Laser radar defect detection method and device for tower crane scene

Abstract

The application provides a laser radar defect detection method and device for a tower crane scene, wherein in a tower crane inspection process, a laser radar device is used to scan a tower crane to be detected to obtain inspection point cloud data of the tower crane to be detected; the inspection point cloud data is input into a component point cloud segmentation model to obtain a component category corresponding to each point in the inspection point cloud data, and the component point cloud segmentation model is obtained by training sample point cloud data of a sample tower crane and a component category label corresponding to each point in the sample point cloud data; a deformation parameter of each component in the tower crane to be detected is calculated according to an inspection point cloud set formed by points of the same component category, and a defect detection result of each component is determined according to the deformation parameter of each component. The application accurately represents the deformation of different categories of components through three-dimensional point cloud, and improves the accuracy of defect detection.

Description

Technical Field

[0001] This invention relates to the field of equipment maintenance technology, and in particular to a method and device for detecting defects in LiDAR for tower crane applications. Background Technology

[0002] In various construction sites, frequent hoisting and unloading operations of equipment and materials are often required to ensure the smooth implementation of equipment installation, component hoisting, and material handling. To accomplish these tasks, tower cranes, gantry cranes, and other specialized hoisting and unloading equipment are widely used on construction sites. Because these specialized equipment typically have large structural dimensions, high operating heights, large load capacities, and high operating frequencies, and operate in complex and ever-changing construction environments for extended periods, their key structural components are prone to wear, deformation, or displacement during use. To ensure the safe and stable operation of these specialized equipment, regular inspections and maintenance are necessary to promptly identify and eliminate potential safety hazards.

[0003] Currently, the main inspection methods for special equipment include manual inspection and drone inspection. Manual inspection is limited by factors such as high operating altitude and complex operating environments, resulting in higher safety risks and lower inspection efficiency. In contrast, drone inspection offers advantages such as operational flexibility, high safety, high efficiency, and wide coverage, and has gradually become an important technical means for the safety monitoring of special equipment.

[0004] Existing drone inspection technology is mainly based on visible light image acquisition. By analyzing the acquired image data and combining it with image recognition or target detection models, it can detect the appearance of special equipment to determine whether there are any abnormalities or defects in the equipment.

[0005] However, existing image-based drone inspection methods still have significant shortcomings in detecting structural deformation of special equipment. When critical components of equipment undergo small or gradual geometric deformation, relying solely on image features for identification is insufficient to accurately reflect the actual spatial position and geometric changes of the components, easily leading to missed detections or misjudgments. Examples include small overall tilt angles of tower cranes, slight bending deformation of the boom or counterweight, small displacement of the counterweight, and hook assembly opening increases exceeding a specified threshold.

[0006] In the above situations, changes in the appearance of equipment are not easily quantified accurately in two-dimensional images, and relying solely on image detection is insufficient to meet the needs of refined and quantitative detection of the structural safety status of special equipment. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a lidar defect detection method and device for tower crane scenarios.

[0008] This invention provides a lidar defect detection method for tower crane scenarios, comprising:

[0009] During the inspection of the tower crane to be inspected, the tower crane to be inspected is scanned using a lidar device to obtain the inspection point cloud data of the tower crane to be inspected.

[0010] The inspection point cloud data is input into the component point cloud segmentation model to obtain the component category corresponding to each point in the inspection point cloud data. The component point cloud segmentation model is trained by using sample point cloud data of a sample tower crane and the component category label corresponding to each point in the sample point cloud data.

[0011] Based on the inspection point cloud set formed by points of the same component category, the deformation parameters of each component in the tower crane to be inspected are calculated, and the defect detection results of each component are determined based on the deformation parameters of each component.

[0012] According to the present invention, a lidar defect detection method for tower crane scenarios further includes, before inputting the inspection point cloud data into the component point cloud segmentation model:

[0013] The inspection point cloud data is preprocessed, and the preprocessing includes one or more of the following: dust and noise filtering, point filling in obscured areas, multi-frame point cloud fusion, point cloud downsampling, and coordinate scale normalization.

[0014] According to the present invention, a lidar defect detection method for tower crane scenarios further includes, before inputting the inspection point cloud data into the component point cloud segmentation model:

[0015] Convert the inspection point cloud data to the equipment reference coordinate system;

[0016] The equipment reference coordinate system is established based on the structural design parameters and installation reference of the tower crane to be tested.

[0017] The equipment reference coordinate system takes the theoretical design position of the rotation center of the tower crane under test on the foundation plane as the origin, the designed vertical direction of the tower body of the tower crane under test as the Z-axis direction, the theoretical horizontal direction of the boom of the tower crane under test in the initial design installation state as the positive X-axis direction, and determines the Y-axis direction according to the right-hand coordinate system rules.

[0018] According to the present invention, a lidar defect detection method for tower crane scenarios calculates the deformation parameters of each component in the tower crane to be detected based on a component point cloud set formed by points of the same component category, including:

[0019] Collect the reference point cloud data of the tower crane to be tested, label the component category of each point in the reference point cloud data, and obtain the component segmentation box corresponding to each component category;

[0020] Based on the component segmentation boxes corresponding to each component category, the baseline component point cloud is extracted from the baseline point cloud data and the inspection component point cloud is extracted from the inspection point cloud data.

[0021] The consistency of the point cloud of the inspected component and the point cloud of the reference component is verified. If the verification is successful, the deformation parameters of each component in the tower crane to be inspected are calculated based on the point cloud set formed by points of the same component category.

