A point cloud curvature discrimination and anti-noise fitting main pump three-dimensional modeling measurement method

Abstract

The application belongs to the technical field of three-dimensional modeling measurement, and provides a main pump three-dimensional modeling measurement method for point cloud curvature discrimination and noise-resistant fitting, comprising the following steps: continuously scanning high-density on key surfaces of a main pump and a suction conduit to obtain original surface point cloud data; constructing a point cloud space topology structure, calculating characteristic values of curvatures of each point and local normal vectors, discriminating pseudo-feature noise points, and obtaining pure metal reference point cloud; adopting an improved random sample consensus algorithm with local normal vector direction constraint to fit a high-precision flange plane and a cylindrical axis in a noise-resistant manner; and establishing a coordinate system with the plane and the axis as references to complete simulation installation of the main pump and the suction conduit and interference amount calculation. The application can automatically filter micro defects, avoid magnification of small error levers, ensure that modeling precision meets installation requirements, eliminate equipment scratching damage in actual installation, and ensure nuclear power installation safety.

Description

Technical Field

[0001] This invention belongs to the field of 3D modeling and measurement technology, specifically a 3D modeling and measurement method for a main pump with point cloud curvature discrimination and noise-resistant fitting. Background Technology

[0002] The CAP1400 reactor uses a shielded coolant pump (referred to as the main pump). During installation, the main pump impeller and guide vanes must be installed with high precision inside the suction duct. Due to the extremely small installation gap (concentricity deviation requirement <0.19mm), the extremely high installation accuracy requirement, and the obstructed field of vision, conventional measuring tools cannot meet the requirements.

[0003] In actual nuclear power plant engineering sites, existing 3D modeling and measurement methods for main pumps have hidden and fatal technical defects: First, discrete sampling easily introduces pseudo-feature point errors. After transportation and on-site storage, the flange and sealing surfaces of the main pump and suction duct may have slight unevenness in anti-rust coating, oxide layer accumulation, and fine machining marks and scratches. Existing methods use a small number of discrete points for fitting, which easily collects these pseudo-surface data instead of the real metal reference surface. Second, fixed rejection thresholds lead to the absorption and leverage amplification of micro-errors. The fixed error rejection rules of existing methods allow pseudo-feature point errors of 0.05mm to 0.1mm. After this tiny error is absorbed by the fitting algorithm, it will cause a slight angular offset of the main pump's cylindrical axis. Extending along the main pump several meters high, the error at the top impeller will be amplified to the millimeter level, which will lead to metal-to-metal scraping between the impeller and the suction duct during actual installation, causing equipment damage and nuclear safety hazards.

[0004] Therefore, there is an urgent need for a three-dimensional modeling and measurement method for the main pump that can automatically identify the surface coating and micro-defects of the filter and prevent the algorithm from absorbing and amplifying tiny errors, so as to ensure the safe installation of the main pump under extremely narrow gaps.

[0005] Therefore, this invention provides a method for three-dimensional modeling and measurement of a main pump using point cloud curvature discrimination and noise-resistant fitting. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is: In one aspect, the present invention provides a method for three-dimensional modeling and measurement of a main pump using point cloud curvature discrimination and noise-resistant fitting, comprising: Step 1: Using a laser tracker in conjunction with a spatial scanning probe, perform high-density continuous scanning of the flange surface, sealing surface, and cylindrical mating surface of the main pump and suction duct to obtain raw surface point cloud data; Step 2: Mesh the original surface point cloud data to construct a spatial topology. Based on the spatial topology, calculate the curvature feature values ​​of each target data point and its local neighborhood in the original surface point cloud data, and extract the local normal vector of each target data point. Step 3: Calculate the spatial angle between the local normal vector of each target data point and the average normal vector of its local neighborhood. Set the local curvature discrimination threshold and the normal vector angle threshold. Target data points whose curvature feature value exceeds the local curvature discrimination threshold or whose spatial angle exceeds the normal vector angle threshold are identified as pseudo-feature noise points and removed from the original surface point cloud data to obtain a purified metal reference point cloud. Step 4: The random sampling consensus algorithm is used to perform noise-resistant fitting on the purified metal reference point cloud, and the local normal vector of the purified metal reference point cloud is used as the directional constraint condition to construct a high-precision flange plane and cylinder axis. Step 5: Establish a coordinate system based on the high-precision flange plane and cylindrical axis, and perform simulated installation and interference calculation of the main pump and suction pipe.

