Tower material bending nondestructive testing method and device based on unmanned aerial vehicle

Abstract

The application discloses a kind of tower material bending degree nondestructive testing method and device based on unmanned aerial vehicle, belong to electric power engineering nondestructive testing technical field.The method includes: fixing reflective target at the both ends of component and erecting differential positioning base station;Control unmanned aerial vehicle to fly along preset route, and collect multi-source data by laser radar, industrial camera, differential positioning module and inertial measurement unit carried;Point cloud data is denoised and extracted target center as end point reference coordinate;Least square method is used to fit reference axis, the perpendicular distance of feature point to reference axis is calculated, the maximum distance is used as bending degree and compared with allowable value to output determination result;Generation detection report and component identification binding storage.The device includes unmanned aerial vehicle carrying detection platform, three-dimensional data preprocessing and calibration system, bending degree intelligent calculation module and intelligent flight and data tracing system.The application realizes the quick, nondestructive, automatic bending degree detection of pile storage component.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing technology in power engineering, specifically relating to a method and device for detecting the curvature of pole-type materials using sensors mounted on a drone. Background Technology

[0002] The curvature of pole and tower materials (including steel angle steel, steel pipe components, concrete pole sections, and pre-assembled poles and towers in factories) is an important indicator for measuring their structural stability and safe service life, directly affecting the erection quality and operational safety of power lines. Currently, existing technologies have attempted to utilize drones equipped with lidar for pole and tower structure inspection. For example, Chinese patent application CN121578322A discloses a digital aerial-ground collaborative detection method for pole and tower structural anomalies. This method uses a drone equipped with lidar to perform aerial scanning to acquire three-dimensional point clouds. It extracts parameters such as pole inclination and crossarm skewness and compares them with thresholds to determine structural anomalies. For anomaly poles, it uses ground-based lidar to perform multi-station scanning and stitching to extract multi-dimensional state quantities such as the curvature of the main materials.

[0003] However, the aforementioned existing technologies primarily serve the screening of structural anomalies in in-service transmission line towers. Their point cloud acquisition strategies are not optimized for undelivered tower materials stored in factory warehouses, and they lack rapid, non-destructive methods for locating reference points at both ends of individual components. This makes them unsuitable for batch inspection needs of large quantities of components placed in various orientations during factory quality inspection and warehouse verification. Furthermore, this solution employs a collaborative operation of airborne and ground-based radar, resulting in a complex data acquisition and processing flow and a long inspection time for individual components, failing to meet the efficiency requirements of rapid quality inspection before shipment. In addition, this solution focuses on identifying key node displacements through multi-stage point cloud registration; its curvature calculation relies on fine scanning by ground-based radar and subsequent stitching processing. It has not yet achieved online automatic calculation and real-time determination of curvature based on data collected from a single UAV operation, and the traceability and automation of the inspection data need improvement.

[0004] Therefore, there is an urgent need for an intelligent and non-destructive testing technology solution that is suitable for the scenario of undelivered materials in the factory, in order to solve the technical problems of poor adaptability to the testing of stacked and stored components, low batch testing efficiency, and insufficient automation of curvature calculation in the existing technology. Summary of the Invention

[0005] To address the technical problems of poor adaptability to the inspection of stacked and stored components, low batch inspection efficiency, and insufficient automation in bending degree calculation in the existing technologies, this invention provides a non-destructive inspection method and device for bending degree of pole-type materials mounted on a drone.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for non-destructive testing of the bending degree of pole-like materials mounted on a drone includes the following steps: The first step is test preparation. Magnetic adsorption reflective targets are fixed at the center positions of both ends of the pole component to be tested, serving as reference points for bending measurement; a portable real-time dynamic differential positioning base station is set up, working in conjunction with the positioning module on the UAV to complete coordinate calibration; the ground station imports the component's length, type, and other dimensional parameters, selects the scanning method, and sets parameters such as flight altitude and speed, then starts the UAV to fly automatically along the preset route.

[0007] The second step is data acquisition. During the drone's flight along its route, the lidar acquires 3D point cloud data of the tower surface, the industrial camera simultaneously captures images of the tower surface, the real-time dynamic differential positioning module provides centimeter-level position coordinates, and the inertial measurement unit records the drone's attitude changes. These four types of data are time-aligned during acquisition and transmitted to the ground station in real time.

[0008] The third step is data preprocessing and benchmark extraction. After receiving the data, the ground station removes environmental noise points and fills in missing point cloud areas caused by occlusion. Before testing, the measurement system is calibrated using a standard calibration rod to eliminate measurement errors inherent in the equipment. Environmental data is collected using temperature and humidity sensors to compensate and correct the laser ranging. Then, reflective targets at both ends of the component are identified from the processed data, and the spatial coordinates of the target center are extracted. When the target is tilted due to uneven placement of the component, tilt correction is automatically performed.

[0009] The fourth step is the curvature calculation. An initial baseline is determined based on the coordinates of the two endpoints. Then, the position and direction of the baseline are optimized using the coordinates of multiple feature points on the component surface through the least squares method. Several feature points are selected from the point cloud data of the component surface, and the vertical distance from each feature point to the baseline line is calculated. The maximum value among these vertical distances is taken as the curvature of the component. The system pre-stores allowable curvature values ​​for different types of towers. The calculated results are compared with the allowable values, automatically providing a pass / fail judgment, displaying the specific deviation value, and marking the locations where the curvature exceeds the limit on the point cloud map.

[0010] The fifth step is result output and data storage. The system automatically generates an inspection report, which includes the component number, inspection time, operator, ambient temperature and humidity, curvature value, pass / fail judgment, and point cloud screenshot. All inspection data is bound to the component's unique identifier, encrypted, and stored in the storage unit, supporting queries by component number, inspection time, and other keywords. Through the interface module, the inspection data can be uploaded to the factory's quality management system or exported as a spreadsheet or document file.

[0011] This invention also provides a drone-mounted non-destructive testing device for the bending of pole-like materials, comprising: The drone is equipped with a detection platform, which carries various sensors and automatically flies along a set route to collect data; A 3D data preprocessing and calibration system is used to perform noise reduction, missing data filling, environmental compensation, and target identification on the acquired point cloud and image data. The intelligent curvature calculation module is used to fit the baseline based on the target coordinates, calculate the distance from the feature point to the baseline, and determine whether the curvature is qualified. The intelligent flight and data traceability system is used to plan flight routes, control drones to continuously inspect multiple components, and bind and store inspection data with component numbers and export them externally. The environmental safety auxiliary system is used to shield electromagnetic interference, ensure low-altitude flight safety, protect operator safety, and adopts a reusable and environmentally friendly design.

