An unmanned aerial vehicle autonomous inspection method and system for tower crane outer surface defects

Abstract

This invention discloses an autonomous UAV inspection method for defects on the outer surface of tower cranes, comprising the following steps: constructing a fused 3D model based on the structural parameters and construction environment data of the tower crane; collecting the tower crane's operating status parameters and the physical data of the UAV's starting point, and comparing and calibrating the parameters with the real-time operating status parameters of the fused 3D model; identifying areas with unrectified defects based on historical inspection data, and updating the trajectory planning weights of the corresponding areas; generating an adaptive inspection trajectory; the UAV autonomously flying according to the 3D inspection trajectory and collecting video of the tower crane's outer surface at each trajectory node; extracting keyframes and preprocessing images from the collected inspection videos, identifying cracks and corrosion defects on the tower crane's outer surface based on image features, and obtaining the defect type, location, and quantitative parameters; and evaluating the defect level based on the defect category, the importance of the area, and historical defect trends, generating a defect evaluation result.

Description

Technical Field

[0001] This invention relates to the field of drone inspection technology, specifically to a drone-based autonomous inspection method and system for defects on the outer surface of tower cranes. Background Technology

[0002] Tower cranes (hereinafter referred to as "tower cranes") are core heavy equipment on construction sites, and the integrity of their external surface structure is directly related to construction safety. Defects such as cracks and corrosion are prone to appear on the outer surface of tower cranes. If these are not detected and rectified in a timely manner, they may lead to structural failure or even collapse accidents. The existing inspection of defects on the outer surface of tower cranes mainly has the following problems:

[0003] Manual inspection is unsafe and inefficient: manual tower crane inspection involves the risk of falling from heights, and is affected by blind spots and physical limitations. Defects such as cracks and minor corrosion are easily missed, and the inspection of a single tower crane takes a long time.

[0004] Traditional drone inspection path planning is inaccurate: existing drones mostly use fixed routes or manual operation, which do not fully take into account the structural characteristics of tower cranes and the distribution of areas with high incidence of defects. This results in problems such as path redundancy and incomplete coverage of key areas. Furthermore, the impact of dynamic obstacles (such as construction elevators) in the construction scene is not considered, leading to a high risk of collision.

[0005] The defect identification process suffers from weak anti-interference capabilities: Construction sites are subject to numerous environmental interference factors such as dust, strong light, and shadows. Traditional identification algorithms are not specifically optimized for these factors, resulting in high false positive and false negative rates for cracks and corrosion defects. Furthermore, the algorithms fail to focus on core defect types, leading to functional redundancy.

[0006] The defect assessment system is unscientific: it lacks quantitative assessment standards for core defects on the outer surface of tower cranes, fails to consider defect development trends and the risk of multiple defects overlapping, and the assessment results are difficult to directly guide rectification work. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this solution proposes an autonomous UAV inspection method for defects on the outer surface of tower cranes, comprising the following steps:

[0008] Step 1: Construct a fusion 3D model of the tower crane and its environment. Based on the structural parameters of the tower crane and the construction environment data, a fusion 3D model is constructed. The fusion 3D model includes the key structural areas of the tower crane, the static obstacle areas, and the dynamic activity range of movable obstacles.

[0009] Step 2: Status Acquisition and Model Calibration. Collect the tower crane's operating status parameters and the physical data of the UAV's starting point, and compare and calibrate them with the real-time operating status parameters of the fused 3D model. Identify areas with unrectified defects based on historical inspection data, and update the trajectory planning weights for the corresponding areas. The parameters compared include, but are not limited to, boom rotation angle, amplitude, lifting height, tower verticality, current lifting weight, wire rope tension, and real-time wind speed and ambient temperature at the tower crane's location. Determine whether each parameter exceeds the preset safety threshold. If so, suspend the inspection process and trigger a safety warning, correct the coordinate parameters of the UAV's starting point in the model, add temporary obstacle annotations, and update the environmental safety boundary.

[0010] Step 3: Generate an adaptive inspection trajectory. Based on the calibrated fused 3D model, and taking into account the importance of each inspection area, environmental factors, and the UAV's endurance, a path planning algorithm is used to generate a 3D inspection trajectory that satisfies spatial obstacle avoidance, defect coverage, and energy constraints.

[0011] Step 4: Perform autonomous inspection. The UAV flies autonomously according to the three-dimensional inspection track and collects video of the tower crane's outer surface at each track node. During the acquisition process, the imaging parameters are automatically adjusted according to the ambient lighting conditions to ensure image quality.

[0012] Step 5: Defect identification and processing. Keyframe extraction and image preprocessing are performed on the collected inspection video. Based on image features, cracks and corrosion defects on the outer surface of the tower crane are identified to obtain the type, location and quantitative parameters of the defects.

[0013] Step 6: Defect assessment and output. Based on the defect category, the importance of the area, historical defect trends, and the risk of multiple defects overlapping, a defect level assessment is performed, and a defect assessment result is generated.