[0008] As a further improvement of the present invention, the specific process for obtaining the original surface point cloud data is as follows: Using pre-deployed spatial control points, a global measurement coordinate system is established for the laser tracker, unifying the laser tracker and the spatial scanning probe into this global measurement coordinate system. The spatial scanning probe scans the flange face, sealing face, and impeller outer cylindrical mating surface of the main pump body, as well as the flange face, inner and outer cylindrical mating surfaces of the suction duct, at a certain scanning frame rate and scanning point spacing, respectively, to obtain the original surface point cloud data.

[0009] As a further improvement of the present invention, the specific calculation process of the curvature characteristic value is as follows: The Kd-tree algorithm or Delaunay triangulation algorithm is used to construct a spatial topology for the original surface point cloud data. Within this topology, a search radius for local neighborhoods is set. For any target data point in the original surface point cloud data, all neighboring points within the search radius are extracted to form a local neighborhood point set. The geometric center of this local neighborhood point set is calculated, and a 3×3 covariance matrix is ​​constructed. Eigenvalue decomposition is performed on the 3×3 covariance matrix to obtain three non-negative eigenvalues ​​and eigenvectors. Finally, the curvature eigenvalues ​​of the target data point are calculated. λ1, λ2, and λ3 are three non-negative eigenvalues, and λ1 ≥ λ2 ≥ λ3.

[0010] As a further improvement of the present invention, the specific process of extracting the local normal vectors of each target data point is as follows: Extract the eigenvector corresponding to the smallest eigenvalue, and define it as the local normal vector of the target data point.

[0011] As a further improvement of the present invention, the specific process for determining a point as a pseudo-feature noise point is as follows: The local normal vectors of all neighboring points within the local neighborhood of the target data point are summed and normalized to obtain the average normal vector of the local neighborhood. The spatial angle between the local normal vector of the target data point and the average normal vector is calculated. A local curvature discrimination threshold and a normal vector angle threshold are set. If the curvature feature value of the target data point is greater than the local curvature discrimination threshold, or the spatial angle is greater than the normal vector angle threshold, then the target data point is determined to be a pseudo-feature noise point.

[0012] As a further improvement of the present invention, the specific process of obtaining the purified metal reference point cloud is as follows: All target data points marked as pseudo-feature noise points are stripped from the original surface point cloud data, and the remaining set of target data points is the purified metal reference point cloud.

[0013] As a further improvement of the present invention, the specific process of constructing a high-precision flange plane is as follows: The purified metal reference point cloud is segmented into a planar point cloud subset; in the planar point cloud subset, the Random Sampling Consensus Algorithm (RANSAC) is used for sampling iteration; in each iteration, three target data points are randomly selected as the minimum sample set, and a candidate flange plane equation model and its theoretical normal vector are calculated; the angle between the local normal vector of these three target data points and the theoretical normal vector is calculated respectively, and a plane parallel angle tolerance threshold is set. If the angle between the local normal vector and the theoretical normal vector of any target data point is less than or equal to the plane parallel angle tolerance threshold, then the minimum sample set is determined to be valid. If the minimum sample set is valid, the orthogonal distance from other target data points to the candidate flange plane is calculated, and the angle between their local normal vector and the theoretical normal vector is calculated. A distance tolerance threshold is set. When the orthogonal distance is less than the distance tolerance threshold, and the angle between the local normal vector and the theoretical normal vector is less than the plane parallel angle tolerance threshold, the target data point is determined to be an interior point. The candidate flange plane equation model with the most interior points is retained, and the least squares method is used to perform plane fitting on all interior points to output a high-precision flange plane.