[0022] According to the present invention, a lidar defect detection method for tower crane scenarios calculates the deformation parameters of each component in the tower crane to be detected based on a component point cloud set formed by points of the same component category, including:

[0023] When the component category is a rod-type component, the point cloud set corresponding to the rod-type component is fitted with the axis using the weighted least squares method to obtain the actual axis. The distance between the actual axis and the theoretical axis corresponding to the rod-type component is calculated. Based on the distance between the actual axis and the theoretical axis, the axis offset corresponding to the rod-type component is calculated as the deformation parameter.

[0024] When the component category is a beam arm component, each point in the component point cloud set corresponding to the beam arm component is projected to the normal direction of the theoretical axis corresponding to the beam arm component, and the vertical distance is calculated. The deflection corresponding to the beam arm component is calculated as the deformation parameter based on the vertical distance of each point to the theoretical axis.

[0025] When the component category is a connection and installation component, the point cloud set corresponding to the connection and installation component is subjected to orientation fitting. Based on the fitting result and the theoretical orientation vector corresponding to the connection and installation component, the angular deviation corresponding to the connection and installation component is calculated as the deformation parameter.

[0026] According to the present invention, a lidar defect detection method for tower crane scenarios determines the defect detection results of each component based on the deformation parameters of each component, including:

[0027] If the axial offset of the rod-type component is greater than the maximum permissible axial offset, it is determined that the rod-type component has a structural defect; otherwise, it is determined that the rod-type component is in a normal state.

[0028] If the deflection of the beam-arm component is greater than the maximum allowable deflection, it is determined that the beam-arm component has a structural defect; otherwise, it is determined that the beam-arm component is in a normal state.

[0029] If the angular deviation corresponding to the connection and installation component is greater than the maximum permissible angular deviation, it is determined that the connection and installation component has a structural defect; otherwise, it is determined that the connection and installation component is in a normal state.

[0030] According to the present invention, a lidar defect detection method for tower crane scenarios determines the defect detection results of each component based on the deformation parameters of each component, including:

[0031] Calculate the ratio between the deformation parameters of each component and the corresponding maximum permissible deviation;

[0032] Determine the defect level corresponding to the range of ratios in which the ratio falls, wherein the range of ratios is pre-associated with the defect level.

[0033] The present invention also provides a lidar defect detection device for tower crane scenarios, comprising:

[0034] The data acquisition module is used to scan the tower crane under inspection using a lidar device during the inspection process of the tower crane under inspection, and to acquire the inspection point cloud data of the tower crane under inspection.

[0035] The segmentation module is used to input the inspection point cloud data into the component point cloud segmentation model to obtain the component category corresponding to each point in the inspection point cloud data. The component point cloud segmentation model is trained by using sample point cloud data of a sample tower crane and the component category label corresponding to each point in the sample point cloud data.

[0036] The calculation module is used to calculate the deformation parameters of each component in the tower crane to be inspected based on the inspection point cloud set formed by points of the same component category, and to determine the defect detection results of each component based on the deformation parameters of each component.

[0037] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the lidar defect detection method for tower crane scenarios as described above.

[0038] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the lidar defect detection method for tower crane scenarios as described above.

[0039] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the lidar defect detection method for tower crane scenarios as described above.

[0040] The lidar defect detection method and device for tower crane scenarios provided by this invention achieve the following technical effects:

[0041] 1. Achieve non-contact deformation detection of special equipment components

[0042] By using drones equipped with lidar for inspection, complete three-dimensional structural information can be obtained, avoiding the safety risks of manual inspection.

[0043] 2. Supports component-level fine-grained deformation analysis

[0044] By dividing the point cloud into specific structural components and performing corresponding analysis, the deformed components and their specific spatial locations can be accurately located, facilitating subsequent maintenance and safety assessment.

[0045] 3. Deformation results are quantifiable and comparable.

[0046] By calculating geometric parameters such as point cloud displacement, axis offset, and cross-sectional changes, a quantitative description of the degree of deformation can be achieved, avoiding the subjectivity of traditional visual inspection.

[0047] 4. A universal testing method applicable to various types of special equipment

[0048] This method is not dependent on a specific equipment model and can flexibly adjust the component segmentation strategy according to structural characteristics. It is applicable to deformation detection of various special equipment such as cranes, tower cranes, and booms. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating the lidar defect detection method for tower crane scenarios provided by the present invention.

[0051] Figure 2 This is a complete flowchart of the lidar defect detection method for tower crane scenarios provided by the present invention.

[0052] Figure 3 This is a schematic diagram of the structure of the lidar defect detection device for tower crane scenarios provided by the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0054] The following is combined Figure 1 This invention describes a lidar defect detection method for tower crane scenarios, comprising:

[0055] Step 101: During the inspection of the tower crane to be inspected, the tower crane to be inspected is scanned using a lidar device to obtain the inspection point cloud data of the tower crane to be inspected.

[0056] Step 102: Input the inspection point cloud data into the component point cloud segmentation model to obtain the component category corresponding to each point in the inspection point cloud data. The component point cloud segmentation model is trained by using sample point cloud data of a sample tower crane and the component category label corresponding to each point in the sample point cloud data.

[0057] Step 103: Based on the inspection point cloud set formed by points of the same component category, calculate the deformation parameters of each component in the tower crane to be inspected, and determine the defect detection results of each component based on the deformation parameters of each component.

[0058] The method for measuring defects in tower crane components based on lidar mainly includes the following five aspects: 1. Training a point cloud component segmentation model for the tower crane. 2. Collecting lidar point cloud data of the equipment during equipment inspection. 3. Component point cloud segmentation, using the segmentation model to segment the component point cloud queue. 4. Component deformation detection, calculating the detection data and comparing it with the equipment design parameters. 5. Comparing the comparison results with a threshold; if the result exceeds the threshold, it is judged as a defect, and an abnormal warning is output.

