image coherence, causing a sense of discontinuity in crack direction and spalling contours during comparison. Damage documentation focuses on single-point descriptions, lacking path-based representation, making it difficult to form a unified understanding of the extension trend. The lack of clear correspondence between regions and types in archiving methods weakens structural differences in a large volume of records, affecting subsequent analysis and maintenance decisions. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and propose a multimodal non-destructive testing system and method for wind turbine blades based on unmanned aerial vehicles (UAVs).

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a multimodal nondestructive testing system for wind turbine blades based on unmanned aerial vehicles (UAVs), the system comprising, The identification number filtering module obtains the wind farm unit number and blade number, lists the number of times the blade number appears in the maintenance record, sorts out whether each number has appeared multiple times in past tasks, fills the frequent numbers into the flight operation table, and obtains the inspection number list. The image acquisition and registration module calls the numbers listed in the inspection number list to set the flight waypoint position, activates the acquisition device during flight to acquire numbered images, writes the angle, device channel name and flight position when the image is acquired, distinguishes all flight images according to the number, and obtains a summary table of part image records. Channel image processing module: retrieves images with the same number from the image record table of the aforementioned parts, aligns the layer boundary lines, observes the light and dark relationship and edge contact direction of the seam area between layers, supplements the shooting source name and action status information, and obtains the fused image registration form; The defect content registration module observes the crack extension path, material peeling area and conductive layer fracture segment based on the fused image in the fused image registration table, writes the length direction, boundary shape and structure name of each location in the image, fills in the wind turbine blade damage content one by one in the order of image number, and obtains the wind turbine blade damage description document. The results archiving and aggregation module, based on the content under the numbered sections in the wind turbine blade damage description document, lists the blade tip, web, and root structural entries alongside the damage descriptions. It outputs the blade region and blade damage type in two columns, merges the text content under each number into a page fragment, completes the table writing, and closes each numbered region. This generates a summary table of blade structural defects.

[0007] As a further aspect of the present invention, the acquisition of wind farm turbine and blade numbers specifically includes turbine number, blade number, blade number maintenance record frequency, flight mission number setting, number mission association, and inspection number list; the image acquisition and registration specifically includes flight waypoint location, image acquisition angle, equipment channel name, flight position label, number-distinguished image content, and part image recording; the channel image processing specifically includes layer boundary alignment, joint area brightness and darkness relationship, joint edge direction, layer fusion status, shooting source, and integration action status information; the defect content registration specifically includes crack path, material peeling area, conductive layer fracture fragment, suspected defect location, boundary morphology, structural location description, and blade current status description; the result archiving and collection specifically includes a blade structural defect summary table, numbered area text merging, blade tip area, web area, blade root area, damage type differentiation, and area complete closure.

[0008] As a further aspect of the present invention, the identification number filtering module includes: The numbering statistics submodule obtains the numbering information of wind farm units and blades, counts the number of times each blade number appears in the maintenance record, analyzes the frequency of number occurrence, marks frequently occurring numbers based on frequency, and obtains the number occurrence frequency. The maintenance record filtering submodule filters out the blade numbers that frequently appear in the maintenance records based on the frequency of their occurrence. It then calls up the marked blade number information to filter out the corresponding maintenance tasks from the maintenance records and obtains a list of maintenance record numbers. The flight mission setting submodule, based on the maintenance record number list, configures flight mission options for frequently occurring blade numbers, ensuring that these blade numbers are applied as priority objects in the flight operation table, and derives an inspection number list.

[0009] As a further aspect of the present invention, the image acquisition and registration module includes: The waypoint setting submodule sets the flight waypoint positions according to the numbers in the inspection number list, obtains the flight path for each number and determines the waypoints, and generates a flight waypoint list. The image acquisition submodule activates the acquisition device according to the flight waypoint list, acquires the image corresponding to each number, records the angle, device channel name and flight position when each image is acquired, and obtains the image acquisition record. The image annotation submodule annotates each image based on the image acquisition record, recording its angle, device channel, and flight position, distinguishing images according to their numbers, and generating a summary table of part image records.

[0010] As a further aspect of the present invention, the channel image processing module includes: The image alignment submodule retrieves multiple sets of images with the same number based on the image record table of the aforementioned area, compares the layer boundary lines, adjusts the position of each layer to align them, and obtains a list of aligned images. The seam inspection submodule, based on the aligned image list, observes the light and dark relationship and edge contact direction of the layer seam area, determines whether there is a discontinuous or unnatural transition at the seam, and obtains the seam inspection result. The layer fusion submodule merges the layers based on the seam inspection results, supplements the shooting source name and integration action status information, and generates a fused image registration form.

[0011] As a further aspect of the present invention, the defect content registration module includes: The crack observation submodule observes the extension path of the cracks based on the images in the fused image registration table, records the length, direction and boundary morphology of each crack, and obtains the crack observation results. The material spalling identification submodule identifies material spalling areas in the image based on the crack observation results, records the shape and location of the spalling areas, and obtains the material spalling identification results. The damage description filling submodule fills in the current status description of the blade line by line based on the material spalling identification results. It records the damage situation of each location in the order of number, including structural damage such as cracks and spalling, and produces a wind turbine blade damage description document.

[0012] As a further aspect of the present invention, the result archiving and aggregation module includes: The structural term extraction submodule extracts structural terms related to the blade tip, web, and root from the wind turbine blade damage description document, and lists them alongside the damage description to obtain a list of structural terms. The region and damage matching submodule, based on the structural term list, performs a two-column match between the blade region and the damage type, organizes the correspondence between each region and the damage, and obtains the region damage matching result. The document merging submodule, based on the regional damage matching results, merges the text information under each number into page fragments, completes the table writing, closes each numbered region, and obtains a summary table of blade structural defects.

