Quality inspection system for prefabricated components based on laser point cloud technology

Abstract

The present invention discloses a quality inspection system for prefabricated components based on laser point cloud technology, which relates to the technical field of quality inspection. The system comprises: fusing laser point cloud data of preset prefabricated components to generate preset laser point cloud models; collecting and fusing point cloud data of prefabricated components to be inspected to generate laser point cloud models to be tested; aligning and comparing the preset laser point cloud model with the laser point cloud model to be tested to generate abnormal point cloud distribution areas; screening edge point cloud data to generate abnormal screening point cloud areas; performing similarity recognition on abnormal screening point cloud areas to generate defect recognition results; and generating quality inspection reports. The present invention solves the technical problems in the prior art that it is difficult to accurately identify the defect types and locations of prefabricated components and the lack of precision in quality inspection, and achieves the technical effect of accurately identifying the defect types and locations and improving the quality inspection precision.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection, and in particular to a quality inspection system for assembled prefabricated components based on laser point cloud technology. Background Art

[0002] Prefabricated components are increasingly being used. During the production and installation of prefabricated components, various defects may occur, such as surface pores, pitting, surface cracks, etc. These defects will affect the quality and performance of the components. Therefore, it is very important to conduct quality inspection on prefabricated components. Traditional quality inspection methods mainly rely on manual inspection, which has large errors and is difficult to detect some minor defects.

[0003] The existing technology has the technical problems of difficulty in accurately identifying the defect types and locations of prefabricated components and lack of precision in quality inspection. Summary of the invention

[0004] The present application provides a prefabricated component quality inspection system based on laser point cloud technology, which is used to solve the technical problems in the prior art that it is difficult to accurately identify the defect types and locations of prefabricated components and the quality inspection lacks precision.

[0005] In view of the above problems, the present application provides an assembled prefabricated component quality inspection system based on laser point cloud technology.

[0006] The present application provides a prefabricated component quality inspection system based on laser point cloud technology, the system comprising:

[0007] A preset laser point cloud model generation module, the preset laser point cloud model generation module is used to perform laser point cloud data fusion on a preset prefabricated component under a preset acquisition environment to generate a preset laser point cloud model, wherein the preset prefabricated component is a quality defect-free sample; a laser point cloud model generation module to be tested, the laser point cloud model generation module to be tested is used to control a three-dimensional laser scanning device to perform point cloud data acquisition and fusion on the prefabricated component to be tested with the preset acquisition environment as a constraint to generate a laser point cloud model to be tested; an abnormal point cloud distribution area generation module, the abnormal point cloud distribution area generation module is used to compare the preset laser point cloud model with the The laser point cloud model to be tested is aligned and compared to generate an abnormal point cloud distribution area; a screening point cloud abnormal area generation module is used to screen the edge point cloud data of the abnormal point cloud distribution area to generate a screening point cloud abnormal area; a defect recognition result generation module is used to call a preset probability similarity distribution function, traverse a preset defect point cloud database to perform similarity recognition on the screening point cloud abnormal area, and generate a defect recognition result; a quality inspection report generation module is used to generate a quality inspection report of the prefabricated component to be tested based on the defect recognition result.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] The preset laser point cloud model generation module is used to fuse the laser point cloud data of the preset prefabricated components under the preset acquisition environment to generate the preset laser point cloud model; the test laser point cloud model generation module is used to control the three-dimensional laser scanning device to collect and fuse the point cloud data of the prefabricated components to be tested to generate the test laser point cloud model; the abnormal point cloud distribution area generation module is used to align and compare the preset laser point cloud model with the laser point cloud model to be tested to generate the abnormal point cloud distribution area; the screening point cloud abnormal area generation module is used to screen the edge point cloud data of the abnormal point cloud distribution area to generate the screening point cloud abnormal area; the defect recognition result generation module is used to call the preset probability similarity distribution function, perform similarity recognition on the screening point cloud abnormal area, and generate the defect recognition result; the quality inspection report generation module is used to generate the quality inspection report. The technical effect of accurately identifying the defect type and location and improving the quality inspection accuracy is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 A structural schematic diagram of an assembled prefabricated component quality inspection system based on laser point cloud technology is provided for an embodiment of the present application;

[0012] Figure 2 A schematic diagram of a flow chart of generating abnormal point cloud distribution areas by an abnormal point cloud distribution area generation module of an assembled prefabricated component quality inspection system based on laser point cloud technology is provided for an embodiment of the present application.

[0013] Explanation of the reference numerals: preset laser point cloud model generating module 10, to-be-tested laser point cloud model generating module 20, abnormal point cloud distribution area generating module 30, screening point cloud abnormal area generating module 40, defect recognition result generating module 50, quality inspection report generating module 60. DETAILED DESCRIPTION

[0014] The present application provides a prefabricated component quality inspection system based on laser point cloud technology to solve the technical problems in the prior art that it is difficult to accurately identify the defect types and locations of prefabricated components and the quality inspection lacks precision. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0015] like Figure 1 As shown, the present application provides a prefabricated component quality inspection system based on laser point cloud technology, the system comprising:

[0016] A preset laser point cloud model generation module 10 is used to perform laser point cloud data fusion on a preset prefabricated component under a preset acquisition environment to generate a preset laser point cloud model, wherein the preset prefabricated component is a quality defect-free sample.