[0012] Compared with existing technologies, this invention achieves the following beneficial effects through the organic integration and synergistic effect of multiple technical means: 1. Implemented adaptive detection for stacking storage scenarios. Existing technologies primarily target structural anomaly screening for in-service transmission line towers, and their point cloud acquisition strategies are not optimized for undelivered tower materials stored in factory stacks. This invention achieves rapid, non-destructive positioning of reference points at both ends of components through the coordinated operation of a magnetic adsorption reflective target and a real-time dynamic differential positioning module. Specifically, the magnetic adsorption target can be directly fixed to the steel or concrete surface without drilling, bonding, or welding, avoiding damage to the component surface caused by traditional detection methods. Simultaneously, the extraction of the target's center coordinates provides a precise geometric reference for subsequent curvature calculations, solving the technical problem of difficulty in determining reference points in densely stacked scenarios.

[0013] 2. Improved batch testing efficiency Existing technologies employ a combination of airborne and ground-based radar, resulting in complex data acquisition and processing procedures and lengthy detection times for single components. This invention, through a path planning algorithm integrated into an intelligent flight and data tracing system, achieves automated flight control for continuous multi-component detection. Specifically, the algorithm supports importing tower drawings to automatically generate optimal flight paths, and after completing one component, automatically plans a path to fly to the next target, with path deviation controlled within two centimeters. Combined with one-click switching between parallel and surround scanning modes, batch scanning of stacked components, vertically stored components, and pre-assembled towers can be completed without manual intervention, improving detection efficiency by five to ten times compared to manual measurement.

[0014] 3. Online automatic calculation and real-time determination of curvature have been achieved. Existing technologies for bending degree calculation rely on fine scanning by ground-based radar and subsequent stitching processing, and have not yet achieved online automatic calculation based on data collected by a single UAV. This invention, through the integrated design of a bending degree intelligent calculation module and multi-source data acquisition methods, forms a complete automated closed loop from data acquisition to judgment output. Specifically, LiDAR, industrial cameras, real-time dynamic differential positioning modules, and inertial measurement units simultaneously acquire data and align them in time, providing high-precision 3D point cloud data and spatial positioning benchmarks for bending degree calculation. Based on this, the bending degree intelligent calculation module uses the least squares method to fit the benchmark axis, calculates the perpendicular distance from feature points to the benchmark line through vector cross products, automatically takes the maximum value as the component bending degree, and compares it with a preset threshold to output the judgment result. The entire process requires no manual post-processing, significantly improving the automation level of the detection.

[0015] 4. Enabled full-process traceability of testing data. Existing technologies struggle to automatically integrate test data with factory quality inspection systems, resulting in poor data traceability. This invention addresses this by binding test data to unique component identifiers via a data traceability module. Comprehensive information includes test time, operator, equipment number, environmental parameters, curvature values, judgment results, and both raw and processed point cloud data. Combined with the interface module's support for exporting to common formats such as CSV, Excel, and PDF, and its integration with the manufacturer's quality inspection system, a closed-loop data process from test execution to quality acceptance is achieved, meeting compliance and traceability requirements.

[0016] 5. A precision assurance system based on air-ground collaboration and multi-dimensional calibration has been established. This invention integrates multiple calibration methods at the device level, forming a systematic measurement accuracy assurance mechanism. A standard calibration rod is used to establish a measurement benchmark before testing, eliminating systematic equipment errors; a temperature compensation sensor corrects measurement deviations by considering the different thermal expansion coefficients of steel and concrete; and a humidity sensor compensates for the influence of atmospheric refraction on laser ranging. These calibration methods, in conjunction with the raw measurement data from the lidar, real-time dynamic differential positioning, and inertial measurement unit, ensure that the overall measurement deviation is controlled within ±3 millimeters, meeting the high-precision requirements of national standards for curvature detection.

[0017] 6. It takes into account both operational safety and environmental protection requirements. Existing technologies rely on operators climbing stacks or using aerial work platforms, posing safety risks such as falls from heights and injuries from falling heavy objects. This invention utilizes a drone equipped with a detection platform to achieve non-contact measurement, operating entirely from the ground and avoiding the risks of working at heights. Simultaneously, the environmental safety auxiliary system integrates multiple safety mechanisms, including electromagnetic interference shielding, low-altitude flight protection, and emergency shutdown, making it adaptable to the complex electromagnetic environments of equipment such as welding machines and frequency converters within the factory. In terms of environmental protection, it employs rechargeable lithium batteries and reusable magnetic adsorption targets, with digital storage replacing paper records, achieving zero pollutant emissions and low resource consumption. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of the UAV-mounted non-destructive testing method for the bending degree of pole-type materials according to Embodiment 1 of the present invention; Figure 2 This is a flowchart of step 3 (three-dimensional data preprocessing and calibration) of Embodiment 1 of the present invention; Figure 3 This is a sub-flowchart of step 4 (intelligent calculation of curvature) in Embodiment 1 of the present invention; Figure 4 This is an overall architecture diagram of the UAV-mounted non-destructive testing device for the bending degree of pole-type materials according to Embodiment 2 of the present invention; Figure 5 This is a structural diagram of the UAV-mounted detection platform according to Embodiment 2 of the present invention. Detailed Implementation

[0019] To facilitate understanding of this invention, some of the terms used in the text are explained as follows: "Pole and tower materials" refer to structural components used in power engineering, such as steel angle steel, steel pipe components, concrete pole sections, and pre-assembled poles and towers in factories, which are used to support power transmission lines.

[0020] "Curvature" refers to the degree to which the axis of a tower component deviates from the ideal reference straight line. Specifically, it is expressed as the maximum vertical distance from each feature point on the component surface to the reference straight line determined by the two endpoints.

[0021] "Target" refers to a marker with reflective recognition function. In this application, a structure combining a reflective QR code and a magnetic adsorption base is used to fix it at the center position of both ends of the component as a geometric reference point for measuring curvature.

[0022] Real-time dynamic differential positioning (RTK) is a real-time dynamic positioning technology based on carrier phase observations, which can achieve centimeter-level positioning accuracy.

[0023] An "Inertial Measurement Unit" (IMU) is a measuring device consisting of an accelerometer and a gyroscope, used to detect the acceleration and angular velocity of an object.

[0024] The technical solution of the present invention will be described in detail below with reference to specific embodiments. It should be noted that these embodiments are only used to explain the present invention and do not constitute any limitation on the scope of protection of the present invention.

[0025] Example 1 This embodiment provides a non-destructive testing method for the bending of pole-type materials carried by a drone. The method is illustrated by taking the testing process of a batch of steel angle steel components (single length 6 meters, national standard requires total bending of no more than 9 mm) in the factory quality inspection process of a power pole manufacturer as an example.

[0026] Figure 1 This is an overall flowchart of the UAV-mounted non-destructive testing method for the bending of pole-like materials according to Embodiment 1 of the present invention. As shown in the figure, the method includes five main steps: Step 1, test preparation; Step 2, multi-source data fusion acquisition; Step 3, three-dimensional data preprocessing and calibration; Step 4, intelligent bending calculation; and Step 5, result output and data storage. Each step is executed sequentially in chronological order, with clear input-output relationships between successive steps. The detailed internal processes of Steps 3 and 4 are described in [the following text is missing from the original extract]. Figure 2 and Figure 3 It unfolds in the middle.