[0014] In some embodiments, step 1 specifically includes:

[0015] The core structural parameters of the tower crane are obtained, including tower height, number and dimensions of standard sections, boom length and segmentation, slewing bearing diameter, and location and quantity of attachment devices. Simultaneously, environmental data of the construction site is collected, covering the coordinates of surrounding fixed obstacles and the location of ground-based safe landing and lifting points. BIM technology is used to integrate the above structural parameters with the environmental data, generating a 1:1 scale fused 3D model of the tower crane and its environment, with a model accuracy of ≤±3mm.

[0016] In the 3D model, key structural areas, namely high-risk areas for cracks / corrosion on the tower crane's outer surface and environmental safety boundaries, are clearly marked. High-risk areas for cracks / corrosion include boom welds, attachment frame connection points, standard tower section connection seams, and the outer surface of the slewing bearing. A dynamic obstacle marking function is added. For mobile devices around the tower crane, the movement trajectory range is fitted based on their historical motion data and marked in the model. A dynamic safety distance of ≥2m is set to ensure that potential collision risks from dynamic obstacles are avoided during subsequent trajectory planning.

[0017] In some embodiments, step 2 specifically includes:

[0018] Step 2-1: Collect the real-time operating status parameters of the tower crane and determine whether they exceed the safety threshold. If so, suspend the inspection process and trigger the early warning mechanism to avoid missed inspections or collision risks when the tower crane is in an abnormal posture. The real-time operating status parameters include, but are not limited to, the current angle of the boom, the rotation position of the slewing bearing, and the verticality deviation of the tower body.

[0019] Step 2-2: Collect the precise physical coordinates of the UAV's starting point and information on surrounding obstacles using the onboard GPS / BeiDou dual-mode positioning module and LiDAR, and compare and calibrate them with the corresponding data of the fused 3D model; if the coordinate deviation is >5cm, automatically calibrate the starting point coordinates and the position of surrounding obstacles of the 3D model to ensure that the model is consistent with the physical environment.

[0020] Steps 2-3: Query the tower crane's historical inspection records to identify key areas with unrectified defects; adjust the node density and search priority of the corresponding areas in subsequent path planning based on the identification results. Specifically, if a key area has unrectified defects, reduce the spacing between path nodes in that area from 0.8m to 0.5m and increase the priority weight of the nodes in path search.

[0021] In some embodiments, step 3 specifically includes:

[0022] Step 3-1: Path search is performed based on the calibrated fused 3D model, tower crane real-time status parameters, and preset inspection priorities. During the path search, node priority weights are assigned to different inspection areas, and an environmental adaptation factor based on real-time wind speed is introduced into the path cost function. The path cost function of the trajectory sampling algorithm... Specifically:

[0023]

[0024] in Basic cost; As an environmental adaptation factor, This refers to the real-time wind speed. The node priority weights are as follows: bolted connections have a weight of 1.5, welds have a weight of 1.2, and other areas have a weight of 1.0.

[0025]

[0026] in and These are the relevant weights, , , The path length is normalized, with the normalization benchmark being the maximum possible inspection path length of the tower crane. The turning angle is normalized, with a normalization reference of 180°. and The values ​​were obtained through multiple sets of on-site measurements at construction sites, and optimized to minimize the combination of flight track energy consumption and shooting shake.

[0027] Step 3-2: Smooth and optimize the generated initial path. For path segments with excessively large turning angles, use B-spline curve fitting to ensure continuous track curvature. For path segments with turning angles > 90°, use cubic B-spline curve fitting. The fitting formula is:

[0028]

[0029] in, This represents the coordinates of the point corresponding to parameter t on the B-spline curve, which is the position of a point on the UAV inspection path. The control vertices of the B-spline curve represent multiple critical path points on the turning path of the UAV. This represents a cubic B-spline basis function. The path search direction is dynamically adjusted using weights and factors to adapt the flight path to real-time environmental conditions, reducing the impact of wind resistance on flight stability. The node density of the flight path is adapted to the regional risk level: node spacing is ≤0.8m in areas with high incidence of cracks / corrosion, and ≤1.2m in general areas. For segments in the global path with turning angles >90°, B-spline curve fitting optimization is used to ensure continuous path curvature and avoid image shake caused by sharp UAV turns.

[0030] Step 3-3: Based on the UAV's endurance parameters and flight path energy consumption prediction (considering flight distance, altitude changes, turning radius, real-time wind speed, node task characteristics, and obstacle constraints), determine whether task segmentation is necessary and set intermediate docking points. Finally, generate an inspection flight path containing flight parameters for each node, including flight speed, dwell time, and shooting angle. Specifically, if the total energy consumption > 80% of the battery capacity, automatically split the flight path into 2-3 segments. Allocate the inspection range of each segment according to the principle of "prioritizing areas with high incidence of cracks / corrosion." Set intermediate docking points near the tower crane's attachment device. Support segmented inspection mode to avoid missing critical defects due to insufficient endurance.