[0017] Specifically, first of all, it is necessary to clearly define the preset acquisition environment, which includes specific lighting conditions, the location and parameter settings of the scanning equipment, and the control of interference factors in the surrounding environment. In such a strictly set acquisition environment, the preset prefabricated components are selected as quality defect-free samples. Use high-precision laser scanning equipment to scan this component in all directions. The laser scanning equipment will emit a large number of laser beams, which will be reflected back when they hit the surface of the component, thereby obtaining the position, distance and other information of many points on the surface of the component, forming a large amount of original point cloud data. Subsequently, the point cloud data fusion stage is entered, and the original point cloud data obtained from different angles and different positions are preliminarily preprocessed, including removing obvious noise points and outliers. Then, geometric transformation is used to accurately align and splice these data from different perspectives. In this process, the spatial transformation relationship between different data sets is calculated by finding common feature points or reference points to achieve seamless splicing. Then, the fused point cloud data is further optimized, and possible holes and missing parts are filled through interpolation, smoothing and other operations to make the point cloud model more complete and continuous. At the same time, the data is streamlined and compressed to reduce the amount of data without losing important information, thereby improving the efficiency of subsequent processing. After a series of rigorous and meticulous operations, the preset laser point cloud model was successfully generated. The preset laser point cloud model with extremely high accuracy and integrity was created, providing an accurate and reliable comparison benchmark for subsequent quality inspection.

[0018] The laser point cloud model generation module 20 to be tested is used to control the three-dimensional laser scanning device to collect and fuse point cloud data of the prefabricated components to be tested, using the preset collection environment as a constraint, and generate a laser point cloud model to be tested.

[0019] Specifically, the laser point cloud model generation module 20 to be tested works strictly in accordance with the detailed constraints established by the preset acquisition environment. These constraints not only cover the precise position and orientation of the laser scanning device in space, including its height, angle, and distance from the prefabricated component to be detected, but also include specific parameters of illumination, such as the type, intensity, angular distribution, and uniformity of the light source. In addition, the scanning resolution is finely set to determine the density and accuracy of the collected point cloud data, and the scanning frequency determines the speed and consistency of data acquisition. In actual operation, precise control instructions are sent to the three-dimensional laser scanning device to start and start working. The laser beam emitted by the device presents a highly dense and orderly distribution, fully irradiating the prefabricated component to be detected from all possible angles. These laser beams are reflected back after contacting the surface of the component, and their reflection time and intensity are accurately captured by the device, thereby obtaining a large amount of original point cloud data. These data contain multi-dimensional features such as the three-dimensional coordinates, color information, and reflectivity of each subtle point on the surface of the component. Subsequently, the point cloud data fusion processing stage is entered, and advanced filtering algorithms are used to remove those data points that are obviously abnormal and erroneous due to environmental interference, equipment noise and other factors. Then, through complex geometric transformation and registration algorithms, the point cloud data collected from different perspectives and at different times are accurately aligned. In this process, multiple common feature points or areas are found, and the spatial transformation relationship between each data set is calculated based on these reference points to achieve seamless splicing. For possible missing data, the interpolation algorithm is used to make reasonable inferences and fill in based on the existing data points in the surrounding area to ensure the integrity of the point cloud model. At the same time, in order to improve the quality of the model and reduce the amount of data, the data will be smoothed and streamlined to remove redundant information, so that the generated point cloud model is both accurate and efficient, and finally a laser point cloud model to be tested is generated that can comprehensively, accurately and meticulously reflect the true form and characteristics of the prefabricated components to be tested.

[0020] The abnormal point cloud distribution area generating module 30 is used to align and compare the preset laser point cloud model with the laser point cloud model to be tested to generate an abnormal point cloud distribution area.

[0021] Specifically, the abnormal point cloud distribution area generation module 30 is a key part of the entire system for accurately identifying and locating abnormal areas. It will carry out in-depth feature extraction work for the preset laser point cloud model, carefully select multiple representative points from the model, and comprehensively and meticulously extract the point cloud distribution features of these points. These features include the three-dimensional spatial coordinates of the point, color characteristics, density of the surrounding point clouds, surface normal vectors and other information. By extracting the features of many points, a point cloud feature distribution network that can fully reflect the distribution of the features of the preset model is constructed. Then, based on the constructed point cloud feature distribution network, according to the set criteria and rules for determining the same features, the network parts with similar features are merged. Through this merger, the originally complex network is divided into multiple point cloud network partitions with obvious identical features, and the features within each partition show a high consistency. Then, the generated multiple point cloud network partitions are compared with the corresponding areas of the laser point cloud model to be tested one by one and accurately. In the comparison process, the features of spatial position, color, density, normal vector and other aspects are compared in detail and the differences are calculated. When the feature difference of a certain area exceeds the preset acceptable range, it is marked as an abnormal area. All the areas marked as abnormal are combined to finally form the abnormal point cloud distribution area. By aligning and comparing the preset laser point cloud model with the laser point cloud model to be tested, the difference between the two can be accurately identified, thereby accurately determining the abnormal point cloud distribution area and improving the accuracy of anomaly detection.