[0014] As a further improvement of the present invention, the specific process of constructing a high-precision cylindrical axis is as follows: The purified metal reference point cloud is divided into a cylindrical point cloud subset; in the cylindrical point cloud subset, the Random Sampling Consensus Algorithm (RANSAC) is used for sampling iteration; in each iteration, the minimum number of target data points that can determine the cylindrical model are randomly selected as the minimum sample set, and combined with its local normal vector direction, a candidate cylindrical model is constructed, and the candidate cylindrical axis vector of the candidate cylindrical model is obtained. Calculate the angle between the local normal vector of each target data point in the minimum sample set and the candidate cylinder axis vector, and set a cylinder vertical angle tolerance threshold; if the angle between the local normal vector of all target data points and the candidate cylinder axis vector is within the cylinder vertical angle tolerance threshold range, then the minimum sample set is determined to be valid. If the minimum sample set is valid, calculate the distance error from other target data points to the candidate cylindrical model, and calculate the angle between its local normal vector and the candidate cylindrical axis vector; set a distance tolerance threshold; when the distance error is less than the distance tolerance threshold, and the angle between the local normal vector and the candidate cylindrical axis vector satisfies a perpendicular relationship, the target data point is determined as an interior point; retain the candidate cylindrical model with the most interior points, and use the least squares method to perform cylindrical fitting on all interior points of the model to output a high-precision cylindrical axis.

[0015] As a further improvement of the present invention, the specific process of simulating the installation of the main pump and the suction tubing and calculating the interference is as follows: For the inhalation catheter, the high-precision cylindrical axis is defined as the reference Z1 axis, the spatial intersection of the high-precision flange plane and the Z1 axis is defined as the origin O1 of the reference coordinate system, and the high-precision flange plane is used as the X1-Y1 reference plane, thereby establishing the local reference coordinate system O1-X1Y1Z1 for the inhalation catheter. For the main pump, the high-precision cylindrical axis is defined as the assembly Z2 axis, and the spatial intersection of the high-precision flange plane and the Z2 axis is defined as the origin O2 of the assembly coordinate system. The high-precision flange plane is used as the X2-Y2 assembly plane, thereby establishing the local assembly coordinate system O2-X2Y2Z2 of the main pump. Using a rigid body transformation matrix, the local reference coordinate system O1-X1Y1Z1 of the inhalation tubing and the local assembly coordinate system O2-X2Y2Z2 of the main pump are transformed into the global measurement coordinate system; Calculate the radial distance in space between the main pump assembly Z2 axis and the suction pipe reference Z1 axis, and take the maximum value as the final concentricity deviation value. Calculate the spatial dihedral angle between the high-precision flange plane of the main pump and the high-precision flange plane of the suction pipe, and convert it into the maximum opening gap at the flange edge as the parallelism deviation value. The calculated concentricity deviation value is compared with the preset limit gap threshold. If the concentricity deviation value is greater than or equal to the limit clearance threshold, it is determined that the outer cylindrical mating surface of the main pump impeller will cause metal interference and scraping with the inner cylindrical mating surface of the suction duct. If the concentricity deviation is less than the limit gap threshold and the parallelism deviation meets the sealing requirements, then the installation is considered successful.