[0031] Step 4: The drone flies along the precise inspection trajectory generated in Step 3, maintaining stability via an onboard attitude sensor to ensure an angle deviation of ≤±3° and accurately matching the preset speed and altitude parameters of the trajectory. An onboard 4K high-definition camera captures key point videos, with each trajectory node capturing for at least 15 seconds at a resolution of ≥3840×2160, clearly showing color changes in rusted areas and the edge morphology of cracks, providing a high-quality data source for subsequent defect identification. The onboard camera is equipped with a light sensor to detect ambient light intensity in real time—if the light intensity is <500 lux, a supplementary light is automatically activated; if the light intensity is >50000 lux, a polarizing filter is automatically activated to reduce the impact of strong light reflection on video clarity, ensuring high-quality capture of rusted area color and crack edges under different lighting conditions. The collected video data and drone flight status are transmitted to a ground terminal in real time. The ground terminal dynamically displays the trajectory execution progress and real-time images, supporting real-time intervention in abnormal situations.

[0032] In some embodiments, step 5 specifically includes:

[0033] Step 5-1: Extract keyframes from the inspection video to obtain image groups covering the area of ​​each node; ensure that the number of extracted frames for a single node is ≥8, while ensuring the complete view of the area covered by the node.

[0034] Step 5-2: Identify the defect categories in the image group and perform image preprocessing for different defect categories;

[0035] Step 5-3: Perform defect identification for images of different defect categories, and output the defect type, location coordinates, and quantization parameters.

[0036] In some embodiments, step 5-2 specifically includes:

[0037] Image preprocessing for crack defects: Gaussian blur filtering is applied sequentially to the extracted image groups to remove high-frequency noise. A Sobel operator in the x-direction is used for convolution to enhance the crack edge gradient. The images are then converted to black-and-white binary images by setting a binarization threshold. Specifically, a Gaussian blur algorithm with a standard deviation σ=0.5 is used to filter the original images, removing high-frequency noise (such as dust particles and lens noise). This is because a small σ is insufficient to remove interference from dust, strong light, and shadows, while a large σ blurs the fine crack edges while removing noise. Convolution operations are performed on the blurred image. Since tower crane cracks are mostly distributed longitudinally / obliquely, the convolution kernel in the x-direction can specifically enhance the crack edge gradient, making the original crack outline clearer and the contrast higher. Finally, a binarization threshold of 128 is set to convert the edge-enhanced image into a black-and-white binary image. When the pixel gray value is ≥128, it is identified as a "crack area" and marked as white; when the pixel gray value is <128, it is identified as a "background area" and marked as black. This threshold is obtained by comparing the gray value distribution of 500 sets of tower crane crack samples.

[0038] Image preprocessing for rust defect categories: Based on the feature that the rust color is similar to that of the tower body, the extracted image groups are sequentially subjected to illumination distribution equalization based on the Retinex image enhancement algorithm and converted to the HSV color space to enhance the color difference contrast between the rust area and the tower, and avoid misjudgment of rust caused by strong light or shadow.

[0039] In some embodiments, step 5-3 specifically includes:

[0040] For crack defect identification: The U-Net semantic segmentation algorithm is used to segment the preprocessed image and extract the crack region to be inspected. The Canny edge detection algorithm is used to capture crack edges in the segmented region, combined with Hough line detection to determine crack continuity. When the detected crack length and width both exceed a preset threshold, it is determined to be a crack defect, and its type, location, length, and width quantification parameters are output. The specific standard is crack length > 5mm and width > 0.2mm, and the threshold is determined by the "Industry Standard and Engineering Specification for Tower Cranes (Special Equipment)" GB / T 3811-2008. Five recognition combinations were tested: U-Net+Canny (accuracy 94.0%), MaskR-CNN+Canny (92.5%), U-Net+Sobel (89.8%), YOLO+Canny (87.3%), and traditional Canny (78.6%). U-Net+Canny is the optimal combination because it can accurately segment the crack region (excluding background interference) and capture slender edges.

[0041] For the identification of corrosion defects: The ResNet50 image classification algorithm is used to perform binary classification identification of "corroded / non-corroded" in the preprocessed image; for the image classified as corroded, the area ratio of the corroded region is further calculated by combining HSV color space analysis. Specifically, when the area ratio of the corroded region to the inspected area is >10%, it is judged as a corrosion defect. The threshold is determined based on the "Safety Specifications and National Standards for Tower Cranes (Special Equipment)" GB / T 3811-2008, GB 50205-2020, and GB / T 8923-2011; when the area ratio of the corroded region exceeds the preset threshold, it is judged as a corrosion defect and its type, location, and area ratio quantitative parameters are output.

[0042] Both U-Net and ResNet50 models were trained using a construction strategy of "pre-training + fine-tuning + training on a dedicated dataset", with the dataset obtained by taking photos at a construction site.

[0043] In some embodiments, step 6 specifically includes:

[0044] Step 6-1: Construct an evaluation index system that includes the severity of defects, the importance of the region, and the weights of defect development trends; calculate the comprehensive evaluation score based on the index system using a weighted summation method. And classify the defect risk level according to the score;

[0045]

[0046] in The weights for the severity of defects are as follows: severe defects have a weight of 0.6, moderate defects have a weight of 0.3, and minor defects have a weight of 0.1. The importance weights of the defect locations are as follows: boom welds 0.4, attached frame 0.3, tower standard section connection joints 0.2, and slewing bearing surfaces 0.1. As a weight for defect development trends, if the same type of defect appears in two consecutive inspections of a certain area, and the defect quantification parameter increases by ≥30%, the defect weight coefficient is increased to 1.5 times the original weight.