[0022] The screening point cloud abnormal area generation module 40 is used to screen the edge point cloud data of the abnormal point cloud distribution area to generate a screening point cloud abnormal area.

[0023] Specifically, the abnormal point cloud area screening generating module 40 first obtains the abnormal point cloud distribution area, and for each point in the abnormal point cloud distribution area, calculates the distribution of points in its neighborhood to determine the curvature. For example, for a certain point, select a certain number of adjacent points (such as 10) around it, and calculate the curvature value of the surface at point P by fitting a local surface. Set a curvature threshold, such as 0.1. When the curvature value of point P is greater than this threshold, it is marked as a potential edge point, and the entire abnormal point cloud distribution area is traversed to complete the curvature calculation and marking of all points. The points marked as potential edge points are further analyzed to remove some isolated or misjudged points caused by noise. For example, if there are no other marked edge points around a potential edge point within a certain distance range, it is regarded as a noise point and removed. Finally, the screened point cloud abnormal areas are obtained, which accurately reflect the edge features of the abnormal areas. By screening the edge point cloud data in the abnormal point cloud distribution area, the real abnormal area can be defined more accurately, and some edge point clouds that may be misjudged as abnormal can be excluded, thereby improving the accuracy of abnormal area determination.

[0024] The defect recognition result generation module 50 is used to call a preset probability similarity distribution function, traverse a preset defect point cloud database to perform similarity recognition on the abnormal area of ​​the screened point cloud, and generate a defect recognition result.

[0025] Specifically, the defect identification result generation module 50 first triggers the call of the preset probability similarity distribution function. This function has a complex mathematical structure and calculation logic, and can perform similarity evaluation and probability calculation on the input data, and traverse the preset defect point cloud database. This database pre-stores defect point cloud data samples of various known types and degrees, covering a variety of possible defect modes. For the screened point cloud abnormal area, it is compared and analyzed in detail with each defect sample in the database. Through the probability similarity distribution function, the similarity between the abnormal area and each defect sample is calculated, and the corresponding probability value is generated. According to these probability values ​​and the preset judgment rules, the defect type and degree that are most similar to the screened point cloud abnormal area are determined, thereby generating the final defect identification result. By calling the preset probability similarity distribution function and traversing the preset defect point cloud database, the defect type corresponding to the screened point cloud abnormal area can be accurately identified, thereby improving the accuracy of defect identification.

[0026] The quality inspection report generating module 60 is used to generate a quality inspection report of the prefabricated component to be inspected based on the defect identification result.

[0027] Specifically, after receiving the defect identification results from the defect identification result generation module 50, these results are sorted and analyzed to extract key information, including but not limited to the type of defect (such as cracks, holes, deformation, etc.), the location of the defect (specific coordinates or locations on the prefabricated component), the degree of the defect (mild, moderate, severe, etc.) and the possible impact assessment on the overall performance of the component. Then, according to the preset report template and format requirements, these sorted information are filled in the corresponding positions. The report contains a detailed description of each defect, accompanied by charts, images or schematic diagrams to more intuitively display the situation of the defect. In addition, the overall quality of the prefabricated component will be comprehensively evaluated and graded according to the situation of the defect, which will refer to relevant industry standards and quality specifications to give clear quality judgment conclusions, such as "qualified", "unqualified but repairable", "unqualified and needs to be scrapped", etc. In the last part of the report, treatment suggestions and subsequent quality control measures for the detected defects are proposed to provide guidance and improvement direction for the production and construction process. The generated quality inspection report will be presented in a clear and easy-to-read form, which can be either an electronic document or a printout for relevant personnel to review and use.

[0028] In one possible implementation, Figure 2 As shown, the abnormal point cloud distribution area generation module 30 also includes:

[0029] The point cloud distribution features of multiple points in the preset laser point cloud model are extracted to generate a point cloud feature distribution network; networks with the same features are merged based on the point cloud feature distribution network to generate multiple point cloud network partitions; regional point cloud features are compared with the laser point cloud model to be tested according to the multiple point cloud network partitions to generate the abnormal point cloud distribution area.