[0059] The component point cloud segmentation model in this embodiment can use the PointNet++ model; however, this implementation example does not impose specific limitations on the component point cloud segmentation model. Figure 2 As shown, before using the component point cloud segmentation model to segment the inspection point cloud data, a hybrid training dataset is constructed using real collected point cloud data and synthetic point cloud data to train the component point cloud segmentation model, so as to improve the model's ability to identify and generalize tower crane structures under complex working conditions.

[0060] First, point cloud data samples of the tower crane were collected. Using the tower crane as the target for point cloud acquisition, the acquisition range was determined to cover the tower body, jib, counterweight boom, hook, wire rope, and related connecting components, ensuring the acquired point cloud fully reflects the overall structural form of the tower crane. A 64-line or higher LiDAR, paired with a GNSS dual-mode positioning module, was used to scan the tower crane and acquire corresponding 3D point cloud data. The acquisition attitudes covered at least 20 scenarios, including the tower crane being unloaded, fully loaded, and with the jib rotating from 0° to 360°; environmental conditions included sunny days, cloudy days, and light dust.

[0061] After completing the real point cloud acquisition, a three-dimensional structural model of the tower crane is constructed based on the real point cloud data of the tower crane or the equipment structural parameters provided by the manufacturer. The three-dimensional structural model includes at least structural components such as the tower body, boom, counterweight boom, hook, wire rope, and connecting components, and each structural component is pre-assigned a corresponding semantic category label.

[0062] Then, synthetic point cloud data is generated based on the three-dimensional structural model of the tower crane. By constructing a virtual lidar scanning environment, the laser scanning of the tower crane by a UAV under different flight altitudes, orbiting angles, and pitch angles is simulated to generate corresponding synthetic point cloud data. During the synthesis process, random perturbations are applied to the virtual scanning parameters, including changes in scanning distance, point density, random noise perturbations, and local occlusion simulation, so that the generated synthetic point cloud closely approximates the spatial distribution characteristics of the real lidar acquisition results.

[0063] The noise disturbance can be expressed as:

[0064]

[0065] in: The coordinates of the points in the original synthetic point cloud; The coordinates of the point after adding noise; With a mean of 0 and a standard deviation of Gaussian random noise.

[0066] By performing multi-scene, multi-angle virtual scanning of the 3D structural model, multiple synthetic point cloud samples of tower cranes with different structural postures and environmental characteristics can be generated to expand the scale of training data.

[0067] After generating both real and synthetic point clouds, the obtained raw point cloud data undergoes preprocessing. Preprocessing includes at least noise point removal and data format standardization to eliminate environmental interference points and ensure consistency in coordinate representation and data structure. The preprocessed tower crane point cloud data is then used as standardized output to form the raw point cloud dataset for subsequent point cloud annotation and model training. .

[0068] Then, the point cloud dataset is labeled. The original point cloud dataset... As annotation objects, structural components are distinguished among different spatial regions in the point cloud, giving the point cloud data semantic information for component identification. Based on the structural composition of the tower crane, a set of component categories for point cloud annotation can be constructed using tools such as CloudCompare or Label3D. Annotation processing is then performed on the point cloud data, establishing associations between each data point in the point cloud and its corresponding category in the component category set. Annotation processing can be completed manually, rule-based, or through a combination of manual and automated methods.

[0069] For synthetic point cloud data, since the categories of each structural component are predefined in the 3D structural model, the semantic tags of the corresponding structural components can be inherited synchronously during the process of generating point clouds by virtual laser scanning, so as to realize the automatic annotation of synthetic point cloud data and reduce the workload of manual point cloud annotation.

[0070] Subsequently, a hybrid training dataset was constructed by combining real point cloud data with completed component category annotations with automatically annotated synthetic point cloud data. The synthetic point cloud data was used to enhance the model's learning ability for different structural poses and complex working conditions, while the real point cloud data was used to enhance the model's adaptability to real LiDAR noise, occlusion, and missing point cloud features.

[0071] Preferably, in the hybrid training dataset, synthetic point cloud data accounts for 70% to 90%, and real point cloud data accounts for 10% to 30%.

[0072] Then, each frame of point cloud samples in the mixed training dataset is formatted to construct a point cloud sample set for model training. Random rotation, random scaling, random noise perturbation, and random occlusion enhancement are performed on the point cloud samples to further improve the model's generalization ability.

[0073] Finally, the tower crane point cloud data with completed component category labeling is used as training data. Each frame of point cloud sample is formatted to construct a point cloud sample set for model training.

[0074] Each point cloud sample is represented as: ,in Indicating the first point cloud The three-dimensional spatial coordinates of each point; This indicates the component category label corresponding to that point; This indicates the number of points in a single frame of the point cloud.

[0075] To reduce the size of point cloud data and extract local geometric features, a hierarchical sampling strategy in PointNet++ is adopted to process the input point cloud. The farthest point sampling (FPS) method is used to extract features from the point cloud. Select a set of key points; using the key points as the center, based on a spherical neighborhood or - Construct local point sets using the nearest neighbor method.

[0076] To construct and augment the dataset, the point cloud dataset was divided into training, validation, and test sets in a ratio of 8:1:1, and the dataset was augmented by random rotation from -45° to 45° and translation from -0.1 to 0.1m.

[0077] Training parameter settings: GPU environment, NLLLoss loss function, Adam optimizer (initial learning rate 0.001, step size 20, decay factor 0.5), batch size 8, training epoch=100, weight decay 1e-4.

[0078] The formula for the loss function:

[0079]

[0080] in, For the number of component categories, For the first The sample is predicted to contain the first... The probability of a component of a certain type, where N is the total number of samples; It is an indicator function, when the first... The first sample Labels of each sample The value is 1 when it is category c, and 0 otherwise.