[0016] On the other hand, the present invention provides a main pump 3D modeling and measurement system for point cloud curvature discrimination and noise-resistant fitting, comprising: Point cloud scanning acquisition module: Used with a laser tracker and a spatial scanning probe to perform high-density continuous scanning of the flange surface, sealing surface and cylindrical mating surface of the main pump and suction pipe to acquire raw surface point cloud data; Topology construction and calculation module: used to perform meshing processing on the original surface point cloud data to construct a spatial topology structure. Based on the spatial topology structure, it calculates the curvature feature values ​​of each target data point and the local neighborhood of each target data point in the original surface point cloud data, and extracts the local normal vector of each target data point. Point cloud purification module: used to calculate the spatial angle between the local normal vector of each target data point and the average normal vector of its local neighborhood, set the local curvature discrimination threshold and the normal vector angle threshold, and determine the target data points whose curvature feature value exceeds the local curvature discrimination threshold or whose spatial angle exceeds the normal vector angle threshold as pseudo-feature noise points, and remove them from the original surface point cloud data to obtain purified metal reference point cloud; Noise-resistant fitting module: used to perform noise-resistant fitting on the purified metal reference point cloud using a random sampling consensus algorithm, and to construct a high-precision flange plane and cylinder axis by using the local normal vector of the purified metal reference point cloud as a directional constraint. Assembly interference simulation calculation module: used to establish a coordinate system based on the high-precision flange plane and cylindrical axis, and to simulate the installation and interference of the main pump and suction pipe.

[0017] The beneficial effects of this invention are as follows: 1. Effectively eliminates false feature noise and restores the true metal reference surface: Through a dual threshold discrimination mechanism of curvature feature value and normal vector angle, it can automatically identify and eliminate false feature points such as uneven anti-rust coating, oxide layer accumulation, and fine tool marks and scratches on the surface of the main pump and suction pipe caused by transportation and storage. This solves the technical defect of existing discrete sampling that easily introduces false surface data, and obtains a pure point cloud that can truly reflect the underlying metal machining reference surface, thus ensuring the accuracy of modeling and measurement from the source.

[0018] 2. Eliminate the absorption and amplification of minute errors to ensure installation accuracy: The improved RANSAC algorithm with local normal vector direction constraint is adopted, combined with high-precision tolerance control, to avoid the problem of the algorithm absorbing minute errors of 0.05mm to 0.1mm caused by the fixed rejection threshold in the existing technology. This prevents the minute angular offset of the cylindrical axis from being amplified to the millimeter level along the height direction of the main pump, and ensures that the accuracy of the fitted flange plane and the cylindrical axis meets the installation requirement of concentricity deviation <0.19mm.

[0019] 3. Improve measurement efficiency and reliability, and adapt to complex on-site conditions: Replace traditional discrete sampling with high-density continuous scanning, and combine it with spatial topology construction to achieve rapid point cloud retrieval and feature calculation; the whole process is highly automated, and there is no need for manual intervention to identify false feature points. It is suitable for complex on-site conditions such as obstructed line of sight and narrow installation space in nuclear power plants, avoids errors caused by manual operation, and improves measurement efficiency and result reliability. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart of the steps of a three-dimensional modeling and measurement method for a main pump, which involves point cloud curvature discrimination and noise-resistant fitting, according to the present invention. Figure 2 This is a system module diagram of a main pump 3D modeling and measurement system for point cloud curvature discrimination and noise-resistant fitting according to the present invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0023] Example 1