[0030] Specifically, in the preset laser point cloud model, multiple key points will be selected for in-depth feature extraction. For the extraction of density features, a specific neighborhood range will be determined with each point as the center. In this neighborhood, the number of points is calculated and combined with the volume of the neighborhood to obtain the point cloud density value of the point. This density value reflects the density of points around the point. The extraction of the normal vector is determined by analyzing the local geometry around the point. The covariance matrix of the points in the neighborhood of the point is calculated, and then the eigenvalue decomposition is performed. The eigenvector corresponding to the minimum eigenvalue is the normal vector of the point. The normal vector indicates the approximate direction of the surface where the point is located. The extraction of curvature features is relatively complicated. Using methods such as quadratic surface fitting or principal component analysis, the point is centered, appropriate neighborhood points are selected for surface fitting, and then the curvature of the point is calculated based on the parameters of the fitting surface. Curvature describes the degree of curvature of the surface where the point is located. By extracting the above density, normal vector and curvature features from multiple points, and combining these feature information with the position information of the points, an interconnected network structure, namely the point cloud feature distribution network, is constructed. In this network, each point is regarded as a node, and the connection between nodes represents their relationship in spatial position and features.

[0031] Based on the constructed point cloud feature distribution network, the same features are identified and merged, and a series of standards and rules for determining the same features are defined. For density features, if the difference in the average density values ​​of points in two regions is within a certain small range, they are considered to have similar density features; for normal vector features, if the average deviation of the normal vector directions of points in two regions is less than a certain angle threshold, they are considered to have similar normal vector features; for curvature features, a curvature value difference threshold is also set to judge the similarity. Then, according to these standards and rules, the entire point cloud feature distribution network is traversed and analyzed, and adjacent areas with similar density, normal vector and curvature features are gradually merged. During the merging process, the connection relationship and feature information of the network will be updated to ensure that the merged area can accurately reflect its comprehensive features. By repeating this identification and merging process, the point cloud feature distribution network is eventually divided into multiple relatively independent areas with relatively consistent internal features, that is, multiple point cloud network partitions.

[0032] The multiple previously generated point cloud network partitions are accurately matched with the laser point cloud model to be tested, which means accurately finding the corresponding position and area of ​​each point cloud network partition in the model to be tested. Then, for each corresponding area pair, a comprehensive and in-depth point cloud feature comparison is performed. For density features, the difference in point cloud density at corresponding positions in the two areas is calculated; for normal vector features, the direction and angle deviation of the average normal vector are compared; for curvature features, the difference in curvature values ​​is evaluated. During the comparison process, a specific deviation tolerance threshold is set for each feature. If the difference in density, normal vector or curvature of a region exceeds the corresponding threshold, the region is marked as part of the anomaly. By performing such rigorous comparison and marking of all point cloud network partitions and areas to be tested, all areas marked as abnormal are integrated and summarized, and these integrated abnormal areas constitute the final abnormal point cloud distribution area.

[0033] In a possible implementation, the abnormal point cloud distribution area generating module 30 further includes:

[0034] Based on the point cloud feature distribution network, the density features, normal directions and curvature features of adjacent points are compared for deviations to generate point cloud density deviation, normal direction deviation and curvature deviation; if the point cloud density deviation, the normal direction deviation and the curvature deviation are all within the preset deviation range, the adjacent points are network merged to generate the multiple point cloud network partitions.

[0035] Specifically, in the constructed point cloud feature distribution network, adjacent pairs of points are selected one by one. For each pair of adjacent points, their density features, normal directions and curvature features are analyzed and compared separately. For the density feature, the point cloud density of each point is first calculated. The point cloud density can be measured by the number of points within a certain range around the point. Then, the difference between the point cloud density values ​​of two adjacent points is calculated. This difference is the point cloud density deviation. For the normal direction, the normal vector of each point is determined. The normal vector represents the vertical direction of the surface where the point is located. The normal direction deviation is obtained by calculating the angle between the normal vectors of adjacent points. If the angle is 0 degrees, it means that the normal directions are completely consistent; the larger the angle, the greater the normal direction deviation. For the curvature feature, the curvature value of each point is calculated. The curvature reflects the degree of curvature of the surface where the point is located. Similarly, the difference in the curvature values ​​of adjacent points is calculated to obtain the curvature deviation. By performing such calculations and comparisons on a large number of adjacent points, a series of point cloud density deviation values, normal direction deviation values, and curvature deviation values ​​are generated.

[0036] After completing the calculation of the point cloud density deviation, normal direction deviation and curvature deviation of adjacent points, these deviation values ​​will be compared with the preset deviation range set in advance. The preset deviation range is determined based on factors such as specific detection requirements, accuracy requirements and the characteristics of the detected object. If the calculated point cloud density deviation of a pair of adjacent points is less than or equal to the preset point cloud density deviation threshold, and the normal direction deviation is less than or equal to the preset normal direction deviation threshold, and the curvature deviation is also less than or equal to the preset curvature deviation threshold, then it is considered that the two adjacent points have sufficient similarity in density, normal direction and curvature characteristics. When it is determined that the adjacent points have similar characteristics, they will be merged into a network. Specifically, the representations of the two points in the point cloud feature distribution network are connected and integrated to make them a unified part. By judging and merging all adjacent points according to such rules, adjacent points with similar characteristics are gradually combined together, and finally multiple relatively independent areas with relatively consistent internal characteristics are formed, that is, multiple point cloud network partitions.