[0081] Based on the loss function, the backpropagation algorithm is used to update the PointNet++ network parameters; through multiple rounds of iterative training, the model gradually converges to obtain a point cloud segmentation model that can distinguish different structural components of a tower crane.

[0082] The component point cloud segmentation model takes 3D point cloud data as input and outputs a component-level point cloud set, providing basic data support for subsequent inspection analysis and defect determination. The model supports acquiring raw point cloud data from LiDAR equipment, point cloud files, or data interfaces; it parses the data format of the input point cloud, uniformly representing the point cloud as a set of 3D coordinate points.

[0083]

[0084] Verify the integrity of the point cloud, ensuring that each point in the point cloud contains at least spatial coordinate information. The output is a structured point cloud data object, which serves as input to the point cloud preprocessing module.

[0085] The incoming point cloud data is standardized to meet the input requirements of the component segmentation model and improve segmentation stability and accuracy. Preprocessing includes: noise point removal, which removes outliers and invalid points based on the point cloud neighborhood density or height threshold; point cloud downsampling, which downsamples high-density point clouds to control the number of points; and coordinate scale normalization, which normalizes or constrains the coordinates of the point cloud.

[0086] The processed point cloud is still represented as:

[0087]

[0088] The output is generated standardized point cloud data, which serves as input to the component point cloud segmentation model.

[0089] Using a trained point cloud segmentation model, point-level component classification is performed on the input point cloud to achieve automatic segmentation of different structural components of a tower crane. Specific steps include: loading the trained PointNet++ component segmentation model; inputting the preprocessed point cloud into the model and performing point-level semantic segmentation inference; and predicting the corresponding component category label for each point in the point cloud.

[0090] The model output is represented as follows: ,in For the first Each point corresponds to a tower crane component category identifier, and the output is point cloud data containing component label information.

[0091] The point-level segmentation results are organized, classified, and output to form a component-level point cloud data structure that can be directly used for inspection analysis. Specifically, this includes: aggregating the segmentation results by component category; and regrouping points with the same component category identifier into corresponding component point cloud subsets. ,in The component category is numbered for the tower crane; the output is a set of multiple component-level point clouds, which serves as the direct input for component traversal, deformation calculation, and defect determination.

[0092] During the inspection of tower cranes, lidar equipment is used to scan the tower crane under inspection to obtain inspection point cloud data reflecting its current structural status. When collecting the inspection point cloud, the tower crane is kept stationary, and the scanning range and acquisition parameters of the lidar are set so that the point cloud covers the main structural components such as the tower body, boom, and counterweight boom.

[0093] During the data acquisition process, the lidar performs continuous or discrete scans of the tower crane according to preset scanning parameters to obtain the three-dimensional spatial coordinates of the tower crane's surface, and then uniformly represents the acquired point cloud data as a three-dimensional point set:

[0094]

[0095] in, Indicates the inspection time Tower crane inspection point cloud data, This represents the number of valid points in the inspection point cloud.

[0096] While acquiring point cloud data, basic acquisition information related to the acquisition is recorded, including acquisition time identifiers and scanning parameter descriptions, to characterize the acquisition conditions of the inspection point cloud. The acquired inspection point cloud data undergoes integrity verification, and obviously invalid point cloud points are removed to ensure that the inspection point cloud data can fully reflect the spatial structural morphology of the tower crane.

[0097] During the inspection of tower cranes, lidar equipment is used to scan the tower crane under inspection to obtain inspection point cloud data reflecting its current structural status. When collecting the inspection point cloud, the tower crane is controlled to be stationary or running at low speed, and the scanning range and acquisition parameters of the lidar are set so that the point cloud covers the main structural components such as the tower body, boom, and counterweight boom.

[0098] During the data acquisition process, the lidar performs continuous or discrete scans of the tower crane according to preset scanning parameters to obtain the three-dimensional spatial coordinates of the tower crane's surface, and then uniformly represents the acquired point cloud data as a three-dimensional point set:

[0099]

[0100] in, Indicates the inspection time Tower crane inspection point cloud data, This represents the number of valid points in the inspection point cloud.

[0101] Simultaneously with point cloud acquisition, basic acquisition information related to the point cloud acquisition is recorded, including acquisition time identifiers and scanning parameter descriptions, to characterize the acquisition conditions of the inspection point cloud. The acquired inspection point cloud data undergoes integrity verification, removing obviously invalid point cloud points to ensure that the inspection point cloud data can fully reflect the spatial structural morphology of the tower crane. After the above processing, inspection point cloud data that meets the requirements for subsequent processing is obtained. .

[0102] Inspection point cloud data Input the component point cloud segmentation model and output the component segmentation results. The inspection point clouds are grouped according to component categories, forming multiple independent component point cloud sets, and each set is assigned a corresponding component category identifier. Following a preset component processing order, the component point cloud sets are sequentially written into the component point cloud queue. The processing order is set based on the structural relationship or inspection requirements of the tower crane to ensure the stability and consistency of subsequent deformation analysis.

[0103] During queue traversal, the component point cloud set is read one by one according to the queue order. The currently read component point cloud undergoes integrity checks and noise point cloud filtering, eliminating isolated point cloud clusters with insufficient points or significantly deviating from the main component structure, thus obtaining valid component point cloud data for deformation calculation. After processing the current component point cloud, its identification information is associated with and stored with the subsequent deformation calculation results, and the process continues to read the next component point cloud in the queue until all component point clouds in the queue have been traversed.

[0104] Based on the completion of component point cloud segmentation and traversal, for tower crane components with different structural features, according to the equipment design parameters provided by the manufacturer, in the equipment reference coordinate system... The deformation of the point cloud of the inspected components is then quantitatively calculated.