[0047] The single defect assessment is calculated based on the highest level corresponding to the defect quantification parameter (e.g., a crack length of 15mm is calculated as a medium defect with a weight of 0.3; a 15mm crack meets the defect judgment standard of "length > 5mm + width > 0.2mm", falling within the quantification parameter range of "medium defect" (10~30mm) → according to the preset rule of "medium defect corresponding to weight 0.3" → the final weight is determined to be 0.3). The core basis for the classification of defect levels is the proportion of the rusted area to the inspected area. This proportion threshold is determined based on the "Safety Specifications and National Standards for Tower Cranes (Special Equipment)" (GB / T 3811-2008, GB 50205-2020, GB / T 8923-2011), and must be combined with the comprehensive evaluation index system for weighted calculation before final classification.

[0048] Based on the overall score, it is divided into four levels: ≥90 is excellent, 75 ≤ <90 is qualified, 60 ≤ <75 is of concern, and <60 is dangerous.

[0049] Step 6-2: For situations where multiple types of defects, such as cracks and corrosion, exist simultaneously in the same structural area, perform risk superposition calculations;

[0050] Step 6-3: Based on the assessment results, generate rectification suggestions including rectification priorities, rectification timelines, and specific maintenance measures, and generate structural defect assessment results. Rectification priorities are ranked as "Severe Defects > Moderate Defects > Minor Defects," and within the same level, "Crane boom weld / attached frame defects > Tower standard section defects > Slewing bearing surface defects." Rectification timelines are specified as "Severe defects ≤ 24 hours, Moderate defects ≤ 7 days, Minor defects ≤ 30 days." The recommendations are tailored to specific needs—crack defects are recommended for "crack grinding + non-destructive testing review," and rust defects are recommended for "rust cleaning + anti-rust coating repair," with key track nodes requiring post-repair inspection marked.

[0051] In some embodiments, the risk superposition calculation in step 6-3 specifically involves: when multiple types of defects exist in the same area, the comprehensive risk weight of the area is calculated by multiplying the weight of the highest-level defect by a superposition coefficient greater than 1. The superposition coefficient ranges from [1, 1.5]. A superposition coefficient of 1.2 is preferred. If the same critical area (such as a section of weld on the crane boom) simultaneously has cracks and corrosion defects, the comprehensive weight of the area is calculated as "the highest-level defect weight × 1.2 (based on the collaborative risk characteristics of multiple defects in tower crane structures, industry norms and practices, the adaptability of the patent evaluation system, and the optimal value verified by experiments)". This avoids ignoring the safety risks of multiple defect superposition (e.g., if a weld simultaneously has moderate cracks and moderate corrosion, it is calculated as 0.3 × 1.2 = 0.36).

[0052] An autonomous UAV inspection system for defects on the outer surface of tower cranes includes the following modules:

[0053] Fusion Modeling Module: Used to build a fusion 3D model of the tower crane and its environment and to annotate the dynamic activity range;

[0054] Data acquisition and calibration module: used to acquire tower crane status and UAV starting position and calibrate the model;

[0055] The trajectory planning module is used to generate three-dimensional inspection trajectories based on regional weights, environmental parameters, and energy consumption constraints.

[0056] Unmanned aerial vehicle (UAV) inspection module: used to autonomously fly along a planned trajectory and collect video data that adapts to varying lighting conditions;

[0057] Defect identification module: used to perform anti-interference preprocessing on images and identify cracks and corrosion defects;

[0058] Defect assessment module: used to perform risk calculations, trend analysis, and generate rectification suggestions;

[0059] Ground terminal module: Used to display inspection progress, identification results and evaluation reports, and provide early warning prompts.

[0060] Rust identification: The ResNet50 image classification algorithm is used to perform binary classification identification of "rust / non-rust" on the preprocessed image (the model training dataset contains ≥1000 images of tower cranes with different degrees of rust). Combined with HSV color space analysis, the area ratio of the rusted area is calculated. Rust defects are judged according to the standard that "the proportion of the rusted area to the inspected area is >10%" (obtained from the safety specifications and national standards for tower cranes (special equipment) GB / T 3811-2008, GB 50205-2020, GB / T 8923-2011). Identification results output: The ground terminal records and outputs the defect type (cracks / corrosion only), defect location (3D coordinates, associated with track node coordinates), defect quantification parameters (crack length / width, corrosion area ratio), and defect level (minor / moderate / severe) (the core basis is the ratio of the corrosion area to the inspected area, and this ratio threshold is determined based on the "Safety Specifications and National Standards for Tower Cranes (Special Equipment)" (GB / T 3811-2008, GB 50205-2020, GB / T 8923-2011), and needs to be combined with the comprehensive evaluation index system for weighted calculation before final classification). After preprocessing, the crack identification accuracy is ≥94%, and the corrosion identification accuracy is ≥85%.