[0027] Step 1: Test Preparation The operator fixes the magnetic adsorption reflective targets to the center positions of both ends of the angle steel component to be tested, ensuring that the targets are in close contact with the surface of the component. According to experimental verification, when the adsorption force is less than 50 Newtons, the targets may shift or fall off due to airflow disturbances generated by the drone's flight. Therefore, in this embodiment, a magnetic adsorption base with an adsorption force of not less than 50 Newtons is selected.

[0028] A portable RTK base station is set up at the center of the testing area. The base station's coverage radius must cover the entire testing area. In factory storage areas, the coverage radius is generally no more than 200 meters. Considering signal attenuation and obstruction, this embodiment selects a base station with a coverage radius of no less than 500 meters to meet the testing needs of general factory areas and some outdoor storage areas. Positioning calibration with the UAV-borne RTK module is completed to ensure the spatial coordinate system of the entire testing area is unified.

[0029] The operator imported the dimensional parameters (6 meters in length, 75 mm × 75 mm cross-section) of the batch of angle steel components into the ground station system, selected the "parallel scan" mode, and set the flight altitude to 0.8 meters above the component surface, the flight speed to 3 meters per second, and the scan interval to 0.2 meters. The parameters were set based on the following: a flight altitude that is too low (less than 0.5 meters) poses a collision risk, while a flight altitude that is too high (greater than 1.5 meters) reduces the point cloud density and affects measurement accuracy. Orthogonal experiments determined that 0.8 meters above the surface is the balance point between accuracy and safety. The flight speed of 3 meters per second is based on the matching calculation between the lidar point cloud density (100,000 points per second) and the required coverage; too high a speed would result in sparse point clouds along the length direction. A scan interval of 0.2 meters ensures a 15% to 20% overlap between adjacent scan strips, avoiding data omissions. The ground station automatically generates a flight path based on these parameters, initiates the UAV's automatic takeoff, and hovers to the initial detection position.

[0030] Step 2: Multi-source data fusion and acquisition The drone flies at a constant speed along the length of the angle steel component along a preset route. Its onboard lidar emits a laser beam at a point cloud density of at least 100,000 points per second to acquire three-dimensional point cloud data of the angle steel surface. This point cloud density is determined as follows: when the density is below 100,000 points / second, for an angle steel component with a width of only 75 mm, only 3 to 5 points can be obtained in the width direction, making it difficult to accurately fit the surface contour; 100,000 points / second ensures 8 to 10 points are obtained in the width direction, meeting the contour accuracy requirements for curvature calculation. The lidar's ranging accuracy is ±1 mm, a typical indicator for current industrial-grade lightweight lidars, meeting national standards for curvature measurement accuracy.

[0031] The industrial camera simultaneously captures images of the angle steel surface at a resolution of no less than 20 megapixels. This resolution was chosen because, at a shooting distance of 0.8 meters, 20 megapixels allows for a ground resolution of 0.2 millimeters per pixel, sufficient to clearly identify the details of the QR code on the target (the smallest QR code module is approximately 2 millimeters), ensuring reliable target identification. The camera features optical image stabilization to ensure image sharpness even during movement.

[0032] The RTK positioning module records the three-dimensional spatial coordinates of each sampling point in the flight trajectory in real time. The planar accuracy is ±1 cm and the elevation accuracy is ±2 cm. This level of accuracy is standard for current industrial RTK modules and can meet the positioning requirements of registering point cloud data from different flight paths and times to the same coordinate system. Experimental verification shows that the curvature calculation deviation introduced by a ±1 cm planar positioning error is less than 0.5 mm, which is within an acceptable range.

[0033] The inertial measurement unit (IMU) records the pitch, roll, and yaw angle changes of the UAV at a sampling frequency of no less than 100 Hz. The sampling frequency is chosen because the attitude change frequency of the UAV at a speed of 3 m / s is generally between 10 and 20 Hz. According to the Nyquist sampling theorem, the sampling frequency should be no less than twice the highest frequency of the signal. Using 100 Hz ensures at least 5 times oversampling, guaranteeing accurate capture of attitude changes. The measurement error of the pitch and roll angles is no greater than 0.1 degrees. This accuracy can control the point cloud coordinate error introduced by the flight attitude to within 0.17 mm per meter, and its impact on curvature calculation is negligible.

[0034] The four types of data mentioned above are timestamped and aligned using an integrated control board to ensure that point cloud, image, position, and attitude data collected at the same time correspond to each other. The collected data is transmitted to the ground station in real time via a 5G and WiFi dual-mode data transmission module. The dual-mode design is based on the fact that WiFi is suitable for high-speed transmission over short distances (within 50 meters), while 5G is suitable for long-distance or obstacle-crossing transmission. Automatic switching between the two modes adapts to detection scenarios with varying distances and obstructions within the factory area. The parameter settings of a transmission rate of no less than 100 megabits per second and a latency of no more than 10 milliseconds are lower limits calculated based on the amount of data collected in a single detection (approximately 200 megabytes) and real-time processing requirements.

[0035] Step 3: 3D Data Preprocessing and Calibration Figure 2 This is a flowchart of step 3 (3D data preprocessing and calibration) of Embodiment 1 of the present invention. This step takes the multi-source data (LiDAR point cloud, industrial camera image, RTK position data, IMU attitude data) collected in step 2 as input, and sequentially performs operations such as bilateral filtering denoising, point cloud completion, standard rod calibration, temperature compensation correction, humidity compensation correction, YOLO target recognition, and target tilt correction. Finally, it outputs the endpoint reference coordinates P1 and P2 of the two ends of the component, providing a geometric reference for the curvature calculation in step 4.

[0036] After receiving data, the edge computing nodes in the ground station use a bilateral filtering algorithm to remove noise points in the environment (such as dust reflections in the air and interference from surrounding background clutter). The reason for choosing the bilateral filtering algorithm is as follows: this algorithm can preserve the edge features in the point cloud while removing noise, and the sharp edges of the tower components are the key parts for bending measurement, so ordinary Gaussian filtering is not suitable for blurring. After denoising, the signal-to-noise ratio (SNR) is improved to above 45 dB. This SNR threshold has been verified experimentally: when the SNR is below 40 dB, the target recognition accuracy drops significantly (below 95%); when the SNR reaches 45 dB, the recognition accuracy can be stabilized above 99%.

[0037] For point cloud data gaps caused by component stacking and occlusion, a point cloud completion algorithm is used to fill them in, ensuring that the point cloud coverage is not less than 98%. The 98% coverage is based on the following: when the coverage is less than 90%, the curvature calculation may miss the point of maximum curvature, posing a risk of missed detection; according to statistical sampling calculations, when the coverage is not less than 98%, the probability of missing the point of maximum curvature is less than 0.03%.