[0105] Based on the calculation of deformation parameters of various components, and according to the equipment design parameters and corresponding allowable deviation ranges provided by the tower crane manufacturer, the structural condition of each component of the inspected tower crane is assessed for defects. The equipment design parameters include the theoretical geometric dimensions, spatial relationships, and installation angle requirements of the components, while the allowable deviation range characterizes the permissible structural deformation limits of the tower crane under safe operating conditions.

[0106] The allowable deviation range can be derived from the equipment manufacturer's technical documents, installation technical specifications, or relevant industry standards, and is used as the threshold basis for judging component defects.

[0107] This embodiment uses a drone equipped with a lidar for inspection, acquiring three-dimensional structural information of the tower crane under inspection, which provides more detailed information compared to two-dimensional images. By using a component point cloud segmentation model to divide the point cloud into specific structural components and perform corresponding analysis, the deformed components and their specific spatial locations can be accurately located, facilitating subsequent maintenance and safety assessment. Based on the deformation parameters of each component in the tower crane under inspection, the degree of deformation can be quantitatively described, accurately reflecting the changes in the three-dimensional structure of the equipment and improving the accuracy of defect detection.

[0108] Based on the above embodiments, this embodiment further includes the following step before inputting the inspection point cloud data into the component point cloud segmentation model:

[0109] The inspection point cloud data is preprocessed, and the preprocessing includes one or more of the following: noise point removal, point cloud downsampling, and coordinate scale normalization.

[0110] Based on the above embodiments, this embodiment further includes the following step before inputting the inspection point cloud data into the component point cloud segmentation model:

[0111] Convert the inspection point cloud data to the equipment reference coordinate system;

[0112] The device reference coordinate system Established based on the structural design parameters and installation references of the tower crane to be tested;

[0113] The equipment reference coordinate system uses the theoretically designed position of the slewing center of the tower crane under test on the foundation plane as its origin. The vertical direction of the tower body of the tower crane to be tested is taken as the positive direction of the Z-axis, the theoretical horizontal direction of the boom of the tower crane to be tested in the initial installation state is taken as the positive direction of the X-axis, and the Y-axis direction is determined according to the right-hand coordinate system rule.

[0114] To eliminate the influence of different acquisition positions and scanning postures on point cloud analysis results, and to provide a unified spatial reference for subsequent component deformation calculations, coordinate unification processing was performed on the inspection point cloud data. Coordinate unification was based on an equipment reference coordinate system established according to the equipment structural parameters provided by the tower crane manufacturer. It serves as a unified coordinate reference, rather than being based on historically collected point cloud data.

[0115] The equipment reference coordinate system is used to characterize the theoretical spatial structure of a tower crane under undeformed and defect-free conditions. The component dimensions, relative positions, and axis parameters are all derived from the equipment design parameters provided by the tower crane manufacturer, and are used as a benchmark for subsequent component deformation calculations and defect judgments.

[0116] Based on the acquisition pose relationship corresponding to the inspection point cloud, determine the direction of the inspection point cloud from its original acquisition coordinate system to the device reference coordinate system. Rigid body transformation relationships, including rotation matrices. With translation vector .

[0117] For any point in the inspection point cloud Spatial transformation is performed according to the following coordinate transformation relationship:

[0118]

[0119] This leads to a unified device reference coordinate system. The following is the inspection point cloud data:

[0120]

[0121] in, For inspection time The collected raw inspection point cloud data; To complete the inspection point cloud data after coordinate unification; It is a three-dimensional rotation matrix; It is a three-dimensional translation vector.

[0122] Through the above processing, the inspection point cloud data is made consistent with the equipment reference coordinate system established based on the manufacturer's equipment parameters in terms of spatial position and attitude, providing a unified and stable spatial benchmark for subsequent component segmentation, deformation calculation and defect judgment based on theoretical structural parameters.

[0123] Based on the above embodiments, this embodiment calculates the deformation parameters of each component in the tower crane to be detected based on the component point cloud set formed by points of the same component category, including:

[0124] Collect the reference point cloud data of the tower crane to be tested, label the component category of each point in the reference point cloud data, and obtain the component segmentation box corresponding to each component category;

[0125] Based on the component segmentation boxes corresponding to each component category, the baseline component point cloud is extracted from the baseline point cloud data and the inspection component point cloud is extracted from the inspection point cloud data.

[0126] The consistency of the point cloud of the inspected component and the point cloud of the reference component is verified. If the verification is successful, the deformation parameters of each component in the tower crane to be inspected are calculated based on the point cloud set formed by points of the same component category.

[0127] The baseline point cloud data is the overall point cloud data of the tower crane to be tested. The overall point cloud data is labeled with component categories, and boxes containing points of the same component category are used as component segmentation boxes. .

[0128] In the device reference coordinate system Unify point cloud data and segmentation box parameters to ensure the baseline point cloud Inspection Point Cloud The component segmentation boxes are calculated on the same spatial coordinate reference, providing a consistent data basis for component-level point cloud clipping.

[0129] Component segmentation rule system construction: Component segmentation boxes are used as spatial constraints. Each component segmentation box limits the effective range of the corresponding structural component in three-dimensional space, retaining only the point cloud data within that spatial range and excluding interference from other structural or environmental point clouds. The spatial constraint expression for each component partition box is:

[0130]

[0131] In the device reference coordinate system Next, spatial clipping operations are performed on the baseline point cloud data and the inspection point cloud data respectively. Based on component segmentation boxes. Extract the reference component point cloud of the corresponding component from the reference point cloud:

[0132]

[0133] Based on the same component segmentation box Extract the inspection component point cloud of the corresponding component from the inspection point cloud:

[0134]

[0135] and This refers to the point cloud data of the same structural component under both baseline and inspection states, maintaining a one-to-one correspondence in spatial definition and component numbering. For example, if... and If the component categories at the midpoint are the same or mostly the same, it means that the two are consistent.