[0037] In a possible implementation, the abnormal point cloud distribution area generating module 30 further includes:

[0038] The laser point cloud model to be tested is aligned and divided according to the multiple point cloud network partitions to generate multiple aligned partitions of the laser point cloud model to be tested, and a mapping relationship between the multiple aligned partitions and the multiple point cloud network partitions is established; based on the mapping relationship, a first point cloud network partition and a first aligned partition are extracted; similarity comparison and fusion of point cloud density features, normal directions and curvature features of the first point cloud network partition and the first aligned partition are performed to generate a first similarity index; if the first similarity index is less than a preset similarity, the first aligned partition is added to the abnormal point cloud distribution area.

[0039] Specifically, first, obtain multiple previously generated point cloud network partitions. These partitions have specific spatial position, density distribution, normal vector and curvature and other feature information, as well as the laser point cloud model to be tested. Calculate the feature descriptor for each point cloud network partition, use the fast point feature histogram (FPFH) as the feature descriptor, and calculate the FPFH features of each point in each partition to obtain a comprehensive feature vector to describe the features of the partition. The same operation is performed on the laser point cloud model to be tested, calculate the FPFH features of each point, and preliminarily divide the point cloud model into multiple potential partitions based on these features. Then, by comparing the feature descriptor of each potential partition with the feature descriptor of the point cloud network partition, find the most similar match. For a potential partition in the point cloud model to be tested, calculate the Euclidean distance or other similarity measure between its feature descriptor and the feature descriptor of each point cloud network partition, and the point cloud network partition with the smallest distance or the highest similarity is considered to be its corresponding match. After finding the match, the potential partition is further adjusted and optimized according to the matching result, so that it is closer to the corresponding point cloud network partition in shape, size and feature distribution, thereby generating the final aligned partition. At the same time, a clear mapping relationship is established. For example, if the feature descriptor of partition A of the point cloud network matches the feature descriptor of partition B in the point cloud model to be tested successfully, the mapping relationship between A and B is established.

[0040] The mapping relationship between the point cloud network partition and the aligned partition of the laser point cloud model to be tested has been successfully established. This mapping relationship clearly shows the correspondence between each point cloud network partition and the corresponding aligned partition in the laser point cloud model to be tested. When extracting the first point cloud network partition and the first aligned partition, the first here can be understood as the first partition determined according to a certain pre-set order or rule. Assuming that the order is determined from small to large according to the partition number, the first point cloud network partition is the point cloud network partition numbered 1. By querying the previously established mapping relationship, the corresponding partition in the laser point cloud model to be tested is found. This partition is the first aligned partition. In this way, based on the clear mapping relationship that has been established, we can accurately and quickly extract the first point cloud network partition and the first aligned partition we need for subsequent processing and analysis.

[0041] The specific data of the point cloud density feature, normal direction feature and curvature feature of the first point cloud network partition and the first aligned partition are obtained respectively. For the point cloud density feature, the point cloud density values ​​of the corresponding positions in the two partitions are calculated, and the degree of difference between them is calculated. For example, the difference or ratio of the density values ​​can be calculated to measure the similarity of the density features. For the normal direction feature, the directions of the normal vectors of the corresponding points in the two partitions are compared, and the difference in direction is quantified by calculating the angle between the normal vectors. The smaller the angle, the more similar the normal directions are. For the curvature feature, the curvature values ​​of the corresponding positions in the two partitions are also calculated, and their proximity is analyzed. Then, the similarity data of the point cloud density, normal direction and curvature features are fused. The fusion method is to give a certain weight to the similarity of each feature, and then comprehensively calculate to obtain an overall similarity value. Through such processing, a specific indicator is used to quantify the overall similarity between the first point cloud network partition and the first aligned partition in these key features.

[0042] The first similarity index has been calculated, which reflects the similarity between the first point cloud network partition and the first aligned partition. At the same time, there is a preset similarity value that is set in advance, which serves as a standard threshold for judgment. The calculated first similarity index is compared with the preset similarity. If the first similarity index is less than the preset similarity, it means that there is a large difference between the first aligned partition and the expected standard (that is, the standard represented by the first point cloud network partition). Due to this significant difference, it is judged that the first aligned partition does not meet the normal standard range, so it is added to the abnormal point cloud distribution area, which can accurately identify the area in the laser point cloud model to be tested that is significantly different from the preset standard, thereby improving the accuracy and reliability of anomaly detection.

[0043] In a possible implementation, the screening point cloud abnormal area generation module 40 further includes:

[0044] Obtain a mapping area of ​​the abnormal point cloud distribution area in the preset prefabricated component, and obtain a point cloud distribution rule within the mapping area; perform edge point cloud comparison on the abnormal point cloud distribution area according to the point cloud distribution rule to generate a point cloud with consistent edge distribution; delete the point cloud with consistent edge distribution from the abnormal point cloud distribution area to generate the screened point cloud abnormal area.