[0024] This invention discloses a three-dimensional modeling and measurement method for main pumps, which aims to solve the technical problems of pseudo-feature point errors and the absorption and amplification of small errors in existing three-dimensional modeling and measurement of main pumps, and to ensure the high-precision installation safety of the main pump. The method includes the following steps: Step 1: Using a laser tracker in conjunction with a spatial scanning probe, perform high-density continuous scanning of the flange surface, sealing surface, and cylindrical mating surface of the main pump and suction duct to obtain raw surface point cloud data; In some embodiments, a laser tracker is installed at a stable location in the steam generator chamber or main pump storage area of ​​a nuclear power plant. Known spatial control points pre-deployed around the site are used to relocate and align the laser tracker with a reference, establishing a global measurement coordinate system. The spatial scanning probe (e.g., a handheld 3D laser scanner or structured light scanner) and the laser tracker are optically tracked, locked, and calibrated synchronously to ensure that the local microscopic data collected by the spatial scanning probe can be converted into the global measurement coordinate system of the laser tracker in real time and accurately. The specific process of establishing a global measurement coordinate system is as follows: Using multiple known spatial control points pre-deployed and verified on-site as a reference, one of the reference control points is selected as the origin O0 of the global coordinate system; a right-handed rectangular global measurement coordinate system O0-X0Y0Z0 is constructed with the direction parallel to the main axis of the factory building on the horizontal plane as the X0 axis, the direction perpendicular to the X0 axis on the horizontal plane as the Y0 axis, and the vertical upward direction as the Z0 axis; multiple spatial control points are observed using a laser tracker to complete the coordinate system orientation and adjustment calculation, thereby achieving the unification and locking of the global measurement coordinate system; Before scanning, the surfaces to be tested, including the main pump and suction pipe, are cleaned non-destructively to remove large dust particles or obvious oil stains without damaging the anti-rust coating and original metal processing texture that came with the equipment at the factory. Under the continuous tracking of the laser tracker, the space scanning probe performs high-density continuous scanning on the key geometric assembly elements of the main pump and the suction duct. The specific scanning areas include, but are not limited to: the flange face, sealing face and impeller outer cylindrical mating surface of the main pump body; the flange face and inner and outer cylindrical mating surfaces of the suction duct. To ensure that the subsequent algorithm can effectively identify microscopic surface defects at the tens of micrometer level (such as anti-rust paint accumulation points or fine machining marks), the high-density continuous scanning needs to meet specific sampling resolution requirements. In this embodiment, the point spacing parameter of the spatial scanning probe is set to a range of 0.01mm-0.05mm. At this resolution, the spatial scanning probe sweeps across the surface of the object to be measured at a certain scanning frame rate (e.g., hundreds of thousands to millions of measurement points per second). After the above scanning operation, the massive amount of spatial coordinate points collected are summarized to generate raw surface point cloud data without any filtering or downsampling processing. This raw surface point cloud data not only includes the macroscopic geometry of the main pump and suction duct (such as flatness and cylindricity), but also completely preserves the microscopic morphological features of its surface (including uneven coating thickness, scratches, tool marks and other minor undulations), providing a sufficient basic data source for subsequent curvature calculation and pseudo feature point identification. Step 2: Perform meshing processing on the original surface point cloud data, calculate the curvature feature values ​​of each data point and the local neighborhood of each data point in the original surface point cloud data, and extract the local normal vector of each data point; In some embodiments, the original surface point cloud data is obtained. Since the data consists of discrete spatial scattered points and lacks topological connections, the original surface point cloud data is first processed into a grid (or a spatial topology index is constructed). Specifically, the Kd-tree (K-Dimensional Tree) algorithm or the Delaunay triangular mesh partitioning algorithm is used to construct a spatial topology structure for the massive amount of scattered original surface point cloud data. The purpose of this step is to enable the rapid and accurate retrieval of the local neighborhood point set around any target data point in the subsequent curvature feature value calculation, thereby greatly improving the computational efficiency of massive data. Based on the constructed spatial topology, the curvature feature values ​​of each data point and its local neighborhood in the original surface point cloud data are calculated, and the local normal vector of each data point is extracted. The core algorithm adopts the principal component analysis