Robot laser vision three-dimensional scanning measurement system for large-scale structural member

Abstract

The invention relates to the technical field of large-scale structure measurement, and discloses a robot laser vision three-dimensional scanning measurement system for a large-scale structural member. The system comprises a robot motion platform, a laser vision sensing module, an ambient light sensing module, a point cloud processing module and a motion control module, wherein the laser vision sensing module is arranged at the tail end of the robot motion platform. The ambient light sensing module collects ambient light intensity distribution data in real time, and the motion control module combines the data with three-dimensional model data of the large-scale structural member to be measured to generate a multi-view dynamic scanning path instruction. The robot motion platform executes an instruction to drive the laser vision sensing module to move, the module projects laser stripes to the surface of a to-be-measured structural member and synchronously collects an image, the point cloud processing module performs adaptive filtering processing on the image and then generates three-dimensional point cloud data, and the requirement for efficient and accurate measurement of a large structural member in a complex industrial environment can be met.

Description

Technical Field

[0001] This invention relates to the field of large structure measurement technology, specifically to a robotic laser vision 3D scanning measurement system for large structural components. Background Technology

[0002] In aerospace, shipbuilding, and heavy machinery industries, the dimensional accuracy and morphological integrity of large structural components directly affect the overall performance and safe operation of the product. Precise measurement is a crucial requirement in manufacturing and quality inspection. Currently, measurement methods for large structural components are mainly divided into two categories: contact measurement and non-contact measurement. Contact measurement, represented by coordinate measuring machines (CMMs), while capable of high-precision measurements, has significant limitations in practical applications: its measurement range is restricted by the machine's structure. For large structural components with lengths reaching tens of meters, multiple relocations and data stitching are often required, which is not only cumbersome and time-consuming but also prone to affecting the final measurement accuracy due to stitching errors. Among non-contact measurement technologies, laser scanning measurement has gradually become an important choice for measuring large structural components due to its advantages such as high measurement speed, high accuracy, and non-contact and non-destructive operation. However, existing laser scanning measurement systems still have many shortcomings when facing complex industrial environments. Changes in ambient light significantly affect the quality of laser stripe image acquisition. In industrial settings, fluctuations in natural light intensity, the switching on or off of lighting equipment in the workshop, and changes in its position can all lead to a decrease in the contrast between the laser stripe and the background, or even the laser stripe being submerged in stray light, thus affecting the accuracy of subsequent image processing and point cloud generation. Most current systems lack real-time perception and dynamic adaptation capabilities to ambient light, and can usually only operate under preset fixed lighting conditions. Once the ambient light exceeds the preset range, the measurement accuracy will be greatly reduced, making it difficult to meet the stable measurement requirements in complex industrial environments. Existing systems often rely on fixed 3D models of the structural components under test for scanning path planning, employing preset fixed-viewpoint scanning methods. This approach is prone to blind spots when dealing with large structural components with complex surface morphology and numerous occlusion areas, resulting in some areas being ineffectively scanned. This necessitates repeated manual adjustments to the scanning viewpoint and path, increasing operational complexity and potentially introducing additional errors due to the uncertainty of human intervention. Furthermore, fixed scanning paths cannot dynamically adjust to environmental changes during the actual measurement process (such as decreased measurement quality in some areas due to variations in ambient light intensity), making it difficult to guarantee the consistency and reliability of data acquisition across all areas throughout the measurement process. In addition, in the point cloud processing stage, existing systems often use fixed parameter settings for filtering algorithms, failing to adaptively adjust based on the quality of laser stripe images acquired under different ambient light conditions. For images severely affected by stray light interference, the filtering effect is poor, easily retaining a large number of noise points, impacting the accuracy of the 3D point cloud data, and consequently adversely affecting subsequent applications such as structural component morphology analysis and dimensional detection. Summary of the Invention

[0003] The purpose of this invention is to provide a robotic laser vision 3D scanning and measurement system for large structural components, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a robotic laser vision 3D scanning and measurement system for large structural components, the system comprising: The robot motion platform includes a laser vision sensing module, an ambient light sensing module, a point cloud processing module, and a motion control module installed at the end of the robot motion platform. The ambient light sensing module collects ambient light intensity distribution data in real time; the motion control module generates multi-view dynamic scanning path instructions based on the ambient light intensity distribution data and the 3D model data of the large structural component under test; the robot motion platform executes the multi-view dynamic scanning path instructions to drive the laser vision sensing module to move; the laser vision sensing module projects laser stripes onto the surface of the large structural component under test and simultaneously collects laser stripe images; the point cloud processing module performs adaptive filtering on the laser stripe images and generates 3D point cloud data.

[0005] Preferably, the laser vision sensing module includes a laser power adjustment unit; the laser power adjustment unit receives real-time ambient light intensity distribution data sent by the ambient light sensing module, and dynamically adjusts the laser emission power of the laser vision sensing module according to a preset laser intensity threshold; the preset laser intensity threshold is greater than the maximum value of the current ambient light intensity distribution data and less than half of the maximum emission power of the laser vision sensing module.

[0006] Preferably, the motion control module includes a surface curvature analysis unit and a path planning unit; the surface curvature analysis unit calculates the surface curvature distribution based on the three-dimensional model data of the large structural component to be tested; the path planning unit generates the multi-view dynamic scanning path instruction according to the surface curvature distribution, the multi-view dynamic scanning path instruction includes scanning speed and scanning view interval; the scanning speed is inversely proportional to the surface curvature distribution, and the scanning view interval is directly proportional to the surface curvature distribution.