[0045] Specifically, when obtaining the mapping area of ​​the abnormal point cloud distribution area in the preset prefabricated component, the spatial coordinate conversion algorithm is used to determine the three-dimensional coordinates of each point in the abnormal point cloud area. Then, these coordinates are matched with the overall point cloud coordinate system of the preset prefabricated component to find the exact position of the abnormal point cloud in the prefabricated component point cloud, thereby delineating the mapping area. For the acquisition of the density distribution rule of the point cloud in the mapping area, a certain volume is used as a unit, such as a small cube, and the number of points in this small cube is calculated to obtain the local point cloud density value. A large number of such small cube calculations are performed on the entire mapping area to generate a series of density values. By analyzing the distribution of these density values, it is determined whether there are areas with density significantly higher or lower than the average value, as well as the gradient of density changes. When analyzing the normal direction characteristics, for each point in the mapping area, the positional relationship of the surrounding points is calculated, and the normal vector of the point is determined using mathematical methods (such as principal component analysis). Then, the normal vectors of all points are statistically and visually displayed. Observe whether the direction of the normal vector shows a certain degree of aggregation or dispersion, and whether there is an area with a local direction mutation. For curvature features, the surface fitting method is used to select an appropriate set of points in the mapping area, use a mathematical function (such as a quadratic surface) to fit the local surface formed by these points, and calculate the curvature value of each point based on the parameters of the fitting surface. These curvature values ​​are statistically analyzed to find areas with larger or smaller curvature values ​​and locations where the curvature values ​​change dramatically.

[0046] The distribution area of ​​abnormal point clouds has been determined. However, in order to more accurately determine the real abnormal part, it is necessary to conduct a more detailed analysis of the edge of this area. According to the previously obtained point cloud distribution rules, including density rules (such as the average point cloud density value within a specific range, density change trend, etc.), normal direction rules (such as the common normal direction range, the smoothness of direction change, etc.) and curvature feature rules (such as reasonable curvature value interval, the law of curvature change, etc.), extract the point cloud data at the edge of the abnormal point cloud distribution area. These edge point clouds are a collection of point clouds within a certain distance from the boundary of the abnormal area. Next, the density, normal direction and curvature characteristics of these edge point clouds are compared in detail with the corresponding rules. For example, for a certain edge point cloud, calculate the point cloud density of the local area where it is located. If the density value is within the range of normal density; then analyze its normal direction and find that it is consistent with the normal direction change trend of the normal area; at the same time, check its curvature characteristics, which also conform to the normal curvature feature rules. Then, mark this point cloud as a point cloud with consistent edge distribution. By performing such comparison and marking on all edge point clouds, a set of point clouds with consistent edge distribution can be generated.

[0047] After completing the comparison of the edge point clouds of the abnormal point cloud distribution area and determining the point clouds with consistent edge distribution, the next step is to perform a deletion operation to accurately identify and mark all point clouds that are determined to have consistent edge distribution. Then, from the point cloud set contained in the abnormal point cloud distribution area, these marked point clouds with consistent edge distribution are removed one by one. Through such deletion processing, the area composed of the remaining point clouds is the screened point cloud abnormal area. This screened point cloud abnormal area is more accurate than the abnormal point cloud distribution area that was initially determined before, and removes those edge point clouds that were originally misjudged as abnormal but actually meet the normal point cloud distribution rules.

[0048] In a possible implementation, the defect identification result generating module 50 further includes:

[0049] Call a preset probability similarity distribution function, calculate the point cloud similarity between the screened point cloud abnormal area and the first defect area sample in the preset defect point cloud database, and obtain a first probability similarity divergence; determine whether the probability similarity divergence is less than a preset divergence threshold, and if so, add the defect type corresponding to the first defect area sample and the position features of the screened point cloud abnormal area into the defect identification result.