of the covariance matrix (PCA algorithm) based on the local neighborhood point set, which specifically includes the following sub-steps: Define the search range of the local neighborhood. Considering that the scanning point spacing set in step one is 0.01mm-0.05mm, in order to be able to accurately capture high-frequency micro-undulations such as rust-preventive coating drips and fine machining marks, while not being disturbed by the macroscopic curved surfaces of the main pump and suction duct (such as the macroscopic curvature of the cylindrical mating surface itself), in this embodiment, the search radius R of the local neighborhood is set to 0.5mm-2.0mm (or the number of nearest neighbor points K for K-nearest neighbor search is set to 30-50). For any target data point in the original surface point cloud data Extract all neighboring points within its search radius R to form a local neighborhood point set; calculate the geometric center (centroid) of this local neighborhood point set. Suppose that the local neighborhood set contains k data points, and the coordinates of each point are... The formula for calculating the centroid is: , , ; Center all data points within the local neighborhood point set to obtain the offset coordinates of each point relative to the centroid. Based on this offset coordinate, a 3×3 covariance matrix M is constructed. The formula for calculating the covariance matrix M is as follows: The specific form of the expanded covariance matrix M is: Where Cov(a,b) represents the covariance between variables a and b, satisfying Cov(a,b)=Cov(b,a), and the matrix is ​​a symmetric positive definite matrix. Eigenvalue decomposition is performed on this 3×3 covariance matrix M, and the solution satisfies... eigenvalues ​​and eigenvectors, where Let V be the eigenvalue and V be the corresponding eigenvector; solve using the determinant. (I is a 3rd order identity matrix), obtain three non-negative eigenvalues, sort them from largest to smallest as λ1, λ2, λ3 (i.e. λ1≥λ2≥λ3≥0), and obtain the corresponding three orthogonal eigenvectors v1, v2, v3 by substituting the eigenvalues ​​into the system of linear equations. Based on the above decomposition results, the curvature characteristic value of the target data point Pi is calculated (defined here as the surface variation coefficient). The calculation formula is as follows: The curvature eigenvalue In a physical sense, this reflects the degree to which the local neighborhood deviates from the absolutely smooth tangent plane. If the region is a smooth flange surface or a smooth cylindrical mating surface, the minimum eigenvalue λ3 approaches 0. Approaching the minimum value; if the target data point Located precisely at the edge of coating buildup or the depression of tool marks, the randomness of the three-dimensional spatial distribution increases, and λ3 increases significantly. At this point, the curvature eigenvalue... Significant mutations will occur; Extract the feature vector v3 corresponding to the smallest feature value λ3, and define it as the target data point. Local normal vector To ensure the consistency of subsequent directional constraints, the viewpoint orientation method or minimum spanning tree (MST) method is used to adjust the orientation of all extracted local normal vectors to ensure that all local normal vectors point uniformly to the external free space of the main pump or suction duct surface, thus avoiding the problem of random flipping of the positive and negative orientation of the normal vectors. By iterating through each point in the original surface point cloud data and repeating the covariance matrix analysis process described above, each data point is finally assigned a precise curvature feature value and a local normal vector attribute consistent with the orientation, providing a multi-dimensional basis for subsequent micromorphological identification. Step 3: Calculate the spatial angle between the local normal vector of each target data point and the local neighborhood average normal vector of the target data point; set a local curvature discrimination threshold and a normal vector angle threshold, and determine the target data points whose curvature feature value exceeds the local curvature discrimination threshold or whose spatial angle exceeds the normal vector angle threshold as pseudo-feature noise points caused by surface coating or processing tool marks, and remove them from the original surface point cloud data to obtain a purified metal reference point cloud; For any target data point in the original surface point cloud data Extract all neighboring points within the local neighborhood determined in step two, sum the local normal vectors of all neighboring points within the local neighborhood, and normalize them to calculate the average normal vector of the local neighborhood. The average normal vector In a physical sense, it represents the target data point. The macroscopic theoretical orientation of the micro-region in which it is located; Calculate the target data point using the vector dot product formula Its own local normal vector With the average normal vector The spatial angle between The calculation formula is as follows: The included angle of this space It reflects the target data points with extreme sensitivity. Whether a microscopic tilt or abrupt change has occurred on the surface; To satisfy the principle of sufficient disclosure, this embodiment sets a local curvature discrimination threshold and a normal vector angle threshold. The