[0061] Beneficial effects:

[0062] Inspection safety is significantly improved: the autonomous inspection mode of drones completely avoids the risks of manual climbing of tower cranes at high altitudes; combined with dynamic obstacle marking and emergency flight path adjustment, the risk of drone collisions is effectively reduced, ensuring safety throughout the inspection process.

[0063] Defect identification is accurate and efficient: Compared with traditional manual inspection (accuracy rate of about 75%) and existing general identification solutions for drones (accuracy rate of about 75%), this invention designs a special identification algorithm for two core external surface defects, cracks and rust (cracks are identified by U-Net semantic segmentation + Canny edge detection, and rust is identified by ResNet50 classification + HSV color enhancement) and an anti-interference preprocessing process (directional noise reduction and illumination equalization), which improves the defect identification accuracy to over 94% and 85% respectively, with a false positive rate of ≤3%, completely solving the problems of missed detection and false detection in traditional identification methods.

[0064] High adaptability of flight path planning: It integrates tower crane structural characteristics, environmental conditions and endurance requirements to optimize the flight path generation logic, achieve full coverage of key areas, smooth flight path and reasonable energy consumption, and solve the pain points of traditional UAV flight path redundancy, incomplete coverage and insufficient endurance.

[0065] The assessment system is scientific and practical: It constructs a quantitative assessment system around core defects, incorporates risk superposition and defect development trend analysis, outputs accurate risk levels and targeted rectification suggestions, directly guides tower crane safety management and maintenance rectification work, and improves the implementation of assessment results.

[0066] The system is highly adaptable and versatile: the methods and system modules are highly matched, and the design of each function is tailored to the actual needs of the construction site. It can be directly applied to the inspection of external surface defects of various types of tower cranes without the need for complex adaptation and modification, and has a wide range of applications. Attached Figure Description

[0067] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0068] Figure 1 This is an overall flowchart of the present invention;

[0069] Figure 2 This is a path planning diagram for an embodiment of the present invention;

[0070] Figure 3 This is a path diagram of an embodiment of the present invention;

[0071] Figure 4 This is a diagram showing the cracks in the tower crane.

[0072] Figure 5 This is a diagram showing the corrosion of the tower crane. Detailed Implementation

[0073] This embodiment describes an autonomous UAV inspection solution for defects on the outer surface of tower cranes. The implementation process of this invention is as follows:

[0074] A QTZ80 tower crane at a construction site has been in operation for 12 months and requires autonomous inspection of its external surface defects. This solution will be used for the inspection:

[0075] like Figure 1 As shown, an autonomous UAV inspection method for defects on the outer surface of a tower crane includes the following steps:

[0076] Step 1: Construct a fusion 3D model of the tower crane and its environment. Based on the structural parameters of the tower crane and the construction environment data, a fusion 3D model is constructed. The fusion 3D model includes the key structural areas of the tower crane, the static obstacle areas, and the dynamic activity range of movable obstacles.

[0077] Step 1 specifically includes:

[0078] Step 1-1: Obtain the core structural parameters of the tower crane and the environmental data of the construction site. The environmental parameters include the coordinates of surrounding fixed obstacles and the location of the ground safety lifting and lowering point.

[0079] Steps 1-2: Construct an initial three-dimensional model containing tower crane geometry and environmental data based on the structural and environmental parameters;

[0080] Steps 1-3: Mark the cracks and critical areas prone to corrosion on the outer surface of the tower crane and the environmental safety boundary in the initial three-dimensional model;

[0081] Steps 1-4: For mobile devices at the construction site, determine their trajectory range based on their historical motion data, and mark them on the model to obtain a fused 3D model.

[0082] like Figure 2 As shown, based on the structural parameters of the tower crane and the site environment data, a 1:1 fused 3D model was constructed using BIM technology. Track points were marked in high-defect areas such as the boom and tower body, and the dynamic safety range of movable equipment was also marked to complete the basic construction of the model. Figure 2 This includes the viewpoint, which is the location where the drone is stationary;

[0083] Step 2: Status Acquisition and Model Calibration. Collect the tower crane's operating status parameters and the physical data of the UAV's starting point, and compare and calibrate them with the real-time operating status parameters of the fused 3D model; identify areas with unrectified defects based on historical inspection data, and update the trajectory planning weights for the corresponding areas;

[0084] Step 2 specifically includes:

[0085] Step 2-1: Collect the real-time operating status parameters of the tower crane and determine whether they exceed the safety threshold. If so, suspend the inspection process.

[0086] Step 2-2: Collect the precise physical coordinates of the UAV's starting point and information on surrounding obstacles, and compare and calibrate them with the corresponding data of the fused 3D model;

[0087] Steps 2-3: Query the tower crane's historical inspection records to identify key areas with unrectified defects; adjust the node density and search priority of the corresponding areas in subsequent trajectory planning based on the identification results.

[0088] The real-time status of the tower crane was collected (lifting boom angle 30°, verticality deviation compliant). The coordinates of the drone take-off and landing points were calibrated using GPS-RTK. Temporary scaffolding on site was added to the model simultaneously. Historical records were retrieved and it was found that there was unrepaired rust on the weld in the middle section of the lifting boom. The spacing between the track nodes in that area was then reduced from 0.8m to 0.5m to complete the calibration of the model and track weights.