[0007] Preferably, the point cloud processing module includes a feature point extraction unit and a registration optimization unit; the feature point extraction unit performs multi-scale feature point detection on the generated 3D point cloud data and generates feature point descriptors; the registration optimization unit constructs a point cloud registration cost function based on the feature point descriptors and uses a multi-objective optimization algorithm to solve for the optimal transformation matrix of the point cloud registration cost function; the point cloud registration cost function includes feature point matching degree constraints and point cloud overlap degree constraints.

[0008] Preferably, the adaptive filtering process includes: the point cloud processing module constructing a spatial filter based on the ambient light intensity distribution data, using the spatial filter to suppress background noise in the laser stripe image, and extracting the center line of the laser stripe through grayscale morphological operations.

[0009] Preferably, the system further includes a multi-station synchronization module; the multi-station synchronization module receives real-time pose data sent by the robot motion platform, and triggers the scanning action of the laser vision sensing module according to the real-time pose data; the triggering time of the scanning action is synchronized with the timestamp of the robot motion platform reaching the preset scanning station in the multi-view dynamic scanning path instruction.

[0010] Preferably, the registration optimization unit stores the optimal transformation matrix and sends the optimal transformation matrix to the motion control module; the motion control module corrects the scanning station coordinates in the subsequent multi-view dynamic scanning path command according to the optimal transformation matrix.

[0011] Preferably, the system includes a closed-loop verification module; the closed-loop verification module receives the three-dimensional point cloud data generated by the point cloud processing module, calculates the point cloud overlap region error between adjacent scanning stations, and sends the point cloud overlap region error to the registration optimization unit; the registration optimization unit updates the weight coefficients of the point cloud registration cost function according to the point cloud overlap region error.

[0012] Preferably, the multi-objective optimization algorithm is a particle swarm optimization algorithm, wherein the particle position vector of the particle swarm optimization algorithm includes rotation and translation parameters, and the fitness function of the particle swarm optimization algorithm is the point cloud registration cost function.

[0013] Preferably, the motion control module includes a path replanning unit; when the error in the overlapping area of ​​the point cloud exceeds a preset error threshold, the path replanning unit regenerates a multi-view dynamic scanning path instruction covering the overlapping area of ​​the point cloud based on the updated point cloud registration cost function.

[0014] Compared with the prior art, the beneficial effects of the present invention are: The ambient light sensing module enables real-time acquisition of light intensity distribution data in the measurement environment, breaking the dependence of existing systems on fixed lighting conditions. This module dynamically captures changes in ambient light and transmits the relevant data to the motion control module in real time. This provides a basis for dynamic adjustment of the subsequent scanning path and optimization of the laser vision sensing module's operating parameters. It allows the system to maintain effective control over the quality of laser stripe image acquisition under complex ambient light conditions such as fluctuations in natural light and changes in workshop lighting, avoiding image contrast degradation or stripe obliteration caused by ambient light interference, and ensuring the stability and reliability of the image acquisition process. The motion control module generates multi-view dynamic scanning path instructions based on ambient light intensity distribution data and the 3D model data of the large structural component under test. Compared with existing fixed-path scanning methods, this offers greater flexibility and adaptability. Based on the 3D model data and considering ambient light distribution, the module automatically plans a scanning path that avoids areas with high interference from stray light and focuses on covering low-interference areas. Simultaneously, it dynamically adjusts the scanning angle for occlusion areas and complex shapes on the structural component's surface, ensuring a blind-spot-free scanning process. This eliminates the need for repeated manual path adjustments, significantly reducing operational procedures and time costs. Furthermore, when ambient light intensity changes abruptly, the motion control module responds quickly, adjusting the scanning path in real time to ensure high-quality image data is obtained for each measurement area even during ambient light fluctuations, further enhancing the system's measurement stability in complex industrial environments. The robot motion platform executes multi-view dynamic scanning path commands, driving the laser vision sensing module to move precisely, ensuring that the module can scan large structural components according to the planned optimal path and perspective. This platform possesses high-precision motion control capabilities, enabling precise adjustment of the laser vision sensing module's position and orientation. This allows the laser stripes to be accurately projected onto the target area on the surface of the structural component under test, while simultaneously ensuring the stability of the module's position during image acquisition. This avoids measurement errors caused by motion deviations, providing a reliable guarantee for acquiring high-quality laser stripe images. The laser vision sensing module simultaneously completes laser stripe projection and image acquisition, and its operation is closely linked with the ambient light sensing module and motion control module. Guided by the dynamic path planned by the motion control module, this module can project clear laser stripes onto the surface of the structural component from the optimal viewing angle and simultaneously acquire laser stripe images, ensuring that the acquired images can accurately reflect the morphological characteristics of the structural component's surface and provide high-quality raw data for subsequent point cloud processing. The point cloud processing module performs adaptive filtering on laser stripe images. Unlike existing fixed-parameter filtering algorithms, this module can automatically adjust the filtering parameters and algorithm based on the ambient light intensity distribution data collected by the ambient light sensing module and the actual quality of the laser stripe image (such as contrast and the degree of stray light interference). This allows for targeted filtering of images of different quality. For images with less stray light interference and higher quality, the filtering process can remove a small amount of noise while preserving the details of the laser stripes. For images with severe stray light interference and lower quality, the filtering process can effectively suppress stray light noise, enhance the contrast between the laser stripes and the background, accurately extract laser stripe features, and generate high-precision, low-noise 3D point cloud data. This accurately restores the surface morphology and dimensional information of the large structural component under test, meeting the needs of various application scenarios such as subsequent structural component quality inspection and morphological analysis. Attached Figure Description

[0015] Figure 1 This is a timing diagram of the robot laser vision three-dimensional scanning and measurement system for large structural components described in this invention. Figure 2 A flowchart for feature extraction and registration optimization in the point cloud processing module; Figure 3 A flowchart for updating the point cloud registration cost function weights for the closed-loop verification module. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 The present invention provides a robot laser vision three-dimensional scanning and measurement system for large structural components. The system includes: a robot motion platform, a laser vision sensing module, an ambient light sensing module, a point cloud processing module, and a motion control module.