[0013] The method for multimodal nondestructive testing of wind turbine blades based on UAVs, wherein the method is executed based on the aforementioned multimodal nondestructive testing system for wind turbine blades based on UAVs, includes the following steps: S1: Obtain the wind farm unit number and blade number, count the number of times the blade number appears in the maintenance record, filter out the frequently appearing blade numbers, and sort out whether each number has appeared multiple times in past tasks. For the frequently appearing numbers, set flight task options and apply the number correspondence to the flight operation table to generate an inspection number list. S2: Based on the number list in the inspection number list, set the flight waypoint position, activate the acquisition device to acquire images, acquire images related to each number during flight, record the angle, device channel name and flight position when each image is acquired, and distinguish all flight image content by number to generate a part image record summary table. S3: Based on multiple sets of images with the same number in the image record table of the aforementioned parts, perform layer boundary line comparison and alignment processing, observe the light and dark relationship and edge contact direction of the layer seam area, complete layer fusion, and supplement the shooting source name and integration action status information to generate a fused image registration form. S4: Based on the images in the fused image registration form, observe the crack extension path, material peeling area, and conductive layer fracture fragments. Write down the length direction, boundary shape, and structural location name of each suspicious location item by item, and fill in the blade status description item by item according to the number order to generate a wind turbine blade damage description document. S5: Based on the content of the wind turbine blade damage description document, list the structural terms such as blade tip, web, and blade root mentioned in the description alongside the damage description, output the blade area and damage type in two columns, merge the text under each number into a page fragment, complete the table writing, close each numbered area, and generate a summary table of blade structural defects.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, task focus is established by analyzing the frequency of numbering, ensuring that inspection planning is based on quantifiable risk indicators. The flight acquisition phase constructs an image coordinate system through the simultaneous registration of angle, channel source, and flight position, giving the image content clear directionality. Multiple image sets are compared using edge tendency comparison, brightness relationship calibration, and layer comparison to form a continuous structure, ensuring stable morphological references for parts with the same number. The damage recording stage uses the itemized writing of crack paths, spalling boundaries, and fracture fragments to form a linear and regional description, focusing damage expression on morphological change patterns. The archiving stage presents the structure name and damage content side-by-side, providing regional and type indications for the content under each number, achieving continuity in information retrieval, damage comparison, and maintenance planning. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart illustrating the acquisition process of the identification number filtering module of the present invention; Figure 3 This is a flowchart illustrating the acquisition process of the image acquisition and registration module of the present invention. Figure 4 This is a flowchart of the acquisition process of the channel image processing module of the present invention; Figure 5 This is a flowchart illustrating the acquisition process of the defect content registration module of this invention. Figure 6 This is a flowchart illustrating the process of obtaining results from the archiving and aggregation module of this invention. Detailed Implementation

[0016] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0017] In this embodiment of the invention, sometimes the subscript such as W1 is written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1This invention provides a technical solution: a multi-modal non-destructive testing system for wind turbine blades based on unmanned aerial vehicles (UAVs), the system comprising: The identification number filtering module obtains the wind farm unit number and blade number, lists the number of times the blade number appears in the maintenance record, sorts out whether each number has appeared multiple times in past tasks, sets flight mission options for frequently appearing numbers, applies the number correspondence to the flight operation table, and obtains the inspection number list. The image acquisition and registration module calls the numbers listed in the inspection number list to set the flight waypoint location, activates the acquisition device during flight to acquire images related to each number, and marks and writes the angle, device channel name and flight position corresponding to each image acquisition. All flight image content is distinguished according to the number, and a summary table of part image records is obtained. The channel image processing module retrieves multiple sets of images with the same number from the overall image record table, aligns the layer boundary lines, observes the brightness and darkness relationship and edge contact direction of the seam area between layers, and after completing the layer fusion action, it supplements the shooting source name and integration action status information, and obtains the fused image registration form. The defect content registration module observes the crack extension path, material peeling area and conductive layer fracture segment in the fused image registration form based on the fused image registration form. It writes down the length direction, boundary shape and structural location name of each suspicious location, fills in the blade status description one by one according to the number order, and obtains the wind turbine blade damage description document. The results archiving and aggregation module, based on the content under each number in the wind turbine blade damage description document, lists the structural terms mentioned in the description such as blade tip, web, and blade root alongside the damage description, outputs the blade region and damage type in two columns, merges the text under each number into a complete page fragment, completes the table writing and closes each numbered area, and obtains a summary table of blade structural defects.

[0020] The acquisition of wind farm turbine and blade numbers specifically includes turbine number, blade number, blade number maintenance record frequency, flight mission number setting, number-mission association, and inspection number list; image acquisition and registration specifically includes flight waypoint location, image acquisition angle, equipment channel name, flight position label, number-distinguished image content, and location image record; channel image processing specifically includes layer boundary alignment, joint area light and dark relationship, joint edge direction, layer fusion status, shooting source, and integration action status information; defect content registration specifically includes crack path, material peeling area, conductive layer fracture fragment, suspected defect location, boundary morphology, structural location description, and blade current status description; result archiving and compilation specifically includes a blade structural defect summary table, numbered area text merging, blade tip area, web area, blade root area, damage type differentiation, and area complete closure.

[0021] Please see Figure 2 The identification number filtering module includes: The numbering statistics submodule obtains the numbering information of wind farm units and blades, counts the number of times each blade number appears in the maintenance record, analyzes the frequency of number occurrence, marks frequently occurring numbers based on frequency, and obtains the number occurrence frequency. The numbering and statistics submodule sequentially reads all wind farm turbine numbers and blade numbers from the original records. During this reading process, it performs character stripping on the number field of each turbine, writing the turbine number into a turbine number list. Then, it extracts the numbers of the three blades corresponding to that turbine, writing the blade numbers into a blade number list. After reading all turbine numbers, it reads the maintenance record table, expanding each record and stripping the blade number field of each record into an independent character field for subsequent comparison. It then iterates through the blade number list, performing a character-by-character match check between each blade number and all maintenance records. This check first compares the character lengths of the two numbers, then performs a character-by-character match check if the lengths are the same. If they match, the number is then... The statistical count is incremented by one. If there is a discrepancy, the comparison continues to the next record. This repeated process accumulates the number of times each number is retrieved in the maintenance records. The obtained count values ​​are then divided into intervals. Maintenance intervals are manually set. For example, with 120 maintenance records, numbers appearing 0 to 3 times are classified as low-frequency, numbers appearing 4 to 7 times as medium-frequency, and numbers appearing 8 times or more as high-frequency. Numbers falling into the high-frequency interval are marked, and the marked field is written to the number attribute table so that it can be read in subsequent modules. For example, blade B5 appears 2 times after retrieval and is classified as low-frequency, blade B9 appears 6 times and is classified as medium-frequency, and blade B12 appears 10 times and is classified as high-frequency and marked, thus obtaining the frequency of number appearance.