[0038] Before testing begins, operators calibrate the measurement system using a 1-meter standard calibration rod. Three sizes of standard calibration rods (1 meter, 3 meters, and 6 meters) are available to accommodate calibration requirements for components of different lengths, with a length error not exceeding 0.01 millimeters. The calibration principle is as follows: the calibration rod is placed at the same spatial position within the testing area, and its length is measured using the measurement system of this invention. The difference between the measured value and the actual value is taken as the system's constant error, and this error is subtracted from all subsequent measurements. The theoretical basis of this method is that the measurement system's error remains stable over a short period, and a single calibration can maintain testing accuracy for 2 to 4 hours.

[0039] During the testing process, a temperature compensation sensor collects the ambient temperature in real time (measuring range from -10°C to 50°C, with an accuracy of ±0.2°C). The system then calculates the temperature value based on the coefficient of thermal expansion of the steel (11.5 × 10⁻⁶). -6 / ℃) to correct measurement deviation. The principle of temperature correction is: the accuracy of laser ranging is affected by temperature. For every 10 degrees Celsius increase, the propagation speed of laser in air changes by about 0.02%; at the same time, the dimensional change of steel components under a 30-degree Celsius temperature difference can reach 0.0345% (about 2 mm for a 6-meter-long component), which exceeds the national standard allowable deviation of 22%. Therefore, temperature compensation is indispensable.

[0040] The humidity sensor collects the ambient relative humidity (measurement range 0% to 100%, accuracy ±2%), and the system compensates for the change in the laser's refractive index in the atmosphere based on the humidity value. The principle of humidity compensation is that in high humidity environments (relative humidity above 80%), changes in the laser's refractive index can cause centimeter-level ranging errors. For applications requiring millimeter-level accuracy, such as curvature measurement, humidity compensation is a necessary correction method.

[0041] After completing the above preprocessing and calibration, the system identifies reflective targets at both ends of the components in the point cloud and image based on the YOLO target detection algorithm. The YOLO algorithm was chosen because it achieves a good balance between real-time performance and accuracy, and can achieve a processing speed of over 30 frames per second on edge computing nodes, meeting the needs of real-time on-site detection. The three-dimensional spatial coordinates of the target center are extracted, with a reference point extraction error of no more than 0.5 mm. This error limit is set based on the fact that a 0.5 mm reference point error introduces a curvature calculation deviation of approximately 0.016 mm on a 6-meter-long component (calculated based on the principle of similar triangles), and its impact on the final judgment result is negligible.

[0042] When the target tilts due to uneven stacking of components, the system automatically performs tilt correction. The correction principle is as follows: by identifying the coordinates of the four corner points of the target's QR code, the system uses perspective transformation formulas to project and transform the QR code image viewed from the front view angle to a standard front view, thereby calculating the true coordinates of the target's center in three-dimensional space. This correction method can compensate for a maximum angle of 30 degrees between the target plane and the component's end face.

[0043] In practice, the depth information of the four corner points acquired synchronously by the lidar can be used, combined with the known physical size of the target (such as 50mm×50mm), and the three-dimensional pose of the target plane can be calculated by the PnP algorithm, thereby correcting the center coordinates.

[0044] Step 4: Intelligent Calculation of Curvature Figure 3 This is a flowchart of step 4 (intelligent calculation of curvature) of Embodiment 1 of the present invention. This step takes the endpoint reference coordinates P1 and P2 output from step 3 and the preprocessed point cloud data as input. First, the least squares method is used to fit the ideal reference line L. Then, feature points Pi are selected at equal intervals from the point cloud data. For each feature point, the vertical distance di from it to the reference line L is calculated using the vector cross product formula. By iterative comparison, the maximum value D among all vertical distances is found as the actual curvature of the component. Finally, D is compared with the national standard threshold T, and the result of qualified or unqualified is output. When unqualified, the position exceeding the standard is marked.

[0045] Based on the extracted endpoint coordinates P1(x1,y1,z1) and P2(x2,y2,z2), an ideal reference line L is fitted using the least squares method. The reason for choosing the least squares method is as follows: Since there is a random error of ±0.5 mm in the target extraction, if the line passing through P1 and P2 is directly used as the reference line, this random error will be completely carried over into the subsequent calculations; the least squares method uses multiple sampling points to fit the line, which can effectively suppress the influence of random noise and make the fitted reference line closer to the true geometric axis of the component.

[0046] The equation of the straight line is expressed as (x-x1) / (x2-x1)=(y-y1) / (y2-y1)=(z-z1) / (z2-z1). Feature points are selected every 5 cm along the length direction from the point cloud data of the angle steel component surface. Sampling is performed at 5 cm intervals, and the density is automatically increased to one point per centimeter near the peak of curvature. The 5 cm interval for feature points is based on the following: for a 6-meter-long component, 120 feature points are selected. Experiments show that when the feature point interval is greater than 10 cm (approximately 60 points), local curvature extreme points may be missed (the wavelength of local curvature is generally between 15 and 30 cm); when the interval is less than 3 cm (more than 200 points), the computational load increases exponentially but the improvement in accuracy is not significant (the improvement is less than 5%). 5 cm represents a balance between accuracy and efficiency.

[0047] The perpendicular distance di from each feature point Pi to the reference line L is calculated using the vector cross product algorithm. The formula is di = |vector P1Pi × vector P1P2| / |vector P1P2|. The principle behind this method is that the cross product of vectors P1Pi and P1P2 results in a vector whose magnitude is equal to the area of ​​the parallelogram with P1Pi and P1P2 as adjacent sides. Dividing this area by the magnitude of P1P2 (i.e., the length of the base) yields the perpendicular distance from point Pi to line L. The advantage of this method is that it avoids slope calculations, thus preventing numerical problems related to infinitely large slopes on vertical or horizontal lines.

[0048] After calculating the vertical distances of all feature points, the maximum value is taken as the actual curvature D of the angle steel member, i.e., D = max(d1, d2, ..., dn). The curvature is defined as the maximum vertical distance, which is the curvature measurement method defined in national standards such as GB / T 2694-2018 and GB / T 4623-2014. This application is consistent with these standards to ensure the compliance of the test results.

[0049] In this embodiment, the calculated maximum vertical distance is 7.2 mm. The system pre-stores allowable bending values ​​for different types of towers. For single steel members, the allowable value is no more than 1.5 mm per meter and the total bending is no more than 0.15% of the member's length. This threshold is directly referenced from Clause 6.2 of GB / T 2694-2018 and requires no additional setting. For a 6-meter-long angle steel member, the allowable total bending is 9 mm. The system compares the calculated 7.2 mm with 9 mm, determines the member to be qualified, outputs the specific deviation value of 7.2 mm, and highlights the location of the maximum bending on the point cloud map.