[0018] The robot motion platform utilizes a six-axis industrial robot, with a laser vision sensing module mounted on its end effector via a flange. The ambient light perception module employs a high dynamic range light intensity sensor, deployed at multiple locations on the measurement site to collect ambient light intensity distribution data. The motion control module, implemented on an industrial computer, executes path planning algorithms and generates control commands. The point cloud processing module, equipped with a GPU computing unit, processes image and point cloud data in real time. During implementation, the ambient light perception module continuously collects ambient light intensity distribution data and transmits it to the motion control module. Based on the ambient light intensity distribution data and a pre-input 3D CAD model of the large structural component under test, the motion control module generates multi-view dynamic scanning path commands. These commands are sent to the robot controller via Ethernet, driving the robot motion platform to move the laser vision sensing module along the planned path. During movement, the laser vision sensing module projects laser stripes onto the workpiece surface and simultaneously acquires laser stripe images. The image data is transmitted to the point cloud processing module via gigabit network. The point cloud processing module uses an adaptive filtering algorithm to process the laser stripe images, extracts the laser stripe centerline, and calculates the 3D point cloud data, ultimately completing the full-size 3D reconstruction of the large structural component.

[0019] Example 1: The laser power adjustment unit of the laser vision sensing module continuously receives real-time ambient light intensity distribution data from the ambient light sensing module through a dedicated data interface. This data is transmitted in the form of a two-dimensional array, with each array element corresponding to the illumination intensity value of a sampling point. The sampling points are distributed according to a spatial grid to cover the entire measurement area. The data processing chip inside the laser power adjustment unit traverses the entire array and finds the maximum value. This maximum value represents the strongest background light interference level in the current environment. The chip then calls the preset laser intensity threshold stored in the non-volatile memory. This threshold is set according to the laser calibration parameters during system initialization. Its value must be strictly greater than the maximum ambient light intensity value and less than half of the maximum output power of the laser. This setting ensures that the laser stripe has sufficient contrast under any ambient light conditions and avoids image overexposure or laser lifespan loss caused by excessive power.

[0020] The data processing chip generates a voltage signal proportional to the required power through a digital-to-analog conversion circuit. This analog voltage signal is sent to the laser's drive circuit. The drive circuit precisely adjusts the injection current of the laser diode according to the voltage value, thereby controlling the magnitude of the laser output power. The entire adjustment process is dynamic and continuous. The ambient light sensing module updates the light intensity data at a frequency of ten times per second, and the laser power adjustment unit adjusts the output in real time to ensure that the laser stripes are always clearly visible in situations where the ambient light changes rapidly. For example, when a moving light source passes by, the ambient light intensity will suddenly increase, and the adjustment unit can increase the laser power within milliseconds to maintain stable image quality. The surface curvature analysis unit of the motion control module reads the 3D CAD model file of the large structural component to be tested from the system database. This file usually contains complete surface geometry information. The analysis unit uses a surface fitting algorithm based on the moving least squares method to discretize the model surface and calculate the Gaussian curvature value of each grid cell. The algorithm selects a set of neighborhood points around each calculation point and obtains the principal curvature parameter of that point through local surface fitting. Then, it calculates the Gaussian curvature that reflects the degree of surface curvature. The curvature values ​​of all calculation points are organized into a curvature distribution map corresponding to the CAD model grid. The map uses color gradients to intuitively identify different curvature regions. High curvature regions, such as edges and corners, are usually marked with warm colors.

[0021] The path planning unit receives curvature distribution map data and generates multi-view dynamic scanning path instructions accordingly. The path planning algorithm first divides the curvature distribution into regions, classifying continuous surfaces with similar curvature values ​​into the same scanning area. For each scanning area, the algorithm automatically calculates two key parameters based on the curvature: scanning speed and scanning angle interval. In high curvature regions, due to the complex surface geometry, it is necessary to reduce the scanning speed and decrease the angle interval to obtain denser measurement data. The scanning speed is inversely proportional to the curvature value; for every unit increase in curvature, the scanning speed decreases by a certain proportion. The scanning angle interval is directly proportional to the curvature value; the greater the curvature, the smaller the angular interval between adjacent scanning stations. The generated path instructions contain a series of spatial coordinates and attitude parameters of scanning stations. Each station is accompanied by a corresponding scanning speed value and viewing angle interval value. These parameters are encoded into a motion instruction format that the robot controller can recognize and sent to the robot motion platform via industrial Ethernet. When performing a scanning task, the robot automatically adjusts its movement speed and the angle change between adjacent stations based on the curvature region of its current position. This dynamic adjustment mechanism allows measurement resources to be focused on areas with complex geometric features, significantly improving overall scanning efficiency while ensuring measurement accuracy. A data interaction channel exists between the laser power adjustment unit and the path planning unit. When the path planning unit instructs the robot to enter a high-curvature region for fine scanning, it simultaneously notifies the laser power adjustment unit to pay close attention to the ambient light conditions in that region. This is because high-curvature regions are often accompanied by complex reflection and shadow effects, requiring more precise laser power control. The collaborative work of the two units ensures optimal scanning data quality under different curvature conditions.

[0022] The entire system operates based on a strict time synchronization mechanism. The sampling period of the ambient light sensing module, the laser power adjustment period, and the robot motion control period are kept as integer multiples. This design avoids control errors caused by timing mismatches between subsystems. All modules are aligned with data through precise timestamps to ensure that each scanning action is performed under preset ambient light conditions and motion parameters.