[0022] The maintenance record filtering submodule filters out the blade numbers that frequently appear in the maintenance records based on the frequency of their occurrence. It then retrieves the marked blade number information, filters out the corresponding maintenance tasks from the maintenance records, and generates a list of maintenance record numbers. The maintenance record filtering submodule, based on the frequency of the numbers obtained in the previous section, performs judgment actions on these numbers sequentially, extracting numbers that meet the high-frequency interval requirement into a high-frequency number list. Then, it performs a record-by-record read operation on the maintenance record table, separating the blade number field of each record into an independent character item as a comparison parameter for this section's filtering. It performs a character consistency check on this character item against each number in the high-frequency number list. In this check, it first checks if the character lengths of the two character items are the same, then checks if each character is identical. If they are identical, the task number field of that maintenance record is extracted and written to the maintenance record number result list; if they are inconsistent... The process continues with the next record. To ensure clear boundaries for the filtering action, a threshold is set for the frequency of occurrence in the previous segment to enter the high-frequency zone. For example, based on 120 maintenance records, an occurrence count of 8 is set as the minimum value for entering the high-frequency zone. When number B12 appears 10 times in the previous segment, it meets the threshold condition and is written into the high-frequency number list. When the filtering action is performed, if the blade number field of a maintenance record R25 is B12, it is considered consistent, and the task number field of that record is written into the result list. By repeatedly processing all maintenance records, a set of maintenance task numbers is formed, resulting in a maintenance record number list.

[0023] The flight mission setting submodule, based on the maintenance record number list, configures flight mission options for frequently occurring blade numbers, ensuring that these blade numbers are applied as priority objects in the flight operation table, and derives an inspection number list.

[0024] The flight mission setting submodule, based on the maintenance record number list, reads each mission number in the list one by one and extracts its corresponding blade number field. Then, it matches this field with the high-frequency number list in the number statistics module. When a match is found, the number is written to the flight mission configuration parameters. In the parameter writing action, the spatial coordinate data of the aircraft to which the number belongs is read and written to the spatial parameter area of ​​the mission configuration table. Finally, based on the priority range set by the frequency of the previous number occurrence, a priority assignment action is performed on the number, assigning priority 1 to low-frequency numbers and priority 1 to mid-frequency numbers. Assigning a priority of 2, and assigning a priority of 3 to the high-frequency zone number, for example, number B12 is classified as a high-frequency zone 10 times in the statistics, and its priority is set to 3. When generating task sorting, it is placed in the previous position of the current number group. When multiple numbers are associated with the same unit, the distance judgment is performed on the unit coordinates. By comparing the coordinate differences, the relatively closer unit is identified and its corresponding task number is placed in the previous position. Then, all sorted task numbers are written into the inspection number list one by one. Through such parameter writing actions, priority setting actions and order sorting actions, all task numbers are sorted and the inspection number list is obtained.

[0025] Please see Figure 3The image acquisition and registration module includes: The waypoint setting submodule sets the flight waypoint locations based on the numbers in the inspection number list, obtains the flight path for each number and determines the waypoints, and generates a flight waypoint list. The waypoint setting submodule performs a reading action based on the numbers in the inspection number list. It extracts each number sequentially, stripping characters from the number field during extraction to split it into a crew number and a blade number. For the crew number, it performs a crew coordinate retrieval action, writing the coordinate values ​​recorded in the crew coordinate table into a temporary parameter area. For the blade number, it performs a blade position parameter extraction action, writing the corresponding three-dimensional position increment in the crew coordinate system into the temporary parameter area. It then performs a position synthesis action on the crew coordinates and blade increment parameters, adding the axial parameters to obtain the spatial point position for that number. Finally, it performs a sequence judgment action on all spatial points corresponding to each number, using the order of the numbers in the inspection list as the initial sorting basis. Based on this, it compares the distance relationships between spatial points, prioritizing points that are closer and prioritizing points that are farther apart. The module also allows for manual setting of distance ranges, defining the 0-80 meter range as the close range and the 80-160 meter range as the close range. The system is divided into a medium-distance zone and a long-distance zone, with distances greater than 160 meters defined. The spatial coordinate difference is used to determine the range of each point. For example, point B12 corresponds to (150, 90, 80), and point B13 corresponds to (220, 110, 85). The difference between these two points falls into the long-distance zone. Therefore, in the sorting process, B12 is written to a earlier position, and B13 to a later position. A waypoint number generation action is then performed on the sorted numbers, writing H1 to the first waypoint, H2 to the second waypoint, and so on, until all waypoint numbers are assigned. During waypoint generation, spatial coordinates, associated inspection number, and flight segment number are written to each waypoint and recorded in the waypoint record area. In the checking action, each waypoint record is checked for the presence of the number field, coordinate field, and waypoint sequence field. If any fields are missing, the corresponding inspection number is re-executed to fill in the missing information. After all waypoint records have been checked, all waypoints are sequentially written into the overall waypoint list to obtain the flight waypoint list.