[0136] Consistency verification is performed on the point cloud data of each component obtained by segmentation to confirm that the point cloud data of the same component in the baseline state and the inspection state comes from the same component segmentation box, so as to avoid component mismatch caused by overall equipment displacement or attitude change.

[0137] At the same time, check the spatial coverage integrity of the component point cloud. When a component point cloud is found to have obvious missing, occluded, or abnormally sparse conditions, record the corresponding component number and mark it in subsequent analysis or trigger a supplementary acquisition process to ensure the reliability of the component deformation calculation results.

[0138] Component point cloud data organization and storage: The segmented component point cloud data is stored in a structured manner according to the component number, and a correspondence is established between "component number - baseline component point cloud - inspection component point cloud".

[0139] Based on the above embodiments, this embodiment calculates the deformation parameters of each component in the tower crane to be detected based on the component point cloud set formed by points of the same component category, including:

[0140] When the component category is a rod-type component, the point cloud set corresponding to the rod-type component is fitted with the axis using the weighted least squares method to obtain the actual axis. The distance between the actual axis and the theoretical axis corresponding to the rod-type component is calculated. Based on the distance between the actual axis and the theoretical axis, the axis offset corresponding to the rod-type component is calculated as the deformation parameter.

[0141] When the component category is a beam arm component, each point in the component point cloud set corresponding to the beam arm component is projected to the normal direction of the theoretical axis corresponding to the beam arm component, and the vertical distance is calculated. The deflection corresponding to the beam arm component is calculated as the deformation parameter based on the vertical distance of each point to the theoretical axis.

[0142] When the component category is a connection and installation component, the point cloud set corresponding to the connection and installation component is subjected to orientation fitting. Based on the fitting result and the theoretical orientation vector corresponding to the connection and installation component, the angular deviation corresponding to the connection and installation component is calculated as the deformation parameter.

[0143] To improve the robustness and engineering adaptability of deformation assessment, an adaptive weighting mechanism is introduced into the deformation calculation process, assigning different weights to the points and structural regions involved in the deformation calculation, thereby achieving multi-factor fusion assessment.

[0144] For any component Its inspection points are cloud-based. Build weight coefficients for each point Defined as:

[0145]

[0146] in, Point cloud quality factor Spatial location factor Structural sensitivity factor : Total number of points in the current component point cloud Point cloud quality factor weights Spatial location factor weights : Weights of structural sensitivity factors.

[0147] satisfy: For example, rod-type components can be set to: α=0.5, β=0.2, γ=0.3; beam-arm-type components can be set to: α=0.3, β=0.2, γ=0.5; and connection and installation components can be set to: α=0.3, β=0.5, γ=0.2. Point cloud quality factor Used to reflect the reliability of point cloud data, it is defined as:

[0148]

[0149] in, Laser reflection intensity Local point cloud density, Noise level : Adjustment coefficient.

[0150] satisfy: , such as λ1=0.4, λ2=0.4, λ3=0.2.

[0151] Spatial location factor Used to highlight the importance of key areas, defined as:

[0152]

[0153] in, : No. Coordinates of a point cloud point, : Coordinates of the center point of the critical structural region.

[0154] Structural sensitivity factor Defined as:

[0155]

[0156] in, Distance from the point to the support end Total beam length Sensitivity enhancement coefficient.

[0157] For weighting coefficients Normalization is performed:

[0158]

[0159] The calculation of axial offset deformation for rod-like components is applicable to rod-like components such as standard tower sections, diagonal braces, and tie rods, where the axis is the primary structural feature. The actual axis is fitted based on the point cloud of inspected rods and compared with the theoretical axis given in the manufacturer's equipment parameters to calculate the axial offset as a deformation index.

[0160] The specific calculation process includes: fitting the point cloud to its axis using the weighted least squares method.

[0161]

[0162] in, For the first The weight of each point For the first The perpendicular distance from each point to the theoretical axis is calculated. Through the above fitting process, the actual axis parameters are obtained. .

[0163] According to the theoretical axis provided by the manufacturer Calculate the distance from each point to the theoretical axis:

[0164]

[0165] in, For the first Point cloud coordinates, For the first The theoretical axis direction vector of each member. Let be any point on the theoretical axis.

[0166] The weighted axis offset of the rod-type component is defined as:

[0167]

[0168] in, For the first The weight of each point For the first The perpendicular distance from each point to the theoretical axis.

[0169] The deflection deformation calculation for beam-arm components is applicable to beam-arm components such as crane booms and counterweight booms, which are primarily characterized by bending deflection along their length. The inspection point cloud is segmented and projected along the theoretical axis of the beam-arm, and the maximum offset of the actual point cloud relative to the theoretical straight line is calculated as an index of beam-arm deflection deformation.

[0170] The specific calculation process includes: obtaining the theoretical axis based on the design parameters. Project the point cloud onto the normal direction of the theoretical axis and calculate the vertical distance:

[0171]

[0172] in, The first in the cloud-dotting of the beam arm One point, Let the direction vector of the theoretical axis of the beam arm be denoted as . Let be a point on the theoretical axis.

[0173] The weighted deflection index for beam-arm type components is defined as follows:

[0174]

[0175] in, For the first The weight of each point This represents the perpendicular distance from the point to the theoretical axis. Weight A larger value is taken in the mid-span region of the beam, and a smaller value is taken at the support end, in order to reflect the deflection distribution characteristics of the beam structure.