[0023] Example 2: See Figure 2The feature point extraction unit of the point cloud processing module performs multi-scale feature point detection on the generated 3D point cloud data. This process first establishes a multi-level resolution data structure and creates multiple point cloud layers from coarse to fine through voxel grid downsampling. In each point cloud layer, the rate of change of the normal vector of each point is calculated. The normal vector is obtained through local plane fitting. The neighborhood radius varies with the point cloud layer. A larger neighborhood radius is used in the coarse resolution layer to capture macroscopic features, while a smaller neighborhood radius is used in the fine resolution layer to capture microscopic details. When the angle between the normal vector of a point and the average normal vector of its neighboring points exceeds a set threshold, the point is marked as a feature point. This multi-scale detection mechanism can comprehensively identify geometric features of different sizes. Based on the extracted feature point set, the feature point descriptor generation algorithm begins operation. This algorithm employs a fast point feature histogram method, calculating the relative positions and normal vector angles between all pairs of points in the neighborhood of each feature point. These relationship parameters are then incorporated into a statistical histogram, forming a descriptor vector with rotation and translation invariance. The descriptor vector contains statistical values ​​across multiple dimensions, comprehensively representing the local geometric distribution characteristics around the feature point. All feature point descriptors form a descriptor database for feature matching in the subsequent point cloud registration process. The point cloud registration cost function constructed by the registration optimization unit includes two core constraint terms: a feature point matching degree constraint term, implemented by calculating the Euclidean distance between feature point descriptors in two point clouds, reflecting the similarity of corresponding feature points; and a point cloud overlap constraint term, implemented by calculating the point cloud projection overlap rate, projecting one point cloud onto the view space of another point cloud and statistically analyzing the proportion of projected points falling within the neighborhood of the target point cloud, reflecting the spatial coverage consistency of the two point clouds. The cost function weighted and fused these two constraint terms to form a comprehensive evaluation index.

[0024] A multi-objective optimization algorithm is applied to solve for the optimal transformation matrix of the point cloud registration cost function. This algorithm searches for the optimal solution in six-dimensional space, which contains three rotational degrees of freedom and three translational degrees of freedom. The optimization process has two objectives: minimizing the feature point matching distance and maximizing the point cloud overlap rate. It iterative calculations are used to find the pose transformation parameters that minimize the weighted cost function value. The optimization result is output as a fourth-order homogeneous transformation matrix, which accurately describes the spatial transformation relationship between two point clouds. Adaptive filtering is performed to construct a spatial filter based on ambient light intensity distribution data. The light intensity data collected by the ambient light sensing module is mapped to the image coordinate system, generating a background light intensity distribution map with the same size as the laser stripe image. This distribution map marks the ambient light interference intensity at each pixel location in the image. Based on this, an adaptive median filter is constructed with a dynamically adjusted window size, which is proportional to the local light intensity variance. A small window is used for fast filtering in areas with uniform illumination, while a large window is used to suppress strong noise in areas with abrupt changes in illumination.

[0025] The filtered laser stripe image enters the grayscale morphological processing stage. First, an opening operation is used to eliminate scattered noise. The size of the structuring element is dynamically determined based on the image signal-to-noise ratio. Smaller structuring elements are used to connect and repair broken areas of the laser stripes, while larger structuring elements are used to enhance the contours in blurred areas at the stripe edges. The binary image after morphological processing then enters the skeleton extraction stage. A thinning algorithm iteratively removes pixels from the stripe edges until a centerline of single-pixel width is obtained. The coordinates of the centerline are recorded and used for subsequent 3D point cloud reconstruction calculations. A data feedback channel is established between the feature point extraction unit and the registration optimization unit. When the registration optimization unit detects a low feature point matching rate in a certain area, it triggers the feature point extraction unit to perform multi-scale feature re-extraction in that area. The re-extraction process uses adjusted neighborhood parameters, reducing the neighborhood radius in areas with missing features to increase feature point density. This collaborative mechanism effectively solves the problem of uneven distribution of feature points on complex curved surfaces. The adaptive filtering module shares ambient light data with the laser power adjustment unit. When the ambient light sensing module detects local strong light interference, this information is sent to both the laser power adjustment unit and the point cloud processing module. The laser power adjustment unit increases the laser power in the corresponding area, while the point cloud processing module increases the size of the filtering window accordingly. This cross-module linkage control significantly enhances the robustness of the system under harsh lighting conditions.

[0026] The entire point cloud processing workflow adopts a pipelined architecture, with image filtering, centerline extraction, 3D reconstruction, feature extraction, and registration optimization operating in parallel. Each stage processes the output data of the previous stage and passes the results to the next stage. The pipeline design fully utilizes the parallel computing power of the GPU, enabling the system to process high-frame-rate scanned image data streams in real time. During point cloud registration, a multi-resolution registration strategy is employed. Initial registration is performed on a coarse-resolution point cloud layer, and then gradually improved to a fine-resolution layer for precise registration. This layered strategy significantly improves registration efficiency and global convergence. Simultaneously, the registration optimization unit records historical registration error data, and automatically initiates a re-registration process when an abnormal error trend is detected, ensuring the stitching accuracy of the final point cloud model.