[0026] The image acquisition submodule activates the acquisition device based on the flight waypoint list, acquires the image corresponding to each number, records the angle, device channel name and flight position when each image is acquired, and obtains the image acquisition record. The image acquisition submodule reads the flight waypoint list during flight, sequentially calling each waypoint. During this process, it extracts the waypoint number, spatial coordinates, and inspection number fields, breaking them down into parameter sets that can be directly written into flight control. Before entering a waypoint, it compares the aircraft's current position with the target waypoint position, determining whether the aircraft has reached the acquisition zone by analyzing the differences between the three coordinate axes. To avoid ambiguity, distance thresholds can be set for the acquisition zone: 0-5 meters is defined as the acquisition zone, 5-15 meters as the zone requiring position fine-tuning, and greater than 15 meters as the non-acquisition zone. The difference in distance determines the zone; for example, if the difference between the aircraft and the target point at waypoint H4 is 3 meters, the aircraft has entered the acquisition zone and can immediately begin acquisition. The module then calls the shooting configuration for the waypoint's inspection number, specifying the shooting angle, etc. The camera channel number and the relative shooting position required by that number are written into the temporary acquisition parameters. During the acquisition process, the imaging equipment is started, and the current equipment channel name is written into the image buffer. Then, the aircraft's nose direction, roll angle, and pitch angle are read and written into the temporary image record. At the moment of image generation, the spatial position parameters are written, and the waypoint coordinates are directly written into the image record line. During the image acquisition process, for example, when shooting at waypoint H7 with number B12, the pitch angle of 32 degrees, roll angle of 4 degrees, nose direction of 85 degrees, equipment channel CH2, and spatial position (210,120,95) are recorded and written into the image acquisition record area. After the flight path is completed, all image records are checked for completeness according to waypoint number. Records with missing angle, channel, or position fields are filled in sequentially. After all records are complete, the image records are written into the image acquisition record in waypoint order to obtain the image acquisition record.

[0027] The image annotation submodule annotates each image based on the image acquisition record, recording its angle, equipment channel, and flight position, distinguishing images by number, and generating a summary table of part image records.

[0028] The image annotation submodule performs a reading action based on the image acquisition records, extracting each image record from the record table one by one. During extraction, it performs a separation action on the angle field of the record, splitting it into three independent fields: nose angle, pitch angle, and roll angle. Then, it performs a character stripping action on the equipment channel field, writing the channel name to a temporary annotation area. For the flight position field, it performs a coordinate separation action, writing the three coordinate values ​​to the position record area. It performs a number matching action on the inspection number associated with the record, comparing its character content digit by digit with the inspection number list to determine its corresponding number. In the judgment action, it performs consistency verification on the character length and performs consistency judgment on each character. When a match is found, the number is written to the annotation cache. Finally, in the annotation action, the number field is written to the image record. The angle, channel, and position fields are written sequentially into the image annotation table. To avoid number matching errors, strict character consistency requirements are set. If any character is inconsistent, the number association is rejected. For example, if the image IMG32 records number B15, angle (head angle 12 degrees, pitch angle 28 degrees, roll angle 6 degrees), equipment channel CH1, and spatial position (185, 102, 91) in the acquisition record, these fields are directly written into the image annotation record row during the annotation action. The same writing action is performed on all image files. After all image annotations are completed, the image records are classified by number. The image records corresponding to each number are written into the number classification area and summarized in number order to obtain the overall table of part image records.

[0029] Please see Figure 4 The channel image processing module includes: The image alignment submodule retrieves multiple sets of images with the same number based on the master table of image records for the body parts, compares the layer boundary lines, adjusts the position of each layer to align them, and generates a list of aligned images. The image alignment submodule performs image record reading based on the master table of part-specific image records. It retrieves each image record from the master table one by one, splitting the character field of the record's ID during the retrieval process. The ID is written to a temporary ID set, and then a duplicate ID check is performed on this set. Character-by-character comparison is used to determine if two images share the same ID. If the comparison results are completely identical, the two images are written into the same ID image group. After forming the ID image group, a layer transformation is performed on each image, converting the original image into a layer object with movable, comparable, and identifiable boundaries. Then, a boundary extraction operation is performed on the boundary lines of the layer objects, independently separating the boundary line region from the entire image and writing it to the boundary buffer. Next, a layer boundary comparison operation is performed on each layer in the same ID image group. During the comparison process, the boundary point coordinates of each boundary line are read sequentially and written to the comparison buffer. Finally, a point-by-point difference judgment is performed on the coordinate differences between each boundary point. Points with differences falling within the range of 0 to 3 pixels are determined as points with no offset, and points with differences falling within the range of 3 to 1 are determined as points with no offset. Points within a 0-pixel range are identified as minor offset points, while points with a difference exceeding 10 pixels are identified as significant offset points. When a layer has a minor offset point at its boundary, a lightweight position adjustment is performed on the layer, moving it unit by unit along the horizontal or vertical axis. For example, if a layer is offset 6 pixels to the right relative to the reference layer on the horizontal axis, it is moved 6 times to the left to ensure boundary overlap. When the offset falls into the significant offset point range, segmented compensation movement is performed on the layer, dividing the boundary line into several segments and performing different numbers of movement movements based on the offset of each segment. After all offset compensation is completed, an alignment check is performed on all layers. The aligned boundary lines are read again and compared for a second round to confirm whether each layer has reached the zero-offset point range. If there are boundary segments still in the minor or significant offset range, the movement movement is repeated until all boundary segments enter the zero-offset range. The current displacement parameters of each layer are written to the displacement record area, and the aligned layers are written to the output buffer in numerical order. The results are then compiled to generate an aligned image result set, resulting in a list of aligned images.

