Accessory quality inspection method and system based on laser scanning data analysis
By constructing a 3D model of the parts using laser scanning and combining it with big data analysis and point cloud segmentation models, the problems of insufficient coverage of the three-dimensional structure and difficulty in identifying small defects in traditional inspection methods have been solved, achieving efficient and accurate quality inspection of the parts.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-10
AI Technical Summary
Traditional methods for inspecting the quality of parts rely on manual visual inspection and two-dimensional image analysis, which makes it difficult to achieve comprehensive coverage of the three-dimensional structure of parts and accurate location of small defects, resulting in poor repeatability and high missed detection rate.
Laser scanning instruments are used to acquire point cloud data of parts, construct 3D models, and combine multi-angle big data analysis and pre-trained point cloud segmentation models to identify surface defects of parts and generate a comprehensive evaluation report.
It enables comprehensive inspection of the three-dimensional structure of parts, improves the accuracy and efficiency of defect identification, reduces the false detection rate, and optimizes the production process.
Smart Images
Figure CN121639614A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of laser measurement data processing technology, in particular to a kind of accessory quality inspection method and system based on laser scanning data processing. BACKGROUND
[0002] Accessory quality inspection is the process of inspecting and evaluating the performance, size, appearance and other indicators of product accessories to ensure that they meet design requirements, production standards and usage specifications. Through accessory quality inspection, defects and problems that occur during the production process of accessories can be found in time, and unqualified products can be prevented from entering the next process, helping to optimize production processes and process parameters, and improving production efficiency and resource utilization.
[0003] Traditional accessory quality inspection usually uses manual visual inspection and manual measurement tools, relying on the subjective experience of operators, affecting the repeatability and reproducibility of measurement, and it is difficult to realize real-time monitoring and feedback of the production process. Although the existing quality inspection method based on machine vision can realize automatic detection of accessory quality, it only obtains planar image data from two-dimensional space, analyzes surface quality problems, and cannot realize comprehensive coverage of the spatial structure of accessories, making it difficult to detect hidden defects in the three-dimensional structure of accessories, accurately locate small quality problems, and need to construct a three-dimensional model of the accessory through a laser measuring instrument to improve the comprehensiveness of accessory quality inspection and ensure product quality. SUMMARY
[0004] To solve the above technical problems of the prior art, the present application provides an accessory quality inspection method and system based on laser scanning data analysis, which constructs a three-dimensional model of the accessory based on the point cloud data of the accessory obtained by a laser measuring instrument, and performs quality inspection on the accessory through a multi-angle big data analysis method. In addition, for accessory defects, a pre-trained point cloud segmentation model is used to deeply analyze the accessory point cloud data to obtain accurate defect recognition results, and then a comprehensive evaluation report of accessory quality inspection is formed.
[0005] The present application provides an accessory quality inspection method based on laser scanning data analysis, comprising: (1) using a laser scanning instrument to comprehensively scan the accessory, collecting point cloud data of the surface of the accessory, and preprocessing the collected accessory point cloud data; (2) gridding the preprocessed accessory point cloud data to construct a three-dimensional model of the accessory, and extracting key features for quality inspection from the model; (3) processing the key features of the accessory through a multi-dimensional analysis method, and using big data analysis to obtain statistical data of the quality of the accessory for quality verification; (4) input the pretreated accessory point cloud data into the pre-trained point cloud segmentation model to obtain the identification result of the accessory surface defect, determine the defect type and calculate the defect rate index; (5) comprehensively analyze the results of big data and the identification results of the point cloud segmentation model to form a complete accessory quality inspection evaluation report and optimize the accessory production process.
[0006] Further, the three-dimensional imaging method can obtain micro feature information such as depth, normal direction and curvature when detecting the surface of an accessory with a concave-convex structure, and model the complete spatial structure of the accessory. On a two-dimensional grayscale image, the defect and non-defect features are very similar, and the proportion of non-defects in the production process is much higher than that of real defects, which is easy to cause false detection and reduce the practicability of defect detection; therefore, three-dimensional imaging has significant effect in accessory quality detection.