[0089] Step 3: Generate adaptive inspection tracks. Based on the calibrated fused 3D model, and taking into account the importance of each inspection area, environmental factors, and the UAV's endurance, a path planning algorithm is used to generate 3D inspection tracks.

[0090] Step 3 specifically includes:

[0091] Step 3-1: A path search is performed using a trajectory sampling algorithm based on the calibrated fused 3D model, tower crane real-time status parameters, and preset inspection priorities. During the path search, node priority weights are assigned to different inspection areas, and an environmental adaptation factor based on real-time wind speed is introduced into the path cost function. The path cost function of the trajectory sampling algorithm... Specifically:

[0092]

[0093] in Basic cost; As an environmental adaptation factor, This refers to the real-time wind speed. The node priority weights are as follows: bolted connections have a weight of 1.5, welds have a weight of 1.2, and other areas have a weight of 1.0.

[0094]

[0095] in and These are the relevant weights, , , The path length is normalized, with the normalization benchmark being the maximum possible inspection path length of the tower crane. The turning angle is normalized, with a normalization reference of 180°. Take 1.2, Take 0.8.

[0096] Step 3-2: Smooth the generated initial path and use B-spline curve fitting for path segments with excessive turning angles to ensure continuous track curvature;

[0097] Step 3-3: Based on the UAV's endurance parameters and flight path energy consumption prediction, determine whether task segmentation is necessary and set intermediate stopping points; finally, generate an inspection flight path containing flight parameters of each node.

[0098] Based on the calibrated model, an improved trajectory sampling algorithm is used to generate 3D tracks according to inspection priorities, incorporating an environmental adaptation factor of level 2 wind speed, and assigning high priority weights to unrectified areas of the crane boom; Figure 3 The path at the junction of the mid-lift boom and the tower body, with large turning angles, is fitted with a B-spline curve to ensure smoothness, ultimately generating... Figure 3 The inspection path shown is used to determine the flight parameters for each node as a speed of 2 m / s and a dwell time of 10 seconds.

[0099] Step 4: Perform autonomous inspection. The UAV flies autonomously according to the described 3D inspection trajectory and collects video of the tower crane's outer surface at each trajectory node; the UAV then proceeds as follows... Figure 3 The aircraft flies autonomously along a path, maintaining stability via attitude sensors. Figure 2 4K video was captured at each flight path node; due to strong on-site light, the polarization filter was automatically activated, and the rusted area was clearly captured at the weld node in the middle section of the crane boom. Figure 5 (The numbers represent confidence levels), crack images were collected at the joint nodes of standard sections of the tower body. Figure 4 (The numbers represent confidence levels), and the data is transmitted back to the ground terminal in real time.

[0100] Step 5: Defect identification and processing. Keyframe extraction and image preprocessing are performed on the collected inspection video. Based on image features, cracks and corrosion defects on the outer surface of the tower crane are identified to obtain the defect type, location and quantitative parameters.

[0101] Step 5 specifically includes:

[0102] Step 5-1: Extract keyframes from the inspection video to obtain image groups covering the regions of each node;

[0103] Step 5-2: Identify the defect categories in the image group and perform image preprocessing for different defect categories;

[0104] Step 5-3: Perform defect identification for images of different defect categories, and output the defect type, location coordinates, and quantization parameters.

[0105] Step 5-2 specifically includes:

[0106] Image preprocessing for crack defect categories: Gaussian blur filtering is applied sequentially to the extracted image groups to remove high-frequency noise, Sobel operator in the x-direction is used for convolution to enhance the gradient of crack edges, and the image is converted into a black and white binary image by setting a binarization threshold.

[0107] Image preprocessing for rust defects: The extracted image groups are sequentially subjected to illumination distribution equalization based on the Retinex image enhancement algorithm and converted to the HSV color space to enhance the color difference contrast between the rusted area and the tower crane.

[0108] Step 5-3 specifically includes:

[0109] For crack defect identification: The U-Net semantic segmentation algorithm is used to segment the preprocessed image and extract the crack region to be detected; the Canny edge detection algorithm is used to capture the crack edge in the segmented region, and the Hough line detection is combined to determine the crack continuity; when the length and width of the detected crack both exceed the preset threshold, it is determined to be a crack defect and its type, location, length and width quantification parameters are output.

[0110] For the identification of rust defects: The ResNet50 image classification algorithm is used to perform binary classification identification of "rust / non-rust" on the preprocessed image; for the image classified as rust, the area ratio of the rust region is further calculated by combining HSV color space analysis; when the area ratio of the rust region exceeds the preset threshold, it is determined to be a rust defect and its type, location and area ratio quantitative parameters are output.

[0111] After extracting keyframes from the captured video, targeting Figure 4 The crack underwent preprocessing including Gaussian blurring and Sobel edge enhancement. After U-Net segmentation and Canny detection, its length was measured to be 12mm and its width 0.6mm, classifying it as a moderate crack. Figure 5 The corrosion was assessed using balanced lighting and HSV color enhancement. After ResNet50 classification and area analysis, the corrosion rate was found to be 15%, which was determined to be slight corrosion.