[0030] The seam inspection submodule, based on the aligned image list, observes the light and dark relationship and edge contact direction of the layer seam area to determine whether there is a discontinuous or unnatural transition at the seam and obtains the seam inspection result. The seam inspection submodule performs image retrieval based on the aligned image list. Multiple aligned images with the same ID are retrieved in layer order. During retrieval, local stripping is performed on the edge regions of each image, dividing the edge regions into brightness segments, color segments, and edge direction segments, and writing them to the inspection cache. Next, a brightness difference judgment is performed on the brightness segments, comparing the brightness values ​​of corresponding positions in adjacent layers point by point. Pixels with a difference of 0 to 5 are classified as normal brightness areas, those with a difference of 5 to 15 as slightly brightness deviation areas, and those with a difference greater than 15 as significantly brightness deviation areas. If multiple consecutive pixel segments fall into the slightly deviation area, these segments are recorded as deviations; similarly, if consecutive pixel segments fall into the significantly deviation area, they are recorded as severely deviations. Finally, a color continuity judgment is performed on the color segments, comparing the order of each color level value point by point. When the color level of a certain area... When a value suddenly changes, the point is recorded in the color level anomaly record. For example, if layer A shows an increasing trend of five consecutive pixels in a certain seam segment, while layer B suddenly drops the value of the sixth pixel at the corresponding position by 20 color levels, then this point is recorded as an anomaly. A direction consistency judgment action is performed on the edge direction segment. By comparing the tilt direction of the edge direction field, it is determined whether the layer edges extend in the same direction. When the direction field shows a change in orientation, the segment is recorded in the direction anomaly record. After all judgment actions are completed, the number of anomaly records is counted. Records with anomaly counts in the range of 0 to 2 are classified as minor anomaly zones, records with anomaly counts in the range of 3 to 6 are classified as moderate anomaly zones, and records with anomaly counts greater than 6 are classified as severe anomaly zones. Then, the anomaly interval level of each layer seam segment is written into the inspection summary record. After all numbered images are inspected, the inspection summary results are written into the seam inspection result table to obtain the seam inspection result.

[0031] The layer fusion submodule merges layers based on the seam inspection results, adds the shooting source name and integration action status information, and generates a fused image registration form.

[0032] The layer blending submodule performs a record reading action based on the seam inspection results. It loads the corresponding layers and the seam inspection conclusions for each group into the blending buffer. During the loading process, it performs a separation action on the layer fields, peeling off the layer pixel structure, shooting source field, layer sequence number field, and action status field one by one and writing them into an independent parameter area. Then, it performs a blending method judgment action based on the seam inspection conclusions. When the conclusion is in the slightly abnormal area, the layers are directly stacked sequentially. A baseline image is written to the first layer, and an overwrite image is written to subsequent layers, so that the layers are stacked sequentially to generate a blended layer. When the conclusion is in the moderately abnormal area, an edge compensation action is performed on the layers, retrieving the sections with slightly abnormal segments and blending the adjacent layers of that segment. After averaging the corresponding pixels, the compensation paragraph is written, and then the layer overlay writing action is performed. When in a severe abnormal area, multiple rounds of compensation and adjustment actions are performed on the layer. First, compensation writing is performed on each abnormal point, and the number of compensations is written to the action status field. For example, if a layer has 7 abnormal points, the number of compensations is recorded as 7. Then, the overlay writing action is performed on the compensated layers in sequence. The shooting source field is sorted, and the shooting source names of all layers with that number are written to the source registration area in the order of appearance. The adjustment action status field is summarized, and the layer number, the number of compensations, and the action process of the fusion process are written to the status field. After the fusion is completed, the layers are written to the fusion cache, and then the records are written to the fusion image registration table in the order of number to obtain the fusion image registration table.

[0033] Please see Figure 5 The defect content registration module includes: The crack observation submodule observes the extension path of cracks based on the images in the fused image registration form, records the length, direction and boundary morphology of each crack, and obtains the crack observation results. The crack observation submodule performs image retrieval based on the fused image registration form. It reads each fused image in the registration form one by one. During the reading process, it extracts characters from the number field of the image record and writes the number to the number buffer. Then, it strips the spatial location field, shooting source field, and pixel area after layer fusion from the image record, breaking them down into a parameter set that can be directly used for crack area judgment. Next, it performs contour extraction on suspicious linear areas in the image, stripping out areas with color gradation changes, continuous changes in brightness differences, and abrupt boundary changes, and writing them to a temporary contour buffer. After reading the linear areas in the temporary buffer, it performs extension direction judgment. By sequentially reading the coordinates of the points in the area, it writes the connection order of each point to the direction recording area. Then, it performs direction interval definition on the direction recording area, setting the horizontal tilt angle from 0 to 15 degrees as the near-horizontal zone, the tilt angle from 15 to 45 degrees as the medium tilt zone, and the tilt angle exceeding 45 degrees as... For steeply inclined areas, the direction and type of cracks are determined. When determining the crack length, the distance difference between the two ends of the crack is statistically analyzed. When the crack is recorded in segments, the position connection of each segment is judged and written into the crack continuous recording area. For example, if a thin line extending from the left to the upper right appears at position B12 in an image, with an inclination angle of about 28 degrees, it can be written into the medium inclined area. The span of the point is statistically analyzed as 82 consecutive points, so the crack length is recorded as the medium length range. Then, the boundary morphology on both sides of the crack is stripped, and the contours of the protruding, concave, and intersecting boundaries are compared segment by segment. The change in the boundary of each segment is judged by the range. The change in the range of 1 to 3 pixels is set as a slight change, the change in the range of 3 to 8 pixels is set as a moderate change, and the change in the range of more than 8 pixels is set as a severe change. After all the recording operations are completed, the crack direction, crack length, and boundary change are written into the crack observation record in numerical order to obtain the crack observation results.