[0027] Taking the scanning measurement of a wind turbine blade as an example, the point cloud processing module initiates the feature point extraction process. The blade is 60 meters long, with a special coating on its surface and localized damage areas. The feature point extraction unit first performs multi-level meshing on the 3D point cloud. The initial layer uses a 50 mm voxel size for downsampling to capture the overall aerodynamic shape of the blade; the second layer switches to a 20 mm voxel size to focus on the geometric features of the leading edge reinforcement area; the final layer uses a 5 mm voxel size to precisely identify the edges of the bolt holes at the blade root. When calculating the point cloud normal vector at each level, a 200 mm neighborhood radius is used to analyze the overall bending and twisting trend in the blade tip region, while a 50 mm radius is used in the blade root region to capture the details of the flange mounting holes. When the angle between the normal vector of a point and the mean of its neighborhood exceeds a threshold of 15 degrees, feature marking is triggered. At the serrated structure of the trailing edge of the blade, feature points are distributed in a regular band; in the lightning damage area, they form irregular clusters. In the feature point descriptor generation stage, for the leading-edge anti-corrosion coating area, an extended FPFH algorithm is used to increase the statistical dimension of curvature change rate; for the coating peeling area, the surface roughness histogram component is additionally calculated. The generated descriptor database contains 12-dimensional feature vectors, fully characterizing various geometric anomalies on the blade surface. When the registration optimization unit processes point clouds from adjacent stations, the constructed dual-constraint cost function shows significant performance in the blade tip region: the feature point matching degree constraint successfully associates the lightning arrester features captured by the two stations, and their descriptor Euclidean distances are all less than 0.1; the point cloud overlap degree constraint effectively handles the blade flapping deformation problem, compensating for the 0.8 mm elastic deformation through projection overlap rate calculation. In the blade root flange region, the bolt hole feature matching distance reaches 0.3, at which point the overlap degree constraint weight is automatically increased to 0.7, compensating for the feature loss problem through spatial coverage consistency. Adaptive filtering processing faces the challenge of high reflectivity on the blade surface. When the ambient light sensing module detects changes in coating reflectivity, the spatial filter window size expands to 9×9 pixels in the highly reflective area and maintains 3×3 pixels in the matte area. During a scan, a sudden change in cloud cover occurred, increasing light intensity by 300 lux within 3 seconds. The filter increased its window size in real time to suppress glare noise, and morphological processing simultaneously adjusted the size of structural elements: a 5-pixel disk structural element was used for opening operations in glare areas, while a 3-pixel rectangular structural element was used in shadow areas. Laser stripe centerline extraction encountered a special condition—water erosion pits on the blade leading edge caused stripe breakage. Morphological processing initiated multi-level repair: first, a 3-pixel diameter disk was used for closing operations to connect the break points, then a skeleton extraction algorithm was iterated 5 times to obtain a continuous centerline. In areas with abrupt changes in pit depth, the algorithm automatically marked abnormal point cloud data, triggering a local rescanning mechanism. During registration optimization, insufficient feature point density was found in the blade tip region. The feedback mechanism triggered multi-scale feature re-extraction: a 10-millimeter voxel level was added at the 40-meter spanwise position, the neighborhood radius was reduced to 30 mm, and 127 new aerodynamic vortex generator feature points were added.When reconstructing the blade from root to tip, the optimization algorithm employs a layered registration strategy: initial alignment is completed in the coarse-resolution point cloud, and then the accuracy is progressively improved, ultimately achieving complete point cloud reconstruction of a 56-meter-long blade with an accuracy of 0.05 mm. The point cloud processing pipeline maintains real-time performance under GPU acceleration; when the scanning speed reaches 150 mm / s, the processing time for a single frame is controlled within 33 milliseconds. The system specifically addresses the uneven point cloud density caused by changes in blade twist angle, automatically increasing the feature point detection sensitivity threshold in high twist angle regions to avoid mismatches due to sparse point clouds. The entire scanning process generates approximately 12 million point cloud data points, and the registration optimization unit performs a total of 37 multi-objective optimization calculations. In the lightning protection line region of the blade, the feature point matching degree constraint weight is maintained above 0.6; while in the large curvature web region, the overlap constraint weight is increased to 0.8 to address feature symmetry issues. The final point cloud model fully presents the depth gradient of the leading-edge erosion pits and the microscopic geometric features of the trailing-edge serrations.

[0028] Example 3: The multi-station synchronization module receives the pose data stream sent in real time by the robot motion platform via a high-speed industrial Ethernet. This data stream is updated at a kilohertz frequency and includes the position vector and rotation matrix of the robot end flange coordinate system relative to the base coordinate system. The position vector records the three-dimensional spatial coordinates, and the rotation matrix stores the attitude information using the rotation vector representation method. The module maintains a preset scanning station queue. Each station in the queue contains spatial coordinates, attitude angles, and timestamp information accurate to the millisecond level. When the difference between the timestamp of the real-time pose data and the timestamp of a certain station in the queue is less than the synchronization threshold, the synchronization module immediately sends a high-level pulse signal to the trigger port of the laser vision sensing module.

[0029] Upon receiving a pulse signal, the hardware trigger circuit of the laser vision sensing module synchronously starts the exposure cycle of the laser projector and the industrial camera. The laser emits modulated laser stripes within a microsecond delay, and the camera's global shutter opens at a precisely calculated moment. This hardware-level synchronization mechanism ensures that each scanning action occurs the instant the robot's end effector reaches the preset spatial position. The generation process of the scanning station queue has been verified through kinematic simulation to ensure that the robot is in a stable state when it reaches each station, avoiding image blurring due to mechanical vibration. After completing the point cloud registration calculation, the registration optimization unit writes the obtained optimal transformation matrix into the shared memory area. This matrix is ​​represented in homogeneous coordinates and contains 3×3 rotation matrix components and 3×1 translation vector components. Simultaneously, the matrix data is sent to the path planning unit of the motion control module through a cross-process communication mechanism. The path planning unit parses the matrix parameters and calculates the coordinate deviation. The deviation calculation uses a spatial transformation model. ; in: For the preset station coordinates, The actual coordinates of the point cloud center. and Representing rotation and translation transformations respectively, the calculation results are as follows: This represents the position compensation vector for the current station.