[0050] Specifically, the preset probability similarity distribution function is a complex mathematical model that has been carefully designed and constructed. Its purpose is to accurately quantify the similarity between two point cloud areas. When calculating, this function will first perform comprehensive and detailed feature extraction on the samples of the abnormal area and the first defect area of ​​the screened point cloud. In terms of the density distribution of the point cloud, it will divide the two areas into multiple small sub-areas, and then calculate the number of point clouds in each sub-area and form a density distribution curve. For the spatial position relationship of the points, the function will calculate the distance, angle and other parameters of each point relative to the center point of the area, and construct the corresponding position matrix. In terms of the distribution of the normal direction, the normal direction of each point will be measured, and the difference angle of the normal direction at the same position will be counted. As for the change in curvature, the curve of the surface composed of the point cloud will be fitted by mathematical methods, and then the curvature value and its rate of change will be calculated. Next, for the density distribution curve, the function will calculate the mean square error between the two curves; for the position matrix, it will calculate the difference norm between the matrices; for the difference angle of the normal direction, it will calculate its mean value and standard deviation; for the curvature value and its rate of change, it will calculate the correlation coefficient between them. After completing the difference calculation of these individual features, the function will not simply add or average these difference values ​​directly, but will assign different weights according to the importance of these features in judging similarity. The weights are determined based on a large amount of previous experimental data and statistical analysis to ensure that the final similarity assessment is more accurate and reliable. Then, by comprehensively considering all weighted feature differences and applying probability statistics, these difference values ​​are converted into a value between 0 and 1, which is the first probability similarity divergence. If this divergence value is extremely close to 0, it means that after the above comprehensive and detailed feature comparison and calculation, the two regions show a high degree of consistency in almost all important features. This strongly suggests that the abnormal area of ​​the screened point cloud is very similar in nature to the first defect area sample, which greatly increases the possibility that the screened area belongs to the same type of defect as the first defect area sample. For example, when inspecting the prefabricated components of aircraft engine blades, if the first probability similarity divergence obtained by the screened abnormal point cloud area after such a detailed comparison process with a certain fatigue crack defect area sample in the database is close to 0, there is strong evidence that this abnormal area is likely to be a fatigue crack defect.

[0051] The calculated probability similarity divergence is compared with the preset divergence threshold. If the probability similarity divergence is less than the preset divergence threshold, it means that the similarity between the screened point cloud abnormal area and the first defect area sample has reached a high level, which is enough to consider that they are likely to be the same type of defects. In this case, the following operations will be taken: obtain the defect type corresponding to the first defect area sample, such as "cracks", "holes", "wear" and other specific defect descriptions. At the same time, determine the position characteristics of the screened point cloud abnormal area in the entire inspection object, including its coordinate range in three-dimensional space, positional relationship relative to a certain reference point and other detailed information. Then, the defect type corresponding to the first defect area sample and the position characteristics of the screened point cloud abnormal area are added to the defect recognition result. In this way, the final defect recognition result can clearly indicate where and what type of defects are found.

[0052] In a possible implementation, the defect identification result generating module 50 further includes: ;in, To characterize the probability similarity divergence between the first defect area sample and the abnormal area of ​​the screening point cloud, Filter point cloud blocks in abnormal areas With point cloud block The similarity probability of is the point cloud block in the first defect area sample With point cloud block The similarity probability.

[0053] Specifically, Characterizes the probability similarity divergence between the first defect area sample and the abnormal area of ​​the screening point cloud. To filter point cloud blocks in abnormal areas With point cloud block This means that in the screening point cloud abnormal area, the point cloud block is calculated With point cloud block The probability of similarity between . is the point cloud block in the first defect area sample With point cloud block The similarity probability of point cloud blocks is With point cloud block The probability measure of similarity. By calculating the value of this expression, we can get a quantitative probability similarity divergence , which is used to measure the overall similarity between the first defect area sample and the abnormal area of ​​the screened point cloud. When the value of is small, it means that the two regions are close in the probability of similarity of the point cloud blocks, that is, they may have a high similarity; on the contrary, when When the value of is large, it indicates that the similarity between the two regions is low. Such a preset probability similarity distribution function can help the system to more accurately identify and judge the defect types in prefabricated components.

[0054] In a possible implementation, the defect identification result generating module 50 further includes:

[0055] The preset defect point cloud database is formed by collecting defect area samples, wherein the defect area samples include defect areas corresponding to preset defect types; regional point cloud distribution data of each defect area is obtained with the preset collection environment as a constraint, the defect area is identified, and the preset defect point cloud database is generated.

[0056] Specifically, the construction of the preset defect point cloud database is achieved by collecting a series of defect area samples, which cover specific areas corresponding to the preset defect types, such as surface pores, pitting, surface cracks and other types of defect areas of prefabricated components. When collecting these defect area samples, the preset acquisition environment is used as a constraint condition, which means that under the same acquisition environment setting, the regional point cloud distribution data of each defect area is obtained. The point cloud distribution data contains information such as the position, density, and normal direction of each point in the defect area, and can accurately describe the geometric shape and characteristics of the defect area. At the same time, each defect area is identified, and the content of the identification includes information such as the name of the defect type, the specific feature description of the defect, and the severity of the defect. Through these identifications, the defect area can be associated with the corresponding defect information. Finally, according to the point cloud distribution data and identification information of the collected defect area samples, a preset defect point cloud database is generated. This database stores sample data of various defect types, so that in the subsequent quality inspection process, it can be compared and analyzed with the point cloud data of the prefabricated component to be tested, so as to accurately identify the possible defect types and locations in the component. For example, when constructing a preset defect point cloud database, multiple area samples with surface pore defects were collected, and their point cloud distribution data were obtained under the preset collection environment. At the same time, these defect areas were marked, indicating that they were surface pore defects and other related information. Then, these data were integrated to form a part of the preset defect point cloud database containing surface pore defect samples. The same process is also applicable to sample collection and database construction for other preset defect types such as pockmarks and surface cracks. The database contains defect area samples corresponding to the preset defect types, which can provide accurate reference standards for detection, so that similar defects can be more accurately identified when detecting prefabricated components.