[0034] The material spalling identification submodule identifies material spalling areas in images based on crack observation results, records the shape and location of the spalling areas, and obtains the material spalling identification results. The material spalling identification submodule performs a number reading action based on crack observation results, retrieving each crack observation record from the record table one by one. During retrieval, coordinate extraction is performed on the crack location field, and the image coordinates of the crack area are written into the identification starting area. Then, the corresponding numbered images in the fused image registration table are retrieved, and images with the same number are written into the image analysis area. Brightness change extraction is performed on the pixel areas in the analysis area, and large-scale brightness change points are extracted and written into the suspected material spalling buffer area. Then, morphology judgment is performed on the suspected spalling area. By reading the continuous distribution of pixels in the area, the contour direction is determined. The morphology is divided into areas with gentle edges, areas with moderate edge undulation, and areas with severe edge undulation according to the amount of change in contour curvature. Areas with curvature changes falling within the range of 0 to 5 pixels are designated as gentle edges, and areas with changes falling within the range of 5 pixels are designated as gentle edges. The area with a change of 5 to 15 pixels is designated as a moderate fluctuation area, and the area with a change of more than 15 pixels is designated as a severe fluctuation area. Based on this, the range confirmation action is performed on the suspected peeling area, and the continuity of the surrounding pixels is judged. When a certain area has more than 12 pixels with obvious darkening or brightening in 20 adjacent pixels, it is identified as a peeling candidate area. For example, in image B15, there is a dark area with a distribution of about 40×30 pixels in the middle of the leaf. 17 pixels in its continuous pixel inspection show significant changes. This area is written into the peeling candidate area. Then, the morphology supplementation action is performed on the edges around the candidate area. The boundary, center position and coverage of the peeling area are written into the peeling morphology recording area. After multiple rounds of judgment, all peeling morphologies are written into the material peeling identification result record in numerical order to obtain the material peeling identification result.

[0035] The damage description entry submodule fills in the current status description of the blade line by line based on the material spalling identification results. It records the damage situation of each location in the order of the numbers, including structural damage such as cracks and spalling, and produces a wind turbine blade damage description document.

[0036] The damage description entry submodule performs a record-by-record retrieval action based on the material spalling identification results. It extracts spalling records with each number sequentially, and writes the number field during extraction to the damage description number area. Then, it performs a matching call on the crack observation results for the corresponding number, writing the crack direction, length, and boundary morphology of the same number into the temporary damage description area. During the writing process, it judges the crack length range, classifying lengths between 0 and 50 pixels as short cracks, between 50 and 150 pixels as medium cracks, and over 150 pixels as long cracks. It also categorizes crack direction fields, classifying horizontal tendencies as horizontal, moderately inclined as inclined, and steeply inclined as steep. Finally, it reads the spalling areas from the material spalling identification records, judging the size range of the spalling areas, classifying areas with coverage between 100 and 300 pixels as small spalling areas, and areas with coverage between 300 and 900 pixels as medium spalling areas. For each area of ​​detachment, areas exceeding 900 pixels are designated as large-area detachment areas. Spatial classification is then applied to the location field of these detachment areas. For example, areas 0-20 meters from the blade root are classified as root areas, 20-40 meters as mid-section areas, and areas above 40 meters as tip areas. After classification, these areas are written into the location description area. Next, the detachment morphology field is entered, specifying whether the morphology is gentle, moderately undulating, or severely undulating in the morphology description area. After all fields are written, an integration process is performed on the damage content for each number, combining the crack and detachment content sequentially into a paragraph-style descriptive sentence. This sentence is then written into the damage description document line. For example, if number B12 has an 82-pixel-long, moderately inclined crack with moderate boundary changes, and a 400-pixel-area detachment area located in the mid-section of the blade with a moderately undulating morphology, this description is written into the document line. By repeating the above process for all numbers, a summary of blade damage is generated and written into the wind turbine blade damage description document, resulting in the wind turbine blade damage description document.

[0037] Please see Figure 6 The results archiving and aggregation module includes: The structural term extraction submodule extracts structural terms related to the blade tip, web, and root from the wind turbine blade damage description document, lists them alongside the damage description, and generates a list of structural terms. The structural term extraction submodule performs a descriptive text reading action based on the wind turbine blade damage description document. It imports each numbered description from the document into the text parsing area, performing a character-level scan of the description sentences during import. All characters in the sentences are written to the scan cache character by character. Then, phrase splitting is performed on the scan cache, aggregating adjacent characters into segments of two to four characters each, and writing all segments to a temporary segment area. Structural term recognition is then performed on the temporary segment area. This recognition first checks the character length of each segment, setting segments whose length falls within the preset structural term length range as comparable segments. Then, it performs a character-by-character consistency check between the segment characters and the terms in the structural term library. When every character in a segment completely matches the corresponding character in a structural term, the segment is written to the structural segment cache. For example, if the term library contains "blade tip," "stomach," and "blade root," and the character scan detects the segment "blade tip," then the segment is written to the structural segment cache. When "leaf tip" is encountered, it is written into the structural fragment cache. Then, the text before and after the fragment is extended. The subsequent damage description content of the sentence containing the fragment is extracted and written into the associated damage cache to form a parallel relationship between structural terms and damage descriptions. When multiple structural terms appear in a sentence, a binding record between structural fragments and description fragments is created for each term. Then, a duplicate removal action is performed on all binding records. The duplicate structural terms are uniquely retained, and the remaining duplicate content is deleted. At the same time, all structural fragments under different numbers are retained. In the processing example, if the text with number B12 contains "a crack extending along the middle tilt direction appears at the leaf tip position", then "leaf tip" enters the structural fragment, and "crack extending along the middle tilt direction" enters the damage fragment and forms a binding record. After all numbers are parsed, all parallel records of structural fragments and damage fragments are summarized in numerical order and written into the structural term list to obtain the structural term list.