[0112] Step 6: Defect assessment and output. Based on the defect category, the importance of the area, and historical defect trends, a defect level assessment is performed to generate the defect assessment results.

[0113] Step 6 specifically includes:

[0114] Step 6-1: Construct an evaluation index system that includes the severity of defects, the importance of the region, and the weights of defect development trends. Calculate the comprehensive evaluation score based on the index system using a weighted summation method. And classify the defect risk level according to the score;

[0115]

[0116] in The weights for the severity of defects are as follows: severe defects have a weight of 0.6, moderate defects have a weight of 0.3, and minor defects have a weight of 0.1. The importance weights of the defect locations are as follows: boom welds 0.4, attached frame 0.3, tower standard section connection joints 0.2, and slewing bearing surfaces 0.1. As a weight for defect development trends, if the same type of defect appears in two consecutive inspections of a certain area, and the defect quantification parameter increases by ≥30%, the defect weight coefficient is increased to 1.5 times the original weight.

[0117] Based on the overall score, it is divided into four levels: ≥90 is excellent, 75 ≤ <90 is qualified, 60 ≤ <75 is of concern, and <60 is dangerous.

[0118] Step 6-2: For situations where multiple types of defects, such as cracks and corrosion, exist simultaneously in the same structural area, perform risk superposition calculations;

[0119] Step 6-3: Based on the assessment results, generate rectification suggestions that include rectification priorities, rectification deadlines, and specific maintenance measures, and generate structured defect assessment results.

[0120] The risk superposition calculation in step 6-3 is specifically as follows: when there are multiple types of defects in the same area, the comprehensive risk weight of the area is calculated by multiplying the weight of the highest-level defect by a superposition coefficient greater than 1.

[0121] The assessment score was calculated by combining the defect level, regional weight, and development trend: after weighted deduction of moderate cracks (tower body) and slight corrosion (crane boom), the score was 88 points, corresponding to the qualified level; the rectification suggestion is to deal with the tower body cracks within 7 days, repair the crane boom corrosion within 30 days, and maintain a high density of flight track nodes in the area during subsequent inspections to complete the entire process of this autonomous inspection.

[0122] This invention provides a method and system for autonomous inspection of defects on the outer surface of tower cranes using unmanned aerial vehicles (UAVs). Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for autonomous inspection of defects on the outer surface of tower cranes using unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: Step 1: Construct a fusion 3D model of the tower crane and its environment. Based on the structural parameters of the tower crane and the construction environment data, a fusion 3D model is constructed. The fusion 3D model includes the key structural areas of the tower crane, the static obstacle areas, and the dynamic activity range of movable obstacles. Step 2: Status acquisition and model calibration. Collect the tower crane's operating status parameters and the physical data of the UAV's starting point, and compare and calibrate them with the real-time operating status parameters of the fused 3D model. Based on historical inspection data, areas with unrectified defects were identified, and the track planning weights for the corresponding areas were updated. Step 3: Generate adaptive inspection tracks. Based on the calibrated fused 3D model, and taking into account the importance of each inspection area, environmental factors, and the UAV's endurance, a path planning algorithm is used to generate 3D inspection tracks. Step 4: Perform autonomous inspection. The UAV flies autonomously according to the three-dimensional inspection track and collects video of the tower crane's outer surface at each track node. Step 5: Defect identification and processing. Keyframe extraction and image preprocessing are performed on the collected inspection video. Based on image features, cracks and corrosion defects on the outer surface of the tower crane are identified to obtain the defect type, location and quantitative parameters. Step 6: Defect assessment and output. Based on the defect category, the importance of the area, and historical defect trends, a defect level assessment is performed to generate the defect assessment results.

2. The method according to claim 1, characterized in that, Step 1 specifically includes: Step 1-1: Obtain the core structural parameters of the tower crane and the environmental data of the construction site. The environmental parameters include the coordinates of surrounding fixed obstacles and the location of the ground safety lifting and lowering point. Steps 1-2: Construct an initial three-dimensional model containing tower crane geometry and environmental data based on the structural and environmental parameters; Steps 1-3: Mark the cracks and critical areas prone to corrosion on the outer surface of the tower crane and the environmental safety boundary in the initial three-dimensional model; Steps 1-4: For mobile devices at the construction site, determine their trajectory range based on their historical motion data, and mark them on the model to obtain a fused 3D model.

3. The method according to claim 1, characterized in that, Step 2 specifically includes: Step 2-1: Collect the real-time operating status parameters of the tower crane and determine whether they exceed the safety threshold. If so, suspend the inspection process. Step 2-2: Collect the precise physical coordinates of the UAV's starting point and information on surrounding obstacles, and compare and calibrate them with the corresponding data of the fused 3D model; Steps 2-3: Query the tower crane's historical inspection records to identify key areas with unrectified defects; adjust the node density and search priority of the corresponding areas in subsequent trajectory planning based on the identification results.