[0050] Step 5: Outputting Results and Saving Data The system automatically generates a test report, which includes: a unique component identifier (batch number and serial number obtained from a QR code scan), test time (accurate to the second), operator's name, equipment number (UAV fuselage number and sensor number), environmental parameters (temperature and humidity values ​​during testing), curvature value (7.2 mm), judgment result (qualified), coordinates of the location exceeding the standard, and a screenshot of point cloud data. The report content is designed according to the information items that should be included in the factory quality certificate documents as specified in Chapter 9 "Inspection Rules" and Chapter 10 "Marking, Packaging, Transportation and Storage" of GB / T 2694-2018, ensuring that the test report meets the compliance requirements of factory quality inspection.

[0051] All test data is bound to the unique identifier of the component and encrypted and stored in a solid-state storage unit with a storage capacity of no less than 1 terabyte and a read / write speed of no less than 1 gigabyte per second. The capacity selection is based on the fact that the amount of test data for each component (including raw point cloud, processed point cloud, images, reports, etc.) is approximately 200 to 300 megabytes. One terabyte can store the test data of approximately 3,000 to 5,000 components, which meets the test data storage needs of a typical tower manufacturer for one production quarter.

[0052] The data supports searching by keywords such as component identification, inspection time, and operator, and the data retention period is no less than 3 years by default.

[0053] Operators export the test data in CSV format via the interface module and upload it to the factory's quality control platform to complete the component's outgoing quality inspection process. The entire inspection process, from takeoff to landing, takes approximately 1 minute and 50 seconds. This time includes: approximately 10 seconds of takeoff hovering, approximately 60 seconds of round-trip scanning along the component (2 seconds for a 6-meter length at a speed of 3 meters per second, plus approximately 10 seconds for round-trip and hovering reversal, for a total actual data acquisition time of approximately 70 seconds), approximately 10 seconds of landing, and the algorithm processing time (approximately 20 seconds, depending on the computing power of the edge computing nodes).

[0054] Example 2 This embodiment provides a UAV-mounted non-destructive testing device for the bending of pole-like materials. This device is used to perform the method described in Embodiment 1. This embodiment describes the device from the perspective of system architecture and module composition.

[0055] Figure 4This is an overall architecture diagram of the UAV-mounted non-destructive testing device for the bending of pole-type materials according to Embodiment 2 of the present invention. The device adopts a layered architecture design, consisting of five subsystems from front-end data acquisition to back-end data management: a UAV-mounted testing platform in the data acquisition layer; a 5G / WiFi dual-mode data transmission module in the communication layer; a three-dimensional data preprocessing and calibration system and a bending intelligent calculation module in the data preprocessing layer; an intelligent flight and data traceability system in the data management layer; and an environmental protection and safety auxiliary system that runs through all layers. In the diagram, solid arrows indicate the forward flow direction of data, and dashed arrows indicate the support relationship between the environmental protection and safety auxiliary system and each layer.

[0056] 2.1 UAV equipped with detection platform Figure 5 This is a structural diagram of the drone-mounted detection platform according to Embodiment 2 of the present invention. This platform is the data acquisition front end of the device, integrating various sensors and auxiliary modules. The diagram shows the following hierarchical relationship: the drone body (industrial-grade multi-rotor drone, shock-absorbing mounting base, quick-release interface), core sensing components (lightweight LiDAR, 20-megapixel industrial camera, RTK positioning module, inertial measurement unit), and auxiliary modules (three-axis stabilized gimbal, portable RTK base station, ground target, backup lithium battery). Solid arrows in the diagram indicate physical assembly relationships, while dashed arrows indicate signal connection relationships.

[0057] The drone-mounted detection platform is used to carry various sensors and automatically fly along a set route to collect multi-source data. The platform specifically includes: an industrial-grade multi-rotor drone body, a lightweight LiDAR, a 20-megapixel industrial camera, an RTK positioning module, an inertial measurement unit, a three-axis stabilized gimbal, a shock-absorbing mounting base, a quick-release interface, a portable RTK base station, a ground target, a 5G plus WiFi dual-mode data transmission module, and a backup lithium battery.

[0058] The drone's main body must have a payload capacity of at least two kilograms and a flight time of at least twenty minutes. The two-kilogram payload limit is based on the fact that the total weight of the LiDAR, camera, RTK module, IMU, and three-axis stabilized gimbal is approximately 1.2 to 1.5 kilograms, with a 0.5-kilogram margin for quick-release interfaces, cables, and spare lithium batteries. The twenty-minute flight time limit is based on the fact that a typical single inspection takes about two minutes, and twenty minutes of flight time is sufficient to complete continuous inspection of at least 8 to 10 components, meeting batch inspection requirements.

[0059] The vibration damping mounting base has a damping coefficient of no less than 80% and can be detachably connected to the sensor module via a quick-release interface. The 80% damping coefficient is based on the fact that the high-frequency vibrations (50 to 200 Hz) generated during UAV flight have an amplitude of approximately 0.5 to 1 mm. After 80% damping, the amplitude is reduced to 0.1 to 0.2 mm. The residual vibration has a negligible impact of less than 0.1 mm on lidar ranging.

[0060] The lidar has a point cloud density of no less than 100,000 points per second, a ranging accuracy of ±1 millimeter, and a measurement range of 0.5 to 100 meters. The selection criteria for point cloud density and ranging accuracy have been explained in step 2 of Example 1. The lower limit of the measurement range of 0.5 meters corresponds to the setting of the flight safety distance, and the upper limit of 100 meters corresponds to the detection requirements of the tower height (the total height of towers with voltage levels of 220 kV and below generally does not exceed 80 meters).

[0061] Industrial cameras feature optical image stabilization and adjustable focal length. Optical image stabilization compensates for low-frequency shaking (1 to 10 Hz) during flight, ensuring image sharpness even in motion. The adjustable focal length adapts to different shooting distances at varying flight altitudes, maintaining the target at an appropriate size in the image (the optimal recognition range is 5% to 15% of the total image pixels).

[0062] The RTK positioning module supports dual-mode positioning using both BeiDou and GPS, with a horizontal accuracy of ±1 cm and an vertical accuracy of ±2 cm. The purpose of dual-mode positioning is to increase the number of visible satellites, maintaining a fixed solution even in environments with obstructions, such as factory areas. The difference between horizontal and vertical accuracy is based on the fact that satellite navigation systems typically have better horizontal accuracy than vertical accuracy, which is determined by the geometric distribution of satellites.

[0063] The sampling frequency of the inertial measurement unit is no less than 100 Hz, and the measurement errors of pitch and roll angles are no greater than 0.1 degrees. The above four modules are linked together via an integrated control board and fixed to a three-axis stabilized gimbal with a stabilization accuracy of ±0.01 degrees. The function of the three-axis stabilized gimbal is to isolate the influence of UAV attitude changes on sensor pointing, ensuring that the sensors always point towards the component surface.

[0064] The portable RTK base station has a coverage radius of at least 500 meters and supports switching between static and dynamic positioning. The ground target uses a reflective QR code and a magnetic adsorption base with an adsorption force of at least 50 Newtons, adaptable to both steel and concrete surfaces. The data transmission module has a transmission rate of at least 100 megabits per second and a latency of at least 10 milliseconds. The backup lithium battery has a battery life of at least 15 minutes and supports hot-swapping.