[0038] The region and damage matching submodule, based on the list of structural terms, performs a two-column match between blade regions and damage types, organizes the correspondence between each region and damage, and obtains the region damage matching results. The region and damage matching submodule performs a number retrieval action based on the structural term list, retrieving each numbered entry one by one. During the retrieval process, character extraction is performed on the structural term fields, and the structural terms are written to the structural region cache. Then, the damage description fragment for that number is read, splitting crack fragments, spalling fragments, and morphological fragments into independent damage units and writing them to the damage cache. A two-column matching action is performed on the structural region cache and the damage cache. In the matching action, the regional meaning of the structural terms is first mapped to the interval: "leaf root" is mapped to the 0 to 20 meter region, "stomach" to the 20 to 40 meter region, and "leaf tip" to the 40 meter to the blade tip region. Then, the positional text in the damage fragment is parsed. For example, if "located in the middle of the blade" is detected in the damage description, it is mapped to the 20 to 40 meter region; if "located in the blade tip" is detected, it is mapped to the region above 40 meters. After that, a comparison action is performed one by one between the regional intervals. When the structural term... When the location interval matches the location interval of the damaged segment, the damaged segment is written into the matching result group of the structural term. For example, the structural term B15 is "spine", and its region is 20 to 40 meters. The damaged segment record is "the point of the spalling is located in the middle of the blade". In the double column matching, the two regions are consistent, and the damaged segment is bound to the spine region. Then, the same judgment is performed on the crack segment. For example, the structural term B12 is "blade tip", and its damaged segment "the inclined crack is distributed at the tip position" is consistent with the corresponding interval, so the binding is successful. After processing all binding actions, an integrity check is performed on the region binding status of each number. When there is no matching damaged segment under the structural region, "no damage information" is written to the structural region. Then, the matching relationship of each number is sorted in order, and its number, structural region, crack matching status and spalling matching status are written into the output row. After processing all numbers, all double column matching results are written into the regional damage matching result table to obtain the regional damage matching result.

[0039] The document merging submodule, based on the regional damage matching results, merges the text information under each number into page fragments, completes the table writing, closes each numbered area, and obtains a summary table of blade structural defects.

[0040] The document merging submodule performs a numbered call action based on the regional damage matching results. It reads each structural region item corresponding to a number one by one. During the reading process, it performs a character writing action on the number field, writing the number to the table fragment cache. Then, it performs a sequential writing action on the structural region field, writing the three structural positions (leaf tip, web, and leaf root) to the structural segment cache in a fixed order. Next, it performs a segmented reading action on the damage fragments bound to each structural segment, reading and writing the length, orientation, and boundary change fields of the crack fragment field to the crack description segment, and writing the area, coverage location, and morphology fields of the spalling fragment field to the spalling description segment. During the writing process, it performs interval judgment on the fragment parameters; for example, cracks with a length in the 0-50 pixel range are written as short cracks, and those in the 50-150 pixel range are written as short cracks. Cracks are defined as medium-sized cracks, those larger than 150 pixels are defined as long cracks, and the area of ​​the spalling region covering 100 to 300 pixels is defined as small-area spalling, those covering 300 to 900 pixels are defined as medium-area spalling, and those larger than 900 pixels are defined as large-area spalling. The spatial location field is then categorized: 0 to 20 meters is defined as the root region, 20 to 40 meters as the middle region, and above 40 meters as the tip region. After all fields are written, the description segment is closed, merging the crack description segment and the spalling description segment into page fragments in numerical order. When processing multiple numbers, the above actions are repeated to form multiple fragment records. All fragment rows are then written to the table buffer. After all data processing is complete, the fragments corresponding to each number are written to the blade structure defect summary table, resulting in the blade structure defect summary table.

[0041] A multimodal nondestructive testing method for wind turbine blades based on unmanned aerial vehicles (UAVs) includes the following steps: S1: Obtain the wind farm unit number and blade number, count the number of times the blade number appears in the maintenance record, filter out the frequently appearing blade numbers, and sort out whether each number has appeared multiple times in past tasks. For the frequently appearing numbers, set flight task options and apply the number correspondence to the flight operation table to generate an inspection number list. S2: Based on the number list in the inspection number list, set the flight waypoint position, activate the acquisition equipment to acquire images, acquire images related to each number during flight, record the angle, equipment channel name and flight position when each image is acquired, and distinguish all flight image content by number to generate a part image record master table; S3: Based on multiple sets of images with the same number in the image record table, perform layer boundary line comparison and alignment, observe the light and dark relationship and edge contact direction of the layer seam area, complete the layer fusion, and supplement the shooting source name and integration action status information to generate a fused image registration form. S4: Based on the images in the fused image registration form, observe the crack extension path, material peeling area, and conductive layer fracture fragments. Write down the length direction, boundary shape, and structural location name of each suspicious location item by item, and fill in the blade status description item by item according to the number order to generate a wind turbine blade damage description document. S5: Based on the content of the wind turbine blade damage description document, list the structural terms mentioned in the description such as blade tip, web, and blade root alongside the damage description, output the blade area and damage type in two columns, merge the text under each number into a page fragment, complete the table writing, close each numbered area, and generate a summary table of blade structural defects.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multi-modal non-destructive testing system for wind turbine blades based on unmanned aerial vehicles (UAVs), characterized in that, The system includes: The identification number filtering module obtains the wind farm unit number and blade number, lists the number of times the blade number appears in the maintenance record, sorts out whether each number has appeared multiple times in past tasks, fills the frequent numbers into the flight operation table, and obtains the inspection number list. The image acquisition and registration module calls the numbers listed in the inspection number list to set the flight waypoint position, activates the acquisition device during flight to acquire numbered images, writes the angle, device channel name and flight position when the image is acquired, distinguishes all flight images according to the number, and obtains a summary table of part image records. The channel image processing module retrieves images with the same number from the image record table of the aforementioned parts, aligns the layer boundary lines, observes the light and dark relationship and edge contact direction of the seam area between layers, supplements the shooting source name and action status information, and obtains the fused image registration form. The defect content registration module observes the crack extension path, material peeling area, and conductive layer fracture segment based on the fused image in the fused image registration table. It writes down the length direction, boundary shape, and structural name of each location in the image, and fills in the wind turbine blade damage content one by one in the order of image number to obtain the wind turbine blade damage description document.