[0030] The motion control module dynamically corrects the coordinates of subsequent scanning stations based on the compensation vector. The correction process employs a recursive compensation algorithm. First, an error propagation model is established in the global coordinate system. The compensation vector of the current station is decomposed into translational and rotational components. The translational component is directly superimposed on the spatial coordinates of subsequent stations, while the rotational component is distributed to adjacent stations through an attitude interpolation algorithm. For example, when a 0.5-degree deviation around the Z-axis is detected, the yaw angle of the subsequent ten stations is increased by 0.05 degrees in sequence. This progressive compensation strategy avoids mechanical shocks caused by sudden attitude changes. A bidirectional data channel is established between the multi-station synchronization module and the motion control module. When the path planning unit updates the scanning station coordinates, it immediately sends a station queue update command to the synchronization module. The synchronization module receives the new queue data and overwrites the original queue. The update process uses a transaction processing mechanism to ensure data integrity. The scanning trigger signal is paused at the moment of queue switching, and the synchronization trigger function is resumed after the new queue is fully loaded. The entire update process is completed when the robot moves to the transition path between two stations, without affecting continuous scanning operations. The triggering accuracy of the laser vision sensing module is ensured through a time calibration mechanism. During system initialization, a dedicated timing calibration program is executed to measure the delay time between the trigger signal and the stable laser output, as well as the time deviation between the camera exposure signal and the mechanical shutter opening. These delay parameters are written into the compensation register of the synchronization module, and a corresponding time compensation amount is pre-added during actual triggering to ensure that the laser stripe projection and image acquisition strictly match the robot pose during dynamic scanning. The optimal transformation matrix stored in the registration optimization unit has a version identifier and a timestamp. When receiving matrix data, the motion control module verifies the time series and only enables compensation when the matrix timestamp is later than the timestamp generated by the currently executed path instruction, avoiding compensation errors caused by using outdated transformation parameters. The matrix data is stored in a circular buffer structure in the shared memory area, and the latest ten sets of transformation matrices are reserved for compensation trend analysis.

[0031] When the robot motion platform receives the corrected path instructions, its trajectory planner recalculates the joint space path. The trajectory planning employs a seventh-order polynomial interpolation algorithm to ensure smooth motion. In cases of large compensation amounts, it automatically inserts transition path points to prevent overshoot or vibration during high-speed movement. All coordinate correction operations are performed in Cartesian space and converted into joint angle commands through robot inverse kinematics. Throughout the scanning process, pose data, transformation matrices, and compensation parameters are recorded in the operation log. The log file contains detailed time series and spatial coordinate information for offline analysis of scanning accuracy and system stability. The log records are stored in binary format, and the spatial relationships of each scanning station can be reconstructed using dedicated parsing tools. The multi-station synchronization mechanism and coordinate compensation mechanism together constitute a closed-loop control system, eliminating robot positioning errors and environmental disturbances in real time. The spatial positioning accuracy of the scanning stations is always maintained within the sensor measurement accuracy range. This dynamic correction capability is particularly suitable for slowly changing conditions such as thermal deformation of large structural components or foundation settlement.

[0032] Example 4: See Figure 3 The closed-loop verification module continuously receives the 3D point cloud data stream generated by the point cloud processing module. This data stream is organized according to the scanning station order. The point cloud data packet corresponding to each station has spatial coordinate markers and timestamps. An internal point cloud buffer is established to store the complete point cloud data of the five most recent scanning stations. When new point cloud data arrives, the verification algorithm automatically calculates the overlapping area between the point cloud and the point cloud of the previous station. The overlapping area is determined by the spatial grid mapping method, which divides the point cloud space into cubic grid cells and counts the proportion of points in the same grid cell between the two point clouds. The error calculation of the point cloud overlapping area adopts an improved iterative nearest point algorithm. This algorithm first randomly samples corresponding point pairs in the overlapping area, calculates the Euclidean distance between these point pairs, removes outliers whose distance exceeds three times the standard deviation, and calculates the average distance value of the remaining inliers as the error index. The error value is recorded in millimeters. At the same time, the proportion of inliers is counted as the overlap quality coefficient. All error data, along with the corresponding station number, is sent to the registration optimization unit. The registration optimization unit maintains a dynamic weight coefficient table, which records the weight allocation of feature point matching degree constraints and point cloud overlap degree constraints in the point cloud registration cost function. The initial weight value is set to 0.5 for each. When the point cloud overlap region error data sent by the closed-loop verification module is received, the weight adjustment algorithm automatically updates the weight coefficients according to the error value. The update strategy is based on the error threshold mechanism, setting two critical values ​​of 0.05mm and 0.1mm. When the error is below the lower limit, the overlap degree constraint is emphasized, and when the error exceeds the upper limit, the matching degree constraint is emphasized.

[0033] The application of particle swarm optimization (PSO) in point cloud registration employs an adaptive parameter adjustment mechanism. The population size is dynamically set according to the point cloud density, typically one percent of the number of points. The number of iterations is proportional to the point cloud complexity. For point clouds of structural components with complex geometric features, the number of iterations is appropriately increased to ensure convergence. The particle position vector is defined as a six-dimensional solution space, with the first three dimensions representing rotation Euler angles and the last three dimensions representing translation vectors. The initial value of the velocity vector is automatically set according to the point cloud scale. The fitness function, i.e., the point cloud registration cost function, requires significant computational resources. Therefore, a parallel computing architecture is adopted, distributing the fitness calculation task of each particle in the population to multiple computing cores simultaneously. During the calculation, a spatial partitioning tree is used to accelerate the nearest neighbor search, significantly reducing the time overhead of corresponding point matching. After each iteration, the historical trajectory of the global optimal solution is recorded. The iteration automatically terminates when the optimal solution has not improved for ten consecutive generations. Refer to Table 1 to show the adjustment of weight coefficients under different point cloud overlap region error conditions.