[0065] 2.2 Three-dimensional data preprocessing and calibration system The 3D data preprocessing and calibration system is used to perform denoising, missing data filling, environmental compensation, and target identification on acquired point cloud data and image data. The system specifically includes: edge computing nodes, a calibration module, and a data storage unit.

[0066] The edge computing node is a portable ground station equipped with an industrial-grade processor, with a computing speed of no less than 2 gigahertz and a cache of no less than 8 gigabytes. It communicates with the sensor module on the UAV via the OPCUA protocol and supports parallel processing of 64 channels of data. The computing speed was chosen based on the fact that the computational complexity of the point cloud data processing algorithm is approximately O(NlogN). For point cloud data of approximately 2 million points on a single component, the 2 gigahertz processor can complete the denoising and completion calculations within 1 second, meeting real-time requirements.

[0067] The calibration module includes a standard calibration rod, a temperature compensation sensor, and a humidity sensor. The standard calibration rod is available in three lengths: one meter, three meters, and six meters, with a length error of no more than 0.01 millimeters. The selection of the three lengths is based on the following: the length of the calibration rod should be on the same order of magnitude as the length of the component being measured. For general tower components (6 to 15 meters), a 6-meter calibration rod can obtain the best calibration results; the one-meter and two-meter lengths are used for short components and local calibration.

[0068] The temperature compensation sensor measures from -10°C to 50°C with an accuracy of ±0.2°C. This range covers the extreme temperature variations throughout the year in the tower manufacturing area and outdoor storage yard. The humidity sensor measures from 0% to 100% relative humidity with an accuracy of ±2%.

[0069] The data storage unit is solid-state storage with a capacity of no less than one terabyte and a read / write speed of no less than one gigabyte per second. It supports real-time backup of detection data, encrypted storage, and historical data traceability.

[0070] 2.3 Intelligent Bending Degree Calculation Module The intelligent curvature calculation module is used to fit a baseline based on the target coordinates, calculate the distance from the feature point to the baseline, and determine whether the curvature is acceptable. This module is embedded in the edge computing node as software and runs the core algorithm. The algorithm fits the baseline axis using the least squares method, calculates the perpendicular distance from the feature point to the baseline line through vector cross product, and takes the maximum value as the component curvature. The system presets curvature thresholds for different tower types, including: for steel single components, no more than 1.5 mm per meter and the total curvature no more than 0.15% of the component length; for concrete poles, no more than 0.2% of the pole length; and for pre-assembled towers, no more than 0.3% of the total height. These thresholds are directly referenced from the national standards GB / T2694-2018 and GB / T 4623-2014. The system automatically compares the calculation results with the corresponding thresholds, outputs the acceptable or unacceptable result and the specific deviation value, and marks the locations exceeding the standard.

[0071] 2.4 Intelligent Flight and Data Tracking System The intelligent flight and data traceability system is used to plan flight routes, control drones to continuously inspect multiple components, and bind and store inspection data with component numbers for export. The system specifically includes a flight control unit, a data traceability module, and an interface module.

[0072] The flight control unit integrates a path planning algorithm, supporting the import of tower CAD drawings to automatically generate the optimal flight path, and also supporting manual setting of flight parameters (including altitude, speed, scan spacing, and flight mode). The system supports continuous detection of multiple components, automatically planning a path to fly to the next target after completing the detection of one component, with a path deviation of no more than two centimeters. The allowable range of 2 centimeters in path deviation is based on the fact that the minimum distance between adjacent components is generally 50 to 100 centimeters; a deviation of 2 centimeters will not cause the UAV to collide with adjacent components.

[0073] The data traceability module binds the test data to the unique identifier of the component, including all information such as test time, operator, equipment number, environmental parameters (temperature, humidity), curvature value, judgment result, original point cloud data and processed data.

[0074] The interface module supports integration with the manufacturer's quality inspection system and is compatible with data export in formats such as CSV, Excel, and PDF.

[0075] 2.5 Environmental Protection and Safety Auxiliary System The environmental safety auxiliary system provides electromagnetic interference shielding, low-altitude flight protection, personnel safety protection, and environmentally friendly design. Specifically, the system includes: an electromagnetic interference shielding unit, a low-altitude flight protection unit, a personnel safety protection unit, and an environmental design unit.

[0076] The electromagnetic interference shielding unit encapsulates the sensing module and flight control unit within a metal shielding shell, and its electromagnetic interference resistance meets the GB / T 17626.3 standard.

[0077] The low-altitude flight protection unit sets a lower limit for flight altitude (not less than 0.5 meters from the ground) and a safe distance from the tower surface (not less than 0.3 meters). During flight, it monitors the flight attitude in real time and automatically hovers and alarms when abnormalities occur.

[0078] The personnel safety protection unit emits audible and visual warnings when the drone is in operation, and the ground station is equipped with an emergency stop button that can remotely cut off the drone's power.

[0079] The environmentally friendly design unit uses a rechargeable lithium battery (cycle life of no less than 1,000 times), the ground target is reusable (lifespan of no less than 500 times), and the point cloud and image data are stored digitally.

[0080] Example 3 This embodiment provides an application case of concrete pole bending detection based on the technical solution of the present invention, which is applied to the material acceptance scenario before power engineering construction.

[0081] A power engineering company purchased a batch of circular concrete poles, each 12 meters long. According to the GB / T 4623-2014 standard, the total curvature must not exceed 0.2% of the pole length, or 24 millimeters. These poles were stored horizontally in three layers within the factory premises, with five poles arranged side-by-side in each layer. Due to the dense stacking, manual climbing for measurement posed safety risks and was inefficient.

[0082] Operators set up a portable RTK base station next to the stack, fixing magnetically attached targets to the center positions of both ends of the first pole. Despite the rough surface of the concrete pole, the magnetically attached base still provided an attraction force of at least 50 Newtons. The ground station imported parameters for a 12-meter pole length, selected the "parallel scan" mode, and set the flight altitude to 1.2 meters above the pole surface (a safety distance was appropriately increased considering the larger cross-section of the concrete pole compared to steel components), the flight speed to 2.5 meters per second, and the scan interval to 0.25 meters. The flight speed of 2.5 meters per second is slightly lower than the 3 meters per second in Example 1 because the surface roughness of the concrete pole is greater, resulting in a slightly lower point cloud signal-to-noise ratio compared to steel components. Appropriately reducing the flight speed helps increase the point cloud density and improve the signal-to-noise ratio.

[0083] After takeoff, the drone flies at a constant speed along the length of the utility poles, taking approximately 2 minutes to collect data from the first pole. After completing one pole, the flight control unit automatically guides the drone to the starting position of the second pole, with a path deviation of 1.5 cm, allowing for continuous inspection without manual intervention. The total inspection time for all 15 poles in the stack is approximately 35 minutes, with an average inspection time of approximately 2.3 minutes per pole, including the interval time for the drone to transfer between the stacks.