4. The method according to claim 1, characterized in that, Step 3 specifically includes: Step 3-1: A path search is performed using a trajectory sampling algorithm based on the calibrated fused 3D model, tower crane real-time status parameters, and preset inspection priorities. During the path search, node priority weights are assigned to different inspection areas, and an environmental adaptation factor based on real-time wind speed is introduced into the path cost function. The path cost function of the trajectory sampling algorithm... Specifically: in Basic cost; As an environmental adaptation factor, This refers to the real-time wind speed. The node priority weights are as follows: bolted connections have a weight of 1.5, welds have a weight of 1.2, and other areas have a weight of 1.

0. in and These are the relevant weights, , , The path length is normalized, with the normalization benchmark being the maximum possible inspection path length of the tower crane. The turning angle is normalized, with a normalization reference of 180°. Step 3-2: Smooth the generated initial path and use B-spline curve fitting for path segments with excessive turning angles to ensure continuous track curvature; Step 3-3: Based on the UAV's endurance parameters and flight path energy consumption prediction, determine whether task segmentation is necessary and set intermediate stopping points; finally, generate an inspection flight path containing flight parameters of each node.

5. The method according to claim 1, characterized in that, Step 5 specifically includes: Step 5-1: Extract keyframes from the inspection video to obtain image groups covering the regions of each node; Step 5-2: Identify the defect categories in the image group and perform image preprocessing for different defect categories; Step 5-3: Perform defect identification for images of different defect categories, and output the defect type, location coordinates, and quantization parameters.

6. The method according to claim 5, characterized in that, Step 5-2 specifically includes: Image preprocessing for crack defect categories: Gaussian blur filtering is applied sequentially to the extracted image groups to remove high-frequency noise, Sobel operator in the x-direction is used for convolution to enhance the gradient of crack edges, and the image is converted into a black and white binary image by setting a binarization threshold. Image preprocessing for rust defects: The extracted image groups are sequentially subjected to illumination distribution equalization based on the Retinex image enhancement algorithm and converted to the HSV color space to enhance the color difference contrast between the rusted area and the tower crane.

7. The method according to claim 5, characterized in that, Step 5-3 specifically includes: For crack defect identification: The U-Net semantic segmentation algorithm is used to segment the preprocessed image and extract the crack region to be detected; the Canny edge detection algorithm is used to capture the crack edge in the segmented region, and the Hough line detection is combined to determine the crack continuity; when the length and width of the detected crack both exceed the preset threshold, it is determined to be a crack defect and its type, location, length and width quantification parameters are output. For the identification of rust defects: The ResNet50 image classification algorithm is used to perform binary classification identification of "rust / non-rust" on the preprocessed image; for the image classified as rust, the area ratio of the rust region is further calculated by combining HSV color space analysis; when the area ratio of the rust region exceeds the preset threshold, it is determined to be a rust defect and its type, location and area ratio quantitative parameters are output.

8. The method according to claim 1, characterized in that, Step 6 specifically includes: Step 6-1: Construct an evaluation index system that includes the severity of defects, the importance of the region, and the weights of defect development trends. Calculate the comprehensive evaluation score based on the index system using a weighted summation method. And classify the defect risk level according to the score; in The weights for the severity of defects are as follows: severe defects have a weight of 0.6, moderate defects have a weight of 0.3, and minor defects have a weight of 0.

1. The importance weights of the defect locations are as follows: boom welds 0.4, attached frame 0.3, tower standard section connection joints 0.2, and slewing bearing surfaces 0.

1. As a weight for defect development trends, if the same type of defect appears in two consecutive inspections of a certain area, and the defect quantification parameter increases by ≥30%, the defect weight coefficient is increased to 1.5 times the original weight. Based on the overall score, it is divided into four levels: ≥90 is excellent, 75 ≤ <90 is qualified, 60 ≤ <75 is of concern, and <60 is dangerous. Step 6-2: For situations where multiple types of defects, such as cracks and corrosion, exist simultaneously in the same structural area, perform risk superposition calculations; Step 6-3: Based on the assessment results, generate rectification suggestions that include rectification priorities, rectification deadlines, and specific maintenance measures, and generate structured defect assessment results.

9. The method according to claim 8, characterized in that, The risk superposition calculation in step 6-3 is as follows: when there are multiple types of defects in the same area, the comprehensive risk weight of the area is calculated by multiplying the weight of the highest level defect by the superposition coefficient. The superposition coefficient ranges from [1, 1.5].

10. An unmanned aerial vehicle (UAV) autonomous inspection system for defects on the outer surface of tower cranes, characterized in that, Defect inspection using the method described in any one of claims 1-9 includes the following modules: Fusion Modeling Module: Used to build a fusion 3D model of the tower crane and its environment and to annotate the dynamic activity range; Data acquisition and calibration module: used to acquire tower crane status and UAV starting position and calibrate the model; The trajectory planning module is used to generate three-dimensional inspection trajectories based on regional weights, environmental parameters, and energy consumption constraints. Unmanned aerial vehicle (UAV) inspection module: used to autonomously fly along a planned trajectory and collect video data that adapts to varying lighting conditions; Defect identification module: used to perform anti-interference preprocessing on images and identify cracks and corrosion defects; Defect assessment module: used to perform risk calculations, trend analysis, and generate rectification suggestions; Ground terminal module: Used to display inspection progress, identification results and evaluation reports, and provide early warning prompts.

Images