[0176] The angular deviation deformation calculation for connecting and installing components is applicable to hooks, tower and boom connection nodes, slewing bearing installation locations, and diagonal bracing connection structures. Based on the point cloud of the connecting component, its actual installation direction is fitted and compared with the theoretical installation angle specified in the manufacturer's equipment parameters. The angular deviation is then calculated as a deformation index.

[0177] The specific calculation process includes: weighted orientation fitting of the point cloud of the connecting components.

[0178]

[0179] in, Let be the local normal vector or direction vector of the i-th point. For the first The weight of each point.

[0180] According to the theoretical direction vector Calculate the angle deviation:

[0181]

[0182] in, For the angular deviation deformation of connecting components, This is the actual installation direction vector. The theoretical installation direction vector.

[0183] Based on the above embodiments, this embodiment determines the defect detection results of each component according to the deformation parameters of each component, including:

[0184] If the axial offset of the rod-type component is greater than the maximum permissible axial offset, it is determined that the rod-type component has a structural defect; otherwise, it is determined that the rod-type component is in a normal state.

[0185] If the deflection of the beam-arm component is greater than the maximum allowable deflection, it is determined that the beam-arm component has a structural defect; otherwise, it is determined that the beam-arm component is in a normal state.

[0186] If the angular deviation corresponding to the connection and installation component is greater than the maximum permissible angular deviation, it is determined that the connection and installation component has a structural defect; otherwise, it is determined that the connection and installation component is in a normal state.

[0187] For tower body standard sections, diagonal braces, and other rod-like components, the calculated axial offset deformation will be... and the corresponding maximum permissible axis offset Compare them.

[0188] Among them, the maximum permissible axis offset It can be determined according to the manufacturer's design parameters. For example, for a standard tower section, the allowable axial offset can be set to no more than 1 / 1000 to 1 / 1500 of the height of the standard section; or limited to an absolute offset of no more than 5 mm to 10 mm, with the specific value determined according to the tower crane model.

[0189] When satisfied If the condition is not met, the component is determined to have a structural defect; otherwise, it is considered to be in a normal state.

[0190] For beam-arm components such as crane booms and counterweight booms, the calculated maximum deflection deformation will be used as a basis for determining the maximum deflection deformation. and the corresponding maximum permissible deflection value Compare them.

[0191] Wherein, the maximum permissible deflection value It can be set according to the manufacturer's design parameters or industry standards. For example, the allowable deflection of the boom under no-load or rated load conditions can be set to 1 / 500 to 1 / 800 of the boom length; for a boom with a length of 50m, the allowable deflection range can be 60 mm to 100 mm.

[0192] When satisfied If the beam arm component is found to have a structural defect, it is determined to be in a normal state; otherwise, it is determined to be in a normal state.

[0193] For connection and installation components such as hooks, tower body and boom connection nodes, and slewing bearing installation locations, the calculated angular deviation deformation will be used. and the corresponding maximum permissible angle deviation Compare them.

[0194] Among them, the maximum permissible angle deviation The installation requirements can be determined based on the manufacturer's technical specifications. For example, the verticality deviation between the slewing support and the central axis of the tower should not exceed 0.1° to 0.3°; the angle deviation between the diagonal brace and the main structure should not exceed 0.5°.

[0195] When satisfied If the condition is not met, the connection and installation component is determined to have a structural defect; otherwise, it is considered to be in a normal state.

[0196] Based on the above embodiments, this embodiment determines the defect detection results of each component according to the deformation parameters of each component, including:

[0197] Calculate the ratio between the deformation parameters of each component and the corresponding maximum permissible deviation;

[0198] Determine the defect level corresponding to the range of ratios in which the ratio falls, wherein the range of ratios is pre-associated with the defect level.

[0199] After completing the component deformation calculation, the calculated component deformation parameters are compared with the corresponding design allowable deviations. Based on the degree of deformation, the defects are classified into different levels to characterize the structural safety status of the tower crane components.

[0200] For any component Define defect level indicators:

[0201]

[0202] in, These are the deformation parameters of the component under its current inspection status. This refers to the maximum permissible deviation of the component as determined by design parameters or relevant specifications.

[0203] Table 1 shows an example of defect level classification. The defect level determination results for each component are summarized, and a detection result set containing the following information is output: component number, component type, and deformation parameter value. Allowable deviation threshold And defect level. The test results can be used for tower crane structural safety assessment, operational status management, and maintenance decision support.

[0204] Table 1 Example of Defect Level Classification

[0205]

[0206] The following describes the LiDAR defect detection device for tower crane scenarios provided by the present invention. The LiDAR defect detection device for tower crane scenarios described below and the LiDAR defect detection method for tower crane scenarios described above can be referred to and correspond to each other.

[0207] like Figure 3 As shown, the device includes an acquisition module 301, a segmentation module 302, and a calculation module 303, wherein:

[0208] The acquisition module 301 is used to scan the tower crane under inspection using a lidar device during the inspection process of the tower crane under inspection, and to acquire the inspection point cloud data of the tower crane under inspection.

[0209] The segmentation module 302 is used to input the inspection point cloud data into the component point cloud segmentation model to obtain the component category corresponding to each point in the inspection point cloud data. The component point cloud segmentation model is trained by using sample point cloud data of a sample tower crane and the component category label corresponding to each point in the sample point cloud data.

[0210] The calculation module 303 is used to calculate the deformation parameters of each component in the tower crane to be inspected based on the inspection point cloud set formed by points of the same component category, and to determine the defect detection results of each component based on the deformation parameters of each component.