[0084] During data processing, the temperature compensation sensor detected an ambient temperature of 35 degrees Celsius. The system automatically substituted the coefficient of thermal expansion of the concrete poles (1.0 × 10⁻⁵ degrees Celsius) into the measurement deviation correction formula to compensate for dimensional expansion errors under high-temperature conditions. The calculated curvature of one pole was 22 mm, within the allowable range of 24 mm, and was therefore deemed acceptable. The curvature of the other pole was 31 mm, exceeding the allowable value of 24 mm, and was therefore deemed unacceptable by the system. The system marked the location of the excess in the point cloud data as approximately 3.5 meters off-center from the middle of the pole. Operators verified the pole based on the location of the excess and removed it from the acceptable batch, preventing unqualified materials from entering the construction site.

[0085] Example 4 This embodiment provides an application case of pre-assembled tower bending detection based on the technical solution of the present invention, which is applied to the factory quality inspection scenario of pre-assembled whole in the factory.

[0086] A tower manufacturer completed the pre-assembly of a 220 kV transmission line tower in the factory. The tower is 35 meters high. According to the requirements of GB / T 2694-2018 standard, the curvature of the pre-assembled tower shall not exceed 0.3% of the total height, i.e., 105 mm. The pre-assembled tower was fixed vertically in the factory assembly area.

[0087] Operators secured magnetically attached targets at the center of both ends of the tower, at approximately 32 meters in height, avoiding the lightning rod at the very top. Due to the tower's height, securing the top target required a small lifting vehicle, but once secured, subsequent inspections did not require personnel on the tower. The ground station imported the 35-meter tower height parameter, selected the "surround scan" mode, and set a 15-meter surround radius, a layered scanning altitude from 2 meters above the ground to 2 meters above the tower top, and a 0.3-meter interval between each layer. The 15-meter surround radius was chosen because the maximum width of a 35-meter tower is approximately 8 meters; a 15-meter radius ensures a safe flight distance (no less than 0.3 meters) while also preventing the lidar's incident angle from exceeding 45 degrees, thus ensuring point cloud quality. Layered scanning was chosen because in surround scan mode, it is difficult for the UAV to simultaneously collect point clouds from the tower's base and top in a single flight; layered and segmented collection is necessary.

[0088] After the UAV takes off, it performs spiral ascending flight around the tower pole, while completing the collection of point cloud, image, position and attitude data. The total collection time is about 8 minutes. Among them, the collection at the bottom layer (0 to 10 meters) takes about 2 minutes, the collection at the middle layer (10 to 25 meters) takes about 3 minutes, and the collection at the upper layer (25 to 37 meters) takes about 3 minutes.

[0089] During the data processing, due to the complex tower body structure (including multiple components such as main members, diagonal members, cross arms, etc.), the amount of point cloud data is large. After the edge computing node processes the data in parallel, the system identifies the coordinates of two target points and fits the reference straight line of the full height. The algorithm extracts a total of about 350 feature points of each key node on the tower body surface (including main member nodes, cross arm hanging points, etc.), and calculates the perpendicular distance from each feature point to the reference straight line respectively. The maximum perpendicular distance appears at the right hanging point of the middle-phase cross arm, which is 98 mm and does not exceed the allowable value of 105 mm, so it is judged as qualified. After the inspection report is automatically generated, it is directly uploaded to the manufacturer's quality control system through the interface module as the factory qualification certificate of this tower pole.

[0090] Embodiment 5 This embodiment provides an application case of material storage quality review based on the technical solution of the present invention, which is applied to the regular sampling inspection scenario of the material warehouse of the power company.

[0091] There are about 300 pieces of tower pole materials of different batches and different types (including angle steel, steel pipe, concrete pole) stored in a material warehouse of a power company. The warehouse management department needs to conduct quality review on the in-stock materials quarterly, and focuses on whether there is natural bending deformation caused by long-term storage.

[0092] Due to the mixed types of in-stock materials, dense stacking and narrow aisles, it is difficult to efficiently implement traditional detection methods. The operator uses the device of the present invention to set up an RTK base station in the warehouse aisle. The signal coverage radius of the base station is 500 meters, which can cover the entire storage area. The ledger information of the in-stock materials is pre-stored in the ground station, including the unique identification code, type, specification size and storage time of each material. The operator selects the "batch detection" mode, and the system automatically generates an inspection path according to the storage area plan and the material placement position, and guides the UAV to fly to each material storage position in turn.

[0093] For different materials, the system automatically switches scanning modes: flat-stacking angle steel and steel pipes use a "parallel scanning" mode, while upright concrete poles and pre-assembled components use a "surround scanning" mode. Scanning parameters are automatically matched according to the dimensions of the materials. The inspection time for each component is controlled between 1 minute 30 seconds and 2 minutes 30 seconds. Within one workday, a single person and a single machine completed the random inspection of all 300 inventory items. The inspection revealed that the bending of three long-stored angle steel components exceeded the allowable value. The system marked the location of the exceeding standard in the inspection report and prompted warehouse management personnel to move them to the pending processing area. All inspection data is linked to batch entry records, forming a digital archive of the inventory material quality status.

[0094] Example 6 This embodiment provides an implementation example of an alternative sensor solution to illustrate the extended applicability of the technical solution of the present invention.

[0095] In scenarios with lower detection accuracy requirements (measurement deviation allowed to be ±5 mm) and good ambient lighting conditions (indoors or in a shaded factory shed), a binocular vision camera can be used instead of a lidar as the 3D data acquisition sensor. In this embodiment, the lightweight lidar is removed from the quick-release interface and replaced with a binocular vision camera. The binocular vision camera calculates depth information and generates 3D point cloud data by using the parallax of the left and right lenses. Since no laser emitting device is required, the hardware cost is reduced by approximately 60% compared to the lidar solution.

[0096] Correspondingly, due to the low point cloud density of the binocular vision camera (approximately 20,000 to 30,000 points per second) and its significant susceptibility to lighting conditions, an illumination compensation algorithm needs to be added to the data processing steps, and interpolation processing is performed on the point cloud in low-density areas. Target recognition still uses reflective QR codes, but an auxiliary lighting source needs to be added to ensure that the QR codes can be recognized under low-light conditions. The remaining steps (detection preparation, data acquisition, curvature calculation, and result output) are basically the same as the LiDAR solution. Actual measurements show that the curvature measurement deviation of this embodiment for a 6-meter-long steel component is ±4 millimeters, which can meet the needs of some low-precision detection scenarios, but cannot meet the national standard requirement of ±3 millimeters. This embodiment illustrates that the device design of this invention achieves sensor replaceability through a quick-release interface, allowing for flexible selection of sensor configurations based on actual detection accuracy requirements and cost budgets.