[0034] Table 1: Correspondence Table for Point Cloud Registration Weight Coefficient Adjustment Error range (mm) Matching Constraint Weights Overlap Constraint Weight Applicable Scenarios Description <0.05 0.4 0.6 High-precision matching stage 0.05-0.1 0.5 0.5 Standard registration process >0.1 0.7 0.3 Initial coarse registration stage The weight coefficient adjustment process features a smooth transition. When the error value fluctuates near the critical value, linear interpolation is used to calculate the weight value to avoid registration instability caused by sudden coefficient changes. Historical weight change trends are recorded, and when frequent and significant weight adjustments are detected within a short period, a system self-check program is triggered to investigate point cloud quality or environmental interference factors. The inertial weight parameter of the particle swarm optimization algorithm is adjusted in conjunction with the weight coefficient. When the matching degree constraint weight increases, the inertial weight is increased accordingly to enhance global search capability; when the overlap constraint weight increases, the inertial weight is decreased to improve local refinement accuracy. This linkage mechanism allows the algorithm to dynamically balance exploration and utilization based on actual registration needs. In addition to calculating the point cloud overlap region error, the closed-loop verification module also performs point cloud quality assessment. Assessment indicators include point cloud density uniformity, noise point ratio, and feature integrity. These quality indicators are packaged and sent along with the error data, providing a more comprehensive decision-making basis for weight adjustment. When a significant deterioration in point cloud quality is detected, a rescan request is automatically sent to the motion control module. A feedback channel is established between the registration optimization unit and the point cloud processing module. When the registration result after weight adjustment still cannot meet the accuracy requirements, the feature point extraction unit is triggered to recalculate the feature point descriptors, using different feature detection parameters, paying special attention to areas with large previous registration errors. The regenerated descriptor database is used for a new round of registration calculation.

[0035] The entire closed-loop verification process is performed synchronously with point cloud acquisition. Registration parameters are optimized in real time during the scanning and measurement process to ensure the consistency and high accuracy of the final generated overall point cloud model. This real-time optimization mechanism effectively compensates for mechanical deformation and measurement errors caused by changes in ambient temperature, especially for long-cycle scanning tasks of large structural components. After the particle swarm optimization algorithm outputs the optimal transformation matrix, the registration optimization unit verifies the results, calculating the overlap index and feature matching error of the registered point cloud. When the verification index exceeds the acceptance standard, re-registration is automatically initiated. Re-registration uses adjusted weighting coefficients and optimized parameters to ensure the reliability of the output results. All registration results are labeled with a quality assessment tag.

[0036] Example 5: The path replanning unit of the motion control module continuously monitors the point cloud overlap region error data stream sent by the closed-loop verification module. This data stream is updated at a fixed frequency. Each data packet contains the error value, the corresponding station number, and timestamp information. When the error of the point cloud overlap region between adjacent stations exceeds the preset threshold of 0.2 mm three times consecutively, the replanning unit immediately starts the path replanning process. This process first locks the spatial area where the error exceeds the standard and determines the three-dimensional spatial range where the error occurs by analyzing the station coordinates. Taking the leading edge scanning of an aircraft wing as an example, when the closed-loop verification module detects that the point cloud overlap region error between stations S15 and S16 reaches 0.25 mm, the path replanning unit retrieves the original scanning path parameters corresponding to these two stations from the database. The original path sets the station spacing to 120 mm, the scanning speed to 100 mm / s, and the viewing angle interval to 8 degrees in this area. At the same time, it acquires the curvature distribution data of this area, showing that the leading edge curvature value is 0.08 mm. The path replanning unit accesses the updated point cloud registration cost function of the registration optimization unit, extracts the latest adjusted weight coefficient combination in the function, and the current weight shows that the feature point matching degree constraint weight is 0.65 and the point cloud overlap degree constraint weight is 0.35. This weight allocation indicates that there is a significant feature matching problem in this area. The unit calls the 3D CAD model interface to load the original design surface data of the wing leading edge area. By comparing the deviation distribution map between the actual point cloud and the design surface, it is identified that the maximum deviation occurs at the airfoil turning point at 25% of the chord length.

[0037] Based on the deviation analysis results, the replanning algorithm recalculates the scanning path parameters for the coverage problem area. The new path reduces the station spacing to 60 mm, doubles the scanning station density, reduces the viewing angle interval from 8 degrees to 4 degrees, and lowers the scanning speed to 70 mm / s. Simultaneously, an additional circular scanning path is added for the airfoil transition, containing eight equally angularly distributed stations, forming 360-degree coverage around the transition. After passing security checks, the newly generated dynamic scanning path instructions are sent to the multi-station synchronization module to replace the original path segments. The synchronization module immediately loads the new path queue after the robot completes the scan at the current station S16. The new queue starts at station S16.1, ends at station S16.8 (circular scan), and finally connects to the original station S17. The queue switching process is completed during robot movement, and motion continuity is ensured through real-time pose interpolation. When the robot motion platform executes a new path, it adopts adaptive speed control at the airfoil turning point. When the end effector approaches the high curvature area, it automatically reduces the moving speed and restores the standard speed in the flat area. The laser vision sensing module synchronously adjusts the sampling frequency and increases the image acquisition frame rate from 30 Hz to 60 Hz in the high-density station area to ensure that enough data points are acquired.