[0211] This embodiment uses a drone equipped with a lidar for inspection, acquiring three-dimensional structural information of the tower crane under inspection, which provides more detailed information compared to two-dimensional images. By using a component point cloud segmentation model to divide the point cloud into specific structural components and perform corresponding analysis, the deformed components and their specific spatial locations can be accurately located, facilitating subsequent maintenance and safety assessment. Based on the deformation parameters of each component in the tower crane under inspection, the degree of deformation can be quantitatively described, accurately reflecting the changes in the three-dimensional structure of the equipment and improving the accuracy of defect detection.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A lidar defect detection method for tower crane scenarios, characterized in that, include: During the inspection of the tower crane to be inspected, the tower crane to be inspected is scanned using a lidar device to obtain the inspection point cloud data of the tower crane to be inspected. The inspection point cloud data is input into the component point cloud segmentation model to obtain the component category corresponding to each point in the inspection point cloud data. The component point cloud segmentation model is trained by using sample point cloud data of a sample tower crane and the component category label corresponding to each point in the sample point cloud data. Based on the inspection point cloud set formed by points of the same component category, the deformation parameters of each component in the tower crane to be inspected are calculated, and the defect detection results of each component are determined based on the deformation parameters of each component.

2. The lidar defect detection method for tower crane scenarios according to claim 1, characterized in that, Before inputting the inspection point cloud data into the component point cloud segmentation model, the following steps are also included: The inspection point cloud data is preprocessed, and the preprocessing includes one or more of the following: dust and noise filtering, point filling in obscured areas, multi-frame point cloud fusion, point cloud downsampling, and coordinate scale normalization.

3. The lidar defect detection method for tower crane scenarios according to claim 1, characterized in that, Before inputting the inspection point cloud data into the component point cloud segmentation model, the following steps are also included: Convert the inspection point cloud data to the equipment reference coordinate system; The equipment reference coordinate system is established based on the structural design parameters and installation reference of the tower crane to be tested. The equipment reference coordinate system takes the theoretical design position of the rotation center of the tower crane under test on the foundation plane as the origin, the designed vertical direction of the tower body of the tower crane under test as the Z-axis direction, the theoretical horizontal direction of the boom of the tower crane under test in the initial design installation state as the positive X-axis direction, and determines the Y-axis direction according to the right-hand coordinate system rules.

4. The lidar defect detection method for tower crane scenarios according to claim 1, characterized in that, Based on the component point cloud set formed by points of the same component category, the deformation parameters of each component in the tower crane to be detected are calculated, including: The reference point cloud data of the tower crane to be tested is collected, and the component category of each point in the reference point cloud data is labeled to obtain the component segmentation box corresponding to each component category. Based on the component segmentation boxes corresponding to each component category, the baseline component point cloud is extracted from the baseline point cloud data and the inspection component point cloud is extracted from the inspection point cloud data. The consistency of the point cloud of the inspected component and the point cloud of the reference component is verified. If the verification is successful, the deformation parameters of each component in the tower crane to be inspected are calculated based on the point cloud set formed by points of the same component category.

5. The lidar defect detection method for tower crane scenarios according to claim 1, characterized in that, Based on the component point cloud set formed by points of the same component category, the deformation parameters of each component in the tower crane to be detected are calculated, including: When the component category is a rod-type component, the point cloud set corresponding to the rod-type component is fitted with the axis using the weighted least squares method to obtain the actual axis. The distance between the actual axis and the theoretical axis corresponding to the rod-type component is calculated. Based on the distance between the actual axis and the theoretical axis, the axis offset corresponding to the rod-type component is calculated as the deformation parameter. When the component category is a beam arm component, each point in the component point cloud set corresponding to the beam arm component is projected to the normal direction of the theoretical axis corresponding to the beam arm component, and the vertical distance is calculated. The deflection corresponding to the beam arm component is calculated as the deformation parameter based on the vertical distance of each point to the theoretical axis. When the component category is a connection and installation component, the point cloud set corresponding to the connection and installation component is subjected to orientation fitting. Based on the fitting result and the theoretical orientation vector corresponding to the connection and installation component, the angular deviation corresponding to the connection and installation component is calculated as the deformation parameter.

6. The lidar defect detection method for tower crane scenarios according to claim 5, characterized in that, The defect detection results for each component are determined based on its deformation parameters, including: If the axial offset of the rod-type component is greater than the maximum permissible axial offset, it is determined that the rod-type component has a structural defect; otherwise, it is determined that the rod-type component is in a normal state. If the deflection of the beam-arm component is greater than the maximum allowable deflection, it is determined that the beam-arm component has a structural defect; otherwise, it is determined that the beam-arm component is in a normal state. If the angular deviation corresponding to the connection and installation component is greater than the maximum permissible angular deviation, it is determined that the connection and installation component has a structural defect; otherwise, it is determined that the connection and installation component is in a normal state.

7. The lidar defect detection method for tower crane scenarios according to claim 1, characterized in that, The defect detection results for each component are determined based on its deformation parameters, including: Calculate the ratio between the deformation parameters of each component and the corresponding maximum permissible deviation; Determine the defect level corresponding to the range of ratios in which the ratio falls, wherein the range of ratios is pre-associated with the defect level.

8. A lidar defect detection device for tower crane applications, characterized in that, include: The data acquisition module is used to scan the tower crane under inspection using a lidar device during the inspection process of the tower crane under inspection, and to acquire the inspection point cloud data of the tower crane under inspection. The segmentation module is used to input the inspection point cloud data into the component point cloud segmentation model to obtain the component category corresponding to each point in the inspection point cloud data. The component point cloud segmentation model is trained by using sample point cloud data of a sample tower crane and the component category label corresponding to each point in the sample point cloud data. The calculation module is used to calculate the deformation parameters of each component in the tower crane to be inspected based on the inspection point cloud set formed by points of the same component category, and to determine the defect detection results of each component based on the deformation parameters of each component.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the lidar defect detection method for tower crane scenarios as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the lidar defect detection method for tower crane scenarios as described in any one of claims 1 to 7.

Images