2. The multi-modal non-destructive testing system for wind turbine blades of a UAV according to claim 1, characterized in that: The acquisition of wind farm turbine and blade numbers specifically includes turbine number, blade number, blade number maintenance record frequency, flight mission number setting, number mission association, and inspection number list; the image acquisition and registration specifically includes flight waypoint location, image acquisition angle, equipment channel name, flight position label, number-distinguished image content, and part image record; the channel image processing specifically includes layer boundary alignment, joint area brightness and darkness relationship, joint edge direction, layer fusion status, shooting source, and integration action status information; the defect content registration specifically includes crack path, material peeling area, conductive layer fracture fragment, suspected defect location, boundary morphology, structural location description, and blade current status description.

3. The multi-modal non-destructive testing system for wind turbine blades of a drone according to claim 1, characterized in that, The identification number filtering module includes: The numbering statistics submodule obtains the numbering information of wind farm units and blades, counts the number of times each blade number appears in the maintenance record, analyzes the frequency of number occurrence, marks frequently occurring numbers based on frequency, and obtains the number occurrence frequency. The maintenance record filtering submodule filters out the blade numbers that frequently appear in the maintenance records based on the frequency of their occurrence. It then calls up the marked blade number information to filter out the corresponding maintenance tasks from the maintenance records and obtains a list of maintenance record numbers. The flight mission setting submodule, based on the maintenance record number list, configures flight mission options for frequently occurring blade numbers, ensuring that these blade numbers are applied as priority objects in the flight operation table, and derives an inspection number list.

4. The multi-modal non-destructive testing system for wind turbine blades of a UAV according to claim 1, characterized in that, The image acquisition and registration module includes: The waypoint setting submodule sets the flight waypoint positions according to the numbers in the inspection number list, obtains the flight path for each number and determines the waypoints, and generates a flight waypoint list. The image acquisition submodule activates the acquisition device according to the flight waypoint list, acquires the image corresponding to each number, records the angle, device channel name and flight position when each image is acquired, and obtains the image acquisition record. The image annotation submodule annotates each image based on the image acquisition record, recording its angle, device channel, and flight position, distinguishing images according to their numbers, and generating a summary table of part image records.

5. The multi-modal non-destructive testing system for wind turbine blades of a UAV according to claim 1, characterized in that, The channel image processing module includes: The image alignment submodule retrieves multiple sets of images with the same number based on the image record table of the aforementioned area, compares the layer boundary lines, adjusts the position of each layer to align them, and obtains a list of aligned images. The seam inspection submodule, based on the aligned image list, observes the light and dark relationship and edge contact direction of the layer seam area, determines whether there is a discontinuous or unnatural transition at the seam, and obtains the seam inspection result. The layer fusion submodule merges the layers based on the seam inspection results, supplements the shooting source name and integration action status information, and generates a fused image registration form.

6. The multi-modal non-destructive testing system for wind turbine blades of a UAV according to claim 1, characterized in that, The defect content registration module includes: The crack observation submodule observes the extension path of the cracks based on the images in the fused image registration table, records the length, direction and boundary morphology of each crack, and obtains the crack observation results. The material spalling identification submodule identifies material spalling areas in the image based on the crack observation results, records the shape and location of the spalling areas, and obtains the material spalling identification results. The damage description filling submodule fills in the current status description of the blade line by line based on the material spalling identification results. It records the damage situation of each location in the order of number, including structural damage such as cracks and spalling, and produces a wind turbine blade damage description document.

7. The multi-modal non-destructive testing system for wind turbine blades of a UAV according to claim 1, characterized in that, Also includes: The results archiving and aggregation module, based on the content under the numbered sections in the wind turbine blade damage description document, lists the blade tip, web, and root structural entries alongside the damage descriptions, outputs the blade region and blade damage type in two columns, merges the text content under the numbered sections into page fragments, completes the table writing, and closes each numbered region. A summary table of blade structural defects is then obtained. The results archive specifically includes a summary table of blade structural defects, text merging in numbered areas, leaf tip area, web area, leaf root area, damage type differentiation, and complete area closure.

8. The multi-modal non-destructive testing system for wind turbine blades of a UAV according to claim 7, characterized in that, The result archiving and aggregation module includes: The structural term extraction submodule extracts structural terms related to the blade tip, web, and root from the wind turbine blade damage description document, and lists them alongside the damage description to obtain a list of structural terms. The region and damage matching submodule, based on the structural term list, performs a two-column match between the blade region and the damage type, organizes the correspondence between each region and the damage, and obtains the region damage matching result. The document merging submodule, based on the regional damage matching results, merges the text information under each number into page fragments, completes the table writing, closes each numbered region, and obtains a summary table of blade structural defects.

9. A multimodal nondestructive testing method for wind turbine blades based on unmanned aerial vehicles (UAVs), characterized in that, The method is used in the multimodal nondestructive testing system for wind turbine blades of the UAV according to any one of claims 1-8, and includes the following steps: S1: Obtain the wind farm unit number and blade number, count the number of times the blade number appears in the maintenance record, filter out the frequently appearing blade numbers, and sort out whether each number has appeared multiple times in past tasks. For the frequently appearing numbers, set flight task options and apply the number correspondence to the flight operation table to generate an inspection number list. S2: Based on the number list in the inspection number list, set the flight waypoint position, activate the acquisition device to acquire images, acquire images related to each number during flight, record the angle, device channel name and flight position when each image is acquired, and distinguish all flight image content by number to generate a part image record summary table. S3: Based on multiple sets of images with the same number in the image record table of the aforementioned parts, perform layer boundary line comparison and alignment processing, observe the light and dark relationship and edge contact direction of the layer seam area, complete layer fusion, and supplement the shooting source name and integration action status information to generate a fused image registration form. S4: Based on the images in the fused image registration form, observe the crack extension path, material peeling area, and conductive layer fracture fragments. Write down the length direction, boundary shape, and structural location name of each suspicious location item by item, and fill in the blade status description item by item according to the number order to generate a wind turbine blade damage description document. S5: Based on the content of the wind turbine blade damage description document, list the structural terms such as blade tip, web, and blade root mentioned in the description alongside the damage description, output the blade area and damage type in two columns, merge the text under each number into a page fragment, complete the table writing, close each numbered area, and generate a summary table of blade structural defects.

Images