[0038] After sending new instructions, the path replanning unit continuously monitors the closed-loop verification data. When scanning to the new station S16.5, the verification module reports that the error in the overlapping area of ​​the point cloud between this station and the previous station S16.4 has been reduced to 0.12 mm, meeting the system accuracy requirements. The unit records the parameter adjustment records for this replanning, including key data such as the station spacing reduction ratio and speed adjustment range, forming a replanning case stored in the knowledge base. The replanning knowledge base organizes the case data in a tree structure, classifying and storing them according to structural component type and surface features. When scanning a new type of structural component, the system automatically matches historical cases with similar features and preloads optimized parameters as initial settings. For example, when scanning the tenon of an engine blade, the path parameter template in the turbine disk tenon groove scanning case is automatically called. The event log generated throughout the replanning process is recorded in detail. The log includes original error data, CAD model analysis screenshots, new path parameter tables, execution process status codes, etc. The log file is associated with the point cloud data through timestamps, allowing for retrospective viewing of the scanning path version corresponding to each point cloud during later data processing. The path replanning mechanism features abnormal interruption handling capabilities. When the robot encounters unexpected obstacles while executing a new path, it triggers an emergency stop in real time and saves the current state. After the obstacle is cleared, it automatically resumes scanning from the breakpoint. The scan data is seamlessly integrated with the point cloud acquired before the interruption. This robust design ensures the continuity of long-term scanning of large structural components. For segmented scanning scenarios of ultra-large structural components, the replanning unit supports cross-regional collaborative optimization. When multiple regions simultaneously report excessive errors, the unit comprehensively analyzes the spatial relationships of each region and generates a globally optimized path scheme, avoiding path conflicts caused by local optimization. For example, in the scanning of ship hull segments, the scanning path sequence of the deck and side regions is optimized simultaneously.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A robotic laser vision three-dimensional scanning and measurement system for large structural components, characterized in that, include: The robot motion platform includes a laser vision sensing module, an ambient light sensing module, a point cloud processing module, and a motion control module installed at the end of the robot motion platform. The ambient light sensing module collects ambient light intensity distribution data in real time; the motion control module generates multi-view dynamic scanning path instructions based on the ambient light intensity distribution data and the 3D model data of the large structural component under test; the robot motion platform executes the multi-view dynamic scanning path instructions to drive the laser vision sensing module to move; the laser vision sensing module projects laser stripes onto the surface of the large structural component under test and simultaneously collects laser stripe images; the point cloud processing module performs adaptive filtering on the laser stripe images and generates 3D point cloud data.

2. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 1, characterized in that, The laser vision sensing module includes a laser power adjustment unit; the laser power adjustment unit receives real-time ambient light intensity distribution data sent by the ambient light sensing module, and dynamically adjusts the laser emission power of the laser vision sensing module according to a preset laser intensity threshold; the preset laser intensity threshold is greater than the maximum value of the current ambient light intensity distribution data and less than half of the maximum emission power of the laser vision sensing module.

3. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 2, characterized in that, The motion control module includes a surface curvature analysis unit and a path planning unit; the surface curvature analysis unit calculates the surface curvature distribution based on the three-dimensional model data of the large structural component to be tested; the path planning unit generates the multi-view dynamic scanning path instruction according to the surface curvature distribution, the multi-view dynamic scanning path instruction includes scanning speed and scanning view interval; the scanning speed is inversely proportional to the surface curvature distribution, and the scanning view interval is directly proportional to the surface curvature distribution.

4. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 3, characterized in that, The point cloud processing module includes a feature point extraction unit and a registration optimization unit. The feature point extraction unit performs multi-scale feature point detection on the generated 3D point cloud data and generates feature point descriptors. The registration optimization unit constructs a point cloud registration cost function based on the feature point descriptors and uses a multi-objective optimization algorithm to solve for the optimal transformation matrix of the point cloud registration cost function. The point cloud registration cost function includes feature point matching degree constraints and point cloud overlap degree constraints.

5. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 4, characterized in that, The adaptive filtering process includes: the point cloud processing module constructs a spatial filter based on the ambient light intensity distribution data, uses the spatial filter to suppress background noise in the laser stripe image, and extracts the center line of the laser stripe through grayscale morphological operations.

6. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 5, characterized in that, The system also includes a multi-station synchronization module; the multi-station synchronization module receives real-time pose data sent by the robot motion platform, and triggers the scanning action of the laser vision sensing module according to the real-time pose data; the triggering time of the scanning action is synchronized with the timestamp of the robot motion platform reaching the preset scanning station in the multi-view dynamic scanning path instruction.

7. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 6, characterized in that, The registration optimization unit stores the optimal transformation matrix and sends the optimal transformation matrix to the motion control module; the motion control module corrects the scanning station coordinates in the subsequent multi-view dynamic scanning path command according to the optimal transformation matrix.

8. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 7, characterized in that, The system includes a closed-loop verification module; the closed-loop verification module receives three-dimensional point cloud data generated by the point cloud processing module, calculates the point cloud overlap region error between adjacent scanning stations, and sends the point cloud overlap region error to the registration optimization unit; the registration optimization unit updates the weight coefficients of the point cloud registration cost function according to the point cloud overlap region error.

9. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 8, characterized in that, The multi-objective optimization algorithm employs a particle swarm optimization algorithm, wherein the particle position vector of the particle swarm optimization algorithm includes rotation and translation parameters, and the fitness function of the particle swarm optimization algorithm is the point cloud registration cost function.

10. The robotic laser vision three-dimensional scanning and measurement system for large structural components according to claim 9, characterized in that, The motion control module includes a path replanning unit; when the error in the overlapping area of ​​the point cloud exceeds a preset error threshold, the path replanning unit regenerates a multi-view dynamic scanning path instruction covering the overlapping area of ​​the point cloud based on the updated point cloud registration cost function.

Images