[0057] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0058] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0059] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. The quality inspection system for prefabricated components based on laser point cloud technology is characterized by: include: A preset laser point cloud model generation module, wherein the preset laser point cloud model generation module is used to perform laser point cloud data fusion on a preset prefabricated component under a preset acquisition environment to generate a preset laser point cloud model, wherein the preset prefabricated component is a quality defect-free sample; A laser point cloud model generation module to be tested, the laser point cloud model generation module to be tested is used to control the three-dimensional laser scanning device to collect and fuse point cloud data of the prefabricated component to be tested, using the preset collection environment as a constraint, and generate a laser point cloud model to be tested; An abnormal point cloud distribution area generation module, the abnormal point cloud distribution area generation module is used to align and compare the preset laser point cloud model with the laser point cloud model to be tested to generate an abnormal point cloud distribution area; A screening point cloud abnormal area generation module is used to screen edge point cloud data of the abnormal point cloud distribution area to generate a screening point cloud abnormal area; A defect recognition result generation module, wherein the defect recognition result generation module is used to call a preset probability similarity distribution function, traverse a preset defect point cloud database to perform similarity recognition on the abnormal area of ​​the screened point cloud, and generate a defect recognition result; A quality inspection report generating module, the quality inspection report generating module is used to generate a quality inspection report of the prefabricated component to be inspected based on the defect identification result; The screening point cloud abnormal area generation module is used for: Obtaining a mapping area of ​​the abnormal point cloud distribution area in the preset prefabricated component, and obtaining a point cloud distribution rule in the mapping area; Performing edge point cloud comparison on the abnormal point cloud distribution area according to the point cloud distribution rule to generate a point cloud with consistent edge distribution; The edge distribution consistent point cloud is deleted from the abnormal point cloud distribution area to generate the screened point cloud abnormal area.

2. The prefabricated component quality inspection system based on laser point cloud technology according to claim 1, characterized in that: The abnormal point cloud distribution area generation module is used for: Extracting point cloud distribution features of multiple points on the preset laser point cloud model to generate a point cloud feature distribution network; Based on the point cloud feature distribution network, networks with the same features are merged to generate multiple point cloud network partitions; The regional point cloud features are compared with the laser point cloud model to be tested according to the multiple point cloud network partitions to generate the abnormal point cloud distribution area.

3. The prefabricated component quality inspection system based on laser point cloud technology according to claim 2, characterized in that: The abnormal point cloud distribution area generation module is used for: Based on the point cloud feature distribution network, density features, normal directions and curvature features of adjacent points are compared for deviation, and point cloud density deviation, normal direction deviation and curvature deviation are generated; If the point cloud density deviation, the normal direction deviation and the curvature deviation are all within a preset deviation range, the adjacent points are network-merged to generate the multiple point cloud network partitions.

4. The prefabricated component quality inspection system based on laser point cloud technology according to claim 3 is characterized in that: The abnormal point cloud distribution area generation module is used for: Aligning and dividing the laser point cloud model to be measured according to the multiple point cloud network partitions, generating multiple aligned partitions of the laser point cloud model to be measured, and establishing a mapping relationship between the multiple aligned partitions and the multiple point cloud network partitions; Extracting a first point cloud network partition and a first alignment partition based on the mapping relationship; Performing similarity comparison and fusion of point cloud density features, normal direction and curvature features on the first point cloud network partition and the first alignment partition to generate a first similarity index; If the first similarity index is less than a preset similarity, the first alignment partition is added to the abnormal point cloud distribution area.

5. The prefabricated component quality inspection system based on laser point cloud technology according to claim 1, characterized in that: The defect identification result generation module is used for: Calling a preset probability similarity distribution function, performing point cloud similarity calculation on the screened point cloud abnormal area and the first defect area sample in the preset defect point cloud database, and obtaining a first probability similarity divergence; It is determined whether the probability similarity divergence is less than a preset divergence threshold. If so, the defect type corresponding to the first defect area sample and the position feature of the abnormal area of ​​the screening point cloud are added to the defect identification result.

6. The prefabricated component quality inspection system based on laser point cloud technology according to claim 5, characterized in that: The defect identification result generation module is used for: ； in, To characterize the probability similarity divergence between the first defect area sample and the abnormal area of ​​the screening point cloud, To filter out point cloud blocks in abnormal areas With point cloud block The similarity probability of is the point cloud block in the first defect area sample With point cloud block The similarity probability.

7. The prefabricated component quality inspection system based on laser point cloud technology according to claim 6, characterized in that: The preset defect point cloud database is formed by collecting defect area samples, wherein the defect area samples include defect areas corresponding to preset defect types; regional point cloud distribution data of each defect area is obtained with the preset collection environment as a constraint, the defect area is identified, and the preset defect point cloud database is generated.

Images