[0097] Example 7 This embodiment provides an implementation example of an alternative positioning scheme to illustrate the applicability of the technical solution of the present invention in special environments.

[0098] In enclosed factory areas or indoor environments where RTK base stations cannot be set up, visual SLAM positioning can be used as an alternative to RTK base station positioning. In this embodiment, instead of setting up a portable RTK base station, an industrial camera mounted on a drone continuously acquires image sequences of the ground and surrounding environment. The drone's pose is then calculated in real time using a visual SLAM algorithm to establish a local coordinate system for the detection area. This method eliminates the need for additional base stations and is simpler to operate.

[0099] However, visual SLAM localization has the following limitations: in open factory areas with little ground texture (such as cement floors without markers), there are insufficient visual feature points, increasing the localization error to ±3 to 5 centimeters; in densely stacked areas, the visual viewpoint is easily obstructed, leading to tracking loss. Therefore, this embodiment is suitable for indoor factory buildings and other scenarios with rich texture features (such as ground markings and equipment labels) and where high-precision absolute coordinates are not required. In practical applications, the RTK solution should be preferred, and the visual SLAM solution should only be used as an alternative when RTK is unavailable.

[0100] The above-described embodiments are merely preferred embodiments of the present invention, used to help those skilled in the art understand and implement the present invention, and are not intended to limit the scope of protection of the present invention. Equivalent modifications, substitutions, or variations made by those skilled in the art to the specific embodiments based on the technical concepts disclosed in the present invention, without requiring creative effort, or applications of the technical solutions of the present invention to other similar technical fields (such as bending detection of communication poles, wind turbine towers, bridge steel structures, and similar components), all fall within the scope of protection of the present invention. The scope of protection of the present invention is determined by the claims. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for non-destructive detection of tower material bending based on a UAV, characterized in that, Includes the following steps: Test preparation: Fix reflective targets to both ends of the tower components to be tested, and set up real-time dynamic differential positioning base stations; Data acquisition: Control the UAV to fly along a preset route, acquire three-dimensional point cloud data of the tower surface through the onboard LiDAR, simultaneously capture images of the tower surface through the onboard industrial camera, and acquire position coordinates and attitude data through the onboard real-time dynamic differential positioning module and inertial measurement unit, respectively, and transmit the above data to the ground station in real time; Data preprocessing and benchmark extraction: The received point cloud data is denoised and incomplete, the reflective targets at both ends of the component are identified and the spatial coordinates of the target center are extracted as the endpoint benchmark coordinates; Bending calculation: Based on the reference coordinates of the two endpoints, the least squares method is used to fit the reference axis. Multiple feature points are selected from the point cloud data of the component surface, and the vertical distance from each feature point to the reference axis is calculated. The maximum value among all vertical distances is taken as the bending of the component, and compared with the preset bending allowable value to output the judgment result. Results output and data storage: Automatically generate inspection reports containing component numbers, curvature values ​​and judgment results, and bind and store inspection data with the unique identifier of the component.

2. The UAV-based non-destructive testing method for tower material bending according to claim 1, characterized in that, In the curvature calculation, the vector cross product algorithm is used to calculate the vertical distance from the feature point to the reference axis.

3. The method for non-destructive testing of pole material bending based on UAV according to claim 1, characterized in that, In the data preprocessing and benchmark extraction process, after denoising and imputation of the point cloud data, the following steps are also included: calibrating the measurement system using a standard calibration rod to eliminate system errors; and collecting environmental data using temperature and humidity sensors to perform temperature compensation correction and humidity compensation correction on the laser ranging, respectively.

4. The UAV-based non-destructive testing method for tower material bending according to claim 1, wherein, During data acquisition, the point cloud density of the lidar is no less than 100,000 points per second, the resolution of the industrial camera is no less than 20 million pixels, the planar accuracy of the real-time dynamic differential positioning module is ±1 cm, and the sampling frequency of the inertial measurement unit is no less than 100 Hz.

5. A non-destructive testing device for the bending degree of pole materials based on unmanned aerial vehicles (UAVs), characterized in that, The apparatus for performing the method according to any one of claims 1-4 comprises: The drone is equipped with a detection platform, which carries sensors and automatically flies along a set route to collect multi-source data; A 3D data preprocessing and calibration system is used to denoise, fill in missing parts, compensate for environmental conditions, and identify targets in the collected point cloud data and image data in order to extract the endpoint reference coordinates of the two ends of the component. The curvature intelligent calculation module is used to fit the baseline based on the least squares method, calculate the vertical distance from the feature point to the baseline through the vector cross product and take the maximum value as the curvature, and compare the curvature with the preset threshold to output the judgment result. The intelligent flight and data traceability system is used to plan flight routes, control drones to continuously inspect multiple components, and bind and store inspection data with component numbers and export them externally.

6. The unmanned aerial vehicle based non-destructive testing device for tower material bending degree according to claim 5, characterized in that, The reflective target is a magnetic adsorption reflective target with an adsorption force of not less than fifty Newtons, used to adsorb and fix it at the center of both ends of the tower component.

7. The UAV-based non-destructive testing device for tower material bending according to claim 5, wherein, The three-dimensional data preprocessing and calibration system includes an edge computing node and a calibration module. The calibration module includes a standard calibration rod, a temperature compensation sensor, and a humidity sensor. The standard calibration rod is used to establish a measurement benchmark to eliminate system errors. The temperature compensation sensor and humidity sensor are used to perform temperature compensation correction and humidity compensation correction for laser ranging, respectively. 8.The unmanned aerial vehicle based non-destructive testing device for tower material bending according to claim 5, wherein, The intelligent flight and data traceability system includes a flight control unit and a data traceability module. The flight control unit integrates a path planning algorithm, supports continuous detection of multiple components and automatically plans the path to the next target, with a path deviation of no more than two centimeters. The data traceability module binds the detection data with the unique identifier of the component. The detection data includes the detection time, operator, equipment number, environmental parameters, curvature value, judgment result and original point cloud data. 9.The unmanned aerial vehicle based non-destructive testing device for tower material bending according to claim 5, wherein, The UAV-borne detection platform includes an industrial-grade multi-rotor UAV body, and a lightweight lidar, industrial camera, real-time dynamic differential positioning module, and inertial measurement unit mounted on it via a shock-absorbing mounting bracket and quick-release interface; the lidar has a point cloud density of no less than 100,000 points per second and a ranging accuracy of ±1 millimeter; the real-time dynamic differential positioning module supports dual-mode positioning of Beidou and GPS, with a planar accuracy of ±1 centimeter.

10. The UAV-based non-destructive testing device for tower material bending according to claim 5, wherein, It also includes an environmental safety auxiliary system, which includes: an electromagnetic interference shielding unit for encapsulating the sensing module and flight control unit in a metal shielding shell; a low-altitude flight protection unit for setting the lower limit of flight altitude and the safe distance from the tower surface; and an environmentally friendly design unit using rechargeable lithium batteries and reusable targets.

Images