[0007] Three-dimensional imaging often uses photometric stereo method, time-of-flight method, scanning method, stereo vision method and structured light method. The scanning method uses laser to scan the entire target surface to realize three-dimensional measurement, uses the high directionality and high monochromaticity of laser to emit and receive laser, and generates a large number of point coordinates according to the position of laser in the scanning process to form point cloud data and accurately represent the shape of the target object. The spatial information provided by three-dimensional point cloud data helps to obtain more defect feature information, and a deep learning method is often used to learn and extract complex features from a large amount of unstructured data set, so as to extract features from three-dimensional point cloud data and distinguish non-defect and defect areas of the target surface by using a segmentation model.
[0008] The application also provides an accessory quality inspection system based on laser scanning data analysis, comprising: A data acquisition module: a laser scanning instrument is used to comprehensively scan the accessory, collect point cloud data of the surface of the accessory, and pretreat the collected accessory point cloud data; A three-dimensional model construction module: used for gridding the pretreated accessory point cloud data to construct a three-dimensional model of the accessory and extracting key features for quality inspection from the model; A data analysis module: used for processing the key features of the accessory by a multi-dimensional analysis method, obtaining statistical data of the quality of the accessory by big data analysis, and verifying the quality; A defect detection module: used for inputting the pretreated accessory point cloud data into a pre-trained point cloud segmentation model to obtain the identification result of the accessory surface defect, determining the defect type and calculating the defect rate index; A quality evaluation module: used for comprehensively analyzing the results of big data and the identification results of the point cloud segmentation model to form a complete accessory quality inspection evaluation report and optimize the accessory production process.
[0009] The application discloses the following technical effects: The application provides a kind of accessory quality inspection method and system based on laser scanning data analysis, obtains accessory point cloud data by three-dimensional imaging method of laser scanning, constructs accessory three-dimensional model, comprehensively describes the spatial structure of accessory, clearly shows the stereoscopic information of small defect on accessory surface, solves the problem of missed detection caused by defect position hidden and similar to non-defect area characteristics.The application obtains detailed key features from accessory three-dimensional model, uses multidimensional big data analysis method, comprehensively evaluates the quality distribution of accessory, and statistics accessory quality data, digs out product quality fluctuation and abnormal reason, adjusts parameter setting in production process in time;In addition, in the accessory surface quality analysis stage, introduce point cloud segmentation model based on deep learning, utilize the powerful feature extraction capability of neural network to analyze and process accessory point cloud data, capture accessory spatial structure information and defect distribution characteristics, use hierarchical self-attention mechanism to classify point cloud according to sparsity, model point cloud internal structure respectively, help to improve the resolution ability of model to defect and non-defect area, further improve the accuracy of defect detection.The application analyzes laser scanning data of accessory in depth, combines three-dimensional structure and deep learning model, realizes comprehensive, intelligent and automatic of accessory quality inspection, improves quality inspection efficiency. BRIEF DESCRIPTION OF DRAWINGS
[0010] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings of the embodiments of the application will be briefly introduced below, and the flowchart is used to illustrate the operations performed by the system according to the embodiments of the application in this application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can be added to these processes, or one or more steps of operation can be removed from these processes.
[0011] Figure 1 The flowchart of the accessory quality inspection method based on laser scanning data analysis provided by the embodiments of the application.
[0012] Figure 2 The structure diagram of the point cloud segmentation model provided by the embodiments of the application.
[0013] Figure 3 The structure diagram of the continuous two-layer hierarchical point cloud attention module in the point cloud segmentation model provided by the embodiments of the application.
[0014] Figure 4 The structure diagram of the accessory quality inspection system based on laser scanning data analysis provided by the embodiments of the application. DETAILED DESCRIPTION
[0015] The above description is only a summary of the technical solutions of the present application. In order to make the technical means of the present application more clear, the present application can be implemented according to the content of the description, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described.
[0016] In order to make the purposes, technical solutions and advantages of the present application more clear, the following will combine the drawings to make further detailed description of the present application, the described embodiments should not be regarded as limitation of the present application, all other embodiments obtained by the person skilled in the art without creative labor belong to the scope of protection of the present application.
[0017] In the following description, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art of the technology to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.
[0018] Embodiment one, the embodiment of the present application provides a kind of accessory quality inspection method based on laser scanning data analysis, as shown in Figure 1 The method comprises the following steps: Step S10, using laser scanning instrument to scan accessory comprehensively, collects the point cloud data of accessory surface, and pre-processes the collected accessory point cloud data.
[0019] In this embodiment, fixed laser scanning instrument is installed on accessory production line, and four angle scanning accessories are fixed;The scanning range is set to cover the complete accessory surface, the scanning resolution and scanning speed are adjusted, the coordinate system of the point cloud data collected at different angles is unified, the data consistency is ensured, and the coordinates of each point on the accessory surface are determined by triangulation method, including the following processes: The laser scanning instrument emits a beam of laser, which irradiates a certain point on the surface of the target accessory;The sensor inside the instrument receives the reflected light and records the direction and angle of the reflected light; By measuring the angle between the laser emitting point, the target point and the sensor, the distance from the target point to the laser emitting point is calculated using the geometric relationship of the triangle; Combined with the position and direction of the laser emitting point and the calculated distance, the coordinates of the target point in the three-dimensional coordinate space are determined;The coordinate information of all points on the surface of the accessory is obtained, and then the accessory point cloud data is obtained.
[0020] The steps of preprocessing the collected accessory point cloud data are as follows: Align the point cloud data collected at different angles by a feature point matching algorithm: Calculate the normal and curvature of each point in the multi-angle point cloud data, take the points with curvature greater than the threshold as feature points, and obtain the local direction histogram of each feature point as the descriptor of each feature point; After extracting and describing the feature points, match the feature points between the point clouds at different angles, calculate the Euclidean distance between the feature point descriptions, and take the feature points with a distance less than the threshold as matching point pairs; Obtain multiple matching point pairs in the point cloud data at different angles, calculate the rigid body change between the matching point pairs by using the iterative closest point algorithm, including the rotation and translation transformation matrix, apply the matrix to map and transform the feature points, and realize the spatial alignment of the point cloud data at different angles.
[0021] Splice and fuse the spatially aligned multi-angle point cloud data to obtain complete accessory point cloud data; By radius filtering, the radius range covering the complete structure of the accessory is determined, and the outliers beyond the range are removed; Adaptive point cloud density sampling method is used for downsampling of the accessory point cloud data, including: Calculate the Euclidean distance between the accessory point clouds by K nearest neighbor algorithm, select K nearest neighbors for each accessory point cloud, and obtain multiple point cloud sets to capture the local geometric relationship between the accessory point clouds; Use the Gaussian kernel function to perform weighted summation on the point clouds in the neighborhood, obtain the local density of each accessory point cloud and normalize it, use the exponential function to apply weight to the normalized density, and obtain the density adjustment factor; Adjust the distance value of each accessory point cloud and its K nearest neighbors by the density adjustment factor, weaken the geometric influence of dense areas, and improve the structure contribution of sparse areas, and maintain the global balance of point cloud distribution; Quantify the importance of each point in the global geometric structure, filter the significant points by smoothing the correlation calculation, and the global correlation score of each accessory point cloud is represented by the formula:
[0022] Wherein, represents the global correlation score of the accessory point cloud , the density weighted distance of the accessory point cloud and , and , the local density and density adjustment factor of the accessory point cloud , respectively. It is the Gaussian kernel scaling parameter, used to smooth the point cloud of the accessory. and The effect of the distance between them on the Gaussian kernel weights; For each component point cloud, the global correlation scores of its K nearest neighbors are sorted, and the M points with the highest scores are selected as the final sampling results according to the set number of output points M.
[0023] Step S20: Mesh the preprocessed part point cloud data, construct a 3D model of the part, and extract key features for quality inspection from the model.
[0024] In this embodiment, point cloud processing software is used to mesh the preprocessed part point cloud data and generate a 3D model of the part, including the following steps: Import the preprocessed point cloud data into CloudCompare point cloud processing software, select the meshing method based on greedy mapping triangulation, and set the resolution and smoothing parameters according to the overall density of the point cloud. Each 3D point cloud in the accessory point cloud data is projected onto a plane through a normal, and the projected point cloud in the plane is triangulated to obtain the connection relationship of each point cloud. A spatial region growth algorithm based on Delaunay is adopted, a sample triangle is selected as the initial surface, the surface boundary is expanded, and a complete triangular mesh surface is formed. The topological connection order of the point cloud in three-dimensional space is determined based on the connection relationship between the projected point clouds of the accessories, and a triangular grid of fixed size is formed according to the set resolution. Select the filter tool in the software and use the Laplacian smoothing algorithm to smooth the triangular mesh according to the smoothing parameters, thereby reducing mesh surface noise; The number of mesh vertices is reduced by using a quadrilateral simplification algorithm, and the holes in the mesh are filled to obtain a complete 3D model of the accessory.
[0025] CloudCompare point cloud processing software provides feature extraction tools to analyze the 3D model of accessories. The key features extracted include size features, shape features, and angle features. Size features include length, width, height, diameter, and aperture. Shape features include flatness, roundness, cylindricity, and conicity. Angular features include angle and slope.
[0026] Step S30: Process the key features of the parts through multi-dimensional analysis methods, and use big data analysis to obtain statistical data on the quality of the parts for quality verification.
[0027] In this embodiment, the multi-dimensional analysis includes dimensional accuracy, shape deviation, surface quality and assembly gap analysis. Surface quality analysis is achieved through a pre-trained point cloud segmentation model, while other dimensional analyses are achieved by extracting key features from the 3D model of the parts and obtaining statistical indicators through big data analysis.
[0028] For dimensional accuracy analysis, it is necessary to obtain the standard dimensional requirements for each component, calculate the deviation between the measured dimensional characteristic data and the standard data, and determine whether the deviation value is within the tolerance range; and calculate the statistical indicators of all component dimensional data, including mean, standard deviation, maximum and minimum values, to describe the overall data distribution; based on the data distribution, a clustering algorithm is used to classify the quality level according to the dimensional distribution, and evaluate the dimensional quality level of each component. For shape deviation analysis, a standard 3D model of each component is obtained and imported into 3D modeling software along with the real-time acquired component 3D models. The shape differences between the models are analyzed by comparison, the shape feature deviation is calculated, and a deviation distribution histogram is plotted to determine the location where shape deviations are concentrated, which is then a key focus during the production process. Assembly gap analysis focuses on the assembly gaps between components, establishes a regression-based mathematical model to describe the correlation between assembly gaps and dimensional features, predicts data changes, and provides early warnings for quality defects.
[0029] Step S40: Input the pre-processed part point cloud data into the pre-trained point cloud segmentation model to obtain the identification results of surface defects of the part, determine the defect type and calculate the defect rate index.
[0030] In this embodiment, the detailed steps for obtaining the pre-trained point cloud segmentation model include: Collect point cloud data of different types of defects on the surface of the parts, label the type of each point cloud in the point cloud data, and divide the defect types and regions as the dataset for training the point cloud segmentation model. The dataset is divided into training set, validation set and test set in a ratio of 8:1:1. A point cloud segmentation model is established, taking the point cloud data of component defects as input features. The point cloud data is mapped to the feature space through point cloud embedding. A hierarchical point cloud attention module is used to model sparse and dense point clouds according to the point cloud distribution density, learn the correlation between point clouds, capture the defect shape and distribution features, and output point cloud-level classification prediction results. Set the sampling rate of the feature downsampling module in the point cloud segmentation model to 1 / 2, corresponding to the interpolation rate of the linear interpolation method in the upsampling module. Configure the input and output dimensions of each module in the model, and set the maximum number of point clouds processed by each module to 8000. Cross-entropy loss was chosen as the loss function for the model to handle the classification prediction of each accessory point cloud. The Adam optimizer was selected and the initial learning rate was set to 0.0003. The change process of the loss function was recorded during the model training process, and the change curve of the loss function was displayed through TensorBoard. The model performance is evaluated using evaluation metrics, and the model parameters are updated through backpropagation. When the loss function converges, the parameter file of the best-performing model is retained, and the hyperparameter settings of the model are recorded. This is used as a pre-trained point cloud segmentation model to process real-time collected accessory point cloud data.
[0031] Step S50: Based on the comprehensive big data analysis results and the identification results of the point cloud segmentation model, a complete parts quality inspection and evaluation report is generated to optimize the parts production process.
[0032] Example 2: This embodiment of the invention provides a detailed structure of a point cloud segmentation model for detecting surface defects in accessories, such as... Figure 2 As shown: Using preprocessed component point cloud data as the model's input features, the point cloud embedding layer maps the point cloud data to the feature space. Common methods using linear layers or multilayer perceptrons extract point cloud features that only contain color information about the point cloud's own location, lacking local geometric and contextual information. Therefore, kernel convolution is introduced to extract information from relevant points and features surrounding the point cloud. The kernel convolution operation process... Expressed using the following formula:
[0033] in, and These represent the input point cloud and related points within the point cloud's neighborhood, respectively. This represents the local neighborhood of the input point cloud. Represents the kernel function, taken from the input point cloud. Using the relevant points of the center as input The kernel function represents the relevant features corresponding to each relevant point, describing the color information of the point cloud; a set of learnable core points is defined, and feature aggregation is achieved by using the core points around the input point as references.
[0034] in, Indicates regional centralization. This represents the number of relevant points within the neighborhood of the input point cloud. This represents the core points of the input point cloud. Indicates the number of core points. This represents the learnable weight matrix corresponding to the core point; express and The correlation is expressed by the following linear correlation:
[0035] in, This represents the function that takes the maximum value. Indicates the distance between two points. These are weighting coefficients, determined based on the point cloud density; displacement parameters are added to the kernel function to further learn the positional information between points and adaptively fit the shape of each point cloud space.
[0036] The point cloud features embedded in the point cloud are then fed into the hierarchical point cloud attention module, such as... Figure 3 As shown, two consecutively stacked hierarchical convolutional attention modules enable complete interaction between point clouds. Traditional self-attention mechanisms, where the query vector only focuses on the local point cloud within its own window, suffer from a limited effective receptive field and cannot capture long-range contextual dependencies between distant point clouds, leading to incorrect predictions. Therefore, a hierarchical sampling strategy is adopted to filter the point cloud constituting the key vector, guiding the collection of long-range dependencies in the query vector during the attention calculation process; the hierarchical sampling strategy divides the point cloud feature space into a partition of size [missing information]. Non-overlapping cube windows; for each key point cloud, within its own... The module searches for dense points within a small window; additionally, it uses the farthest point sampling processing module to find sparse points within a larger window, combining sparse and dense points to form the final key vector; a hierarchical self-attention mechanism normalizes the original input point cloud features to obtain query and value vectors, which, along with the selected key vectors, participate in attention calculation; the complete hierarchical point cloud attention module includes a pre-normalization layer and a feedforward layer, introducing residual structures between each layer to supplement the original information and prevent overfitting, and in the second module, the divided cubic window is offset. This further enhances the information interaction between windows; through a hierarchical sampling strategy, the receptive field is significantly expanded, and query features can effectively aggregate long-distance context.
[0037] Point cloud features processed by the hierarchical point cloud attention module are downsampled to reduce feature dimensionality. The point cloud segmentation model implements four stages of downsampling and hierarchical point cloud attention processing to model the contextual relationships of multi-scale point cloud features. The output features of the last attention module are upsampled four times to restore the original feature size, and the multi-scale feature information output from each stage is aggregated as the final output of the model, displaying the segmentation and classification results of defective and non-defective regions of the parts.
[0038] Example 3: The component quality inspection system based on laser scanning data analysis provided in this embodiment of the invention can execute the component quality inspection method based on laser scanning data analysis provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. The system structure is as follows: Figure 3 As shown, it includes the following modules: Data acquisition module: Uses a laser scanning instrument to fully scan the parts, collect point cloud data on the surface of the parts, and preprocesses the collected point cloud data of the parts. 3D model building module: used to mesh the preprocessed part point cloud data, build a 3D model of the part, and extract key features for quality inspection from the model; Data Analysis Module: Used to process key features of parts through multi-dimensional analysis methods, and to obtain statistical data on part quality through big data analysis for quality verification; Defect detection module: Used to input pre-processed part point cloud data into a pre-trained point cloud segmentation model, obtain the identification results of surface defects of the parts, determine the defect type and calculate the defect rate index; Quality assessment module: Used to integrate big data analysis results and point cloud segmentation model recognition results to generate a complete parts quality inspection and assessment report, and optimize the parts production process.
[0039] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.
[0040] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
Claims
1. A method for quality inspection of a fitting based on laser scanning data analysis, characterized in that, The method comprises: (1) using a laser scanning instrument to comprehensively scan the accessory, collect point cloud data of the surface of the accessory, and pre-process the collected accessory point cloud data; (2) meshing the pre-processed accessory point cloud data, constructing an accessory three-dimensional model, and extracting key features for quality inspection from the model; (3) processing the key features of the accessory through a multi-dimensional analysis method, using big data analysis to obtain statistical data of the quality of the accessory for quality verification; (4) inputting the pre-processed accessory point cloud data into a pre-trained point cloud segmentation model to obtain the recognition result of the surface defects of the accessory, determining the defect type and calculating the defect rate index; (5) combining the big data analysis result and the recognition result of the point cloud segmentation model to form a complete accessory quality inspection and evaluation report, and optimizing the accessory production process.
2. The method for analyzing the quality of a fitting based on laser scanning data according to claim 1, wherein, In step (1), the process of collecting point cloud data by the laser scanning instrument includes: The laser scanning instrument emits a beam of laser light to illuminate a certain point on the surface of the target accessory; the sensor inside the instrument receives the reflected light and records the direction and angle of the reflected light; By measuring the angle between the laser emission point, the target point and the sensor, the distance from the target point to the laser emission point is calculated using the geometric relationship of a triangle; Combined with the position and direction of the laser emission point and the calculated distance, the coordinates of the target point in the three-dimensional coordinate space are determined; the coordinate information of all points on the surface of the accessory is obtained, and then the accessory point cloud data is obtained.
3. The method for analyzing the quality of a fitting based on laser scanning data according to claim 1, wherein, In step (1), the pre-processing of the collected accessory point cloud data is as follows: Through a feature point matching algorithm, a plurality of matching point pairs in multi-angle point cloud data are obtained, an iterative closest point algorithm is used to calculate the rigid body change between the matching point pairs, the feature points are mapped and transformed, and the point cloud data collected at different angles are aligned; The multi-angle point cloud data after spatial alignment is spliced and fused to obtain complete accessory point cloud data; Through radius filtering, a radius range covering the complete structure of the accessory is determined, and outlier points beyond the range are removed; an adaptive point cloud density sampling method is used to downsample the accessory point cloud data.
4. The method for analyzing the quality of a fitting based on laser scan data according to claim 3, wherein, The adaptive point cloud density sampling method comprises the following steps: Calculate the Euclidean distance between the accessory point clouds by K-nearest neighbor algorithm, select K nearest neighbors for each accessory point cloud, and obtain a plurality of point cloud sets to capture the local geometric relationship between the accessory point clouds; Use a Gaussian kernel function to weight and sum the point clouds in the neighborhood to obtain the local density of each accessory point cloud and normalize it, use an exponential function to weight the normalized density to obtain a density adjustment factor; Adjust the distance values of each accessory point cloud and its K nearest neighbors by the density adjustment factor, calculate the smooth weighted correlation of each accessory point cloud according to the distance values, and select significant points; For each accessory point cloud, sort the global correlation scores of its K nearest neighbors, select the top M points with the highest scores as the final sampling result according to the set output point number M.
5. The method for analyzing the quality of a fitting based on laser scanning data according to claim 4, wherein, In step (2), the pre-processed accessory point cloud data is meshed and an accessory three-dimensional model is generated using point cloud processing software, which comprises the following steps: The pretreated point cloud data is imported into the CloudCompare point cloud processing software, and a meshing method based on greedy mapping triangulation is selected, and the resolution and smoothing parameters are set according to the overall density of the point cloud; Each three-dimensional point cloud in the accessory point cloud data is projected onto a plane by normal projection, and the projected point cloud in the plane is triangulated to obtain the connection relationship of each point cloud; A spatial region growing algorithm based on Delaunay is adopted, a sample triangle is selected as the initial surface, and the surface boundary is expanded to form a complete triangular mesh surface; The topological connection order of the point cloud in the three-dimensional space is determined according to the connection relationship between the accessory projection point clouds, and a triangular grid of a fixed size is formed according to the set resolution; In the software, select the filter tool, and perform smoothing processing on the triangular mesh according to the smoothing parameter through the Laplace smoothing algorithm to reduce the surface noise of the mesh; Through the quadrilateral simplification algorithm, the number of mesh vertices is reduced, and the holes in the mesh are filled to obtain a complete three-dimensional model of the accessory.
6. The method for analyzing the quality of a fitting based on laser scan data according to claim 1, wherein, In step (3), the multi-dimensional analysis includes size accuracy, shape deviation, surface quality and assembly gap analysis, wherein the surface quality analysis is realized by a pre-trained point cloud segmentation model, and the other dimensional analysis is realized by using big data analysis to obtain statistical indicators through key features extracted from the accessory three-dimensional model.
7. The method for analyzing the quality of a fitting based on laser scanning data according to claim 6, wherein, The size accuracy analysis obtains the standard size requirement of each accessory, calculates the deviation between the measured size feature data and the standard data, judges whether the deviation value is within the tolerance range, and calculates the statistical indicators of all accessory size data, including mean, standard deviation, maximum and minimum, to describe the overall data distribution; According to the data distribution, the quality level is divided according to the size distribution by using clustering algorithm, and the size quality level of each accessory is evaluated; The shape deviation analysis obtains the standard three-dimensional model of each accessory, imports the real-time collected accessory three-dimensional model into the three-dimensional modeling software, analyzes the shape difference between the models through model comparison, calculates the shape feature deviation, and obtains the position where the shape deviation is concentrated by statistical deviation distribution histogram; The assembly gap analysis focuses on the assembly gap between the accessories, establishes a regression-based mathematical model to describe the correlation between the assembly gap and the size feature, predicts the data change, and makes early warning for unqualified quality.
8. The method for analyzing the quality of a fitting based on laser scanning data according to claim 1, wherein, In step (4), the detailed steps of obtaining the pre-trained point cloud segmentation model include: Collect point cloud data of different types of defects on the surface of the accessory, label the type of each point cloud in the point cloud data, and divide the defect types and regions as a data set for training the point cloud segmentation model, and divide the data set into training set, validation set and test set; Establish a point cloud segmentation model, input the accessory defect point cloud data as a feature, map the point cloud data to a feature space through point cloud embedding, learn the correlation between point clouds by using a hierarchical point cloud attention module, capture defect shape and distribution features, and output point cloud level classification prediction results; The change process of the loss function is recorded during the model training process, the model performance is evaluated by using evaluation indexes, the model parameters are updated through back propagation, and when the loss function converges, the best model parameters are reserved as the pre-trained point cloud segmentation model for processing the real-time collected accessory point cloud data.
9. The method for analyzing the quality of a fitting based on laser scan data according to claim 8, wherein, The point cloud embedding adopts a kernel point convolution to extract information through related points and related features around the point cloud, an operation process of the kernel point convolution is expressed by the following equation: wherein, and respectively represent the input point cloud and the relevant points in the point cloud neighborhood, representing the local neighborhood of the input point cloud, representing the relevant features corresponding to each relevant point, expressing the color information of the point cloud; representing the kernel function, taking the relevant points centered on the input point cloud as input, calculating the linear correlation between the point clouds through the learnable core point set; by adding the displacement parameter in the kernel function, learning the position information between the points, and adaptively fitting the shape of each point cloud space; The hierarchical point cloud attention module adopts a hierarchical sampling strategy to filter the point cloud constituting the key vector, guides the query vector to collect long-range dependencies during attention calculation, divides the point cloud feature space into non-overlapping cubic windows, searches for close dense points and distant sparse points to jointly constitute the key vector in the self-attention mechanism, and participates in attention calculation.
10. A system for quality inspection of a fitting based on laser scanning data analysis, characterized by, The system is used to implement the accessory quality inspection method based on laser scanning data analysis according to any one of claims 1-9, and the system comprises: A data acquisition module: a laser scanning instrument is used to scan the accessory comprehensively, point cloud data on the surface of the accessory is collected, and the collected accessory point cloud data is preprocessed; A three-dimensional model construction module: for gridding the preprocessed accessory point cloud data, constructing a three-dimensional model of the accessory, and extracting key features of the accessory for quality inspection from the model; A data analysis module: for processing the key features of the accessory by using a multi-dimensional analysis method, and obtaining statistical data of the quality of the accessory by using big data analysis to verify the quality; A defect detection module: for inputting the preprocessed accessory point cloud data into the pre-trained point cloud segmentation model to obtain the identification result of the surface defects of the accessory, determining the defect type and calculating the defect rate index; A quality evaluation module: for comprehensively analyzing the big data analysis result and the identification result of the point cloud segmentation model, forming a complete accessory quality inspection evaluation report, and optimizing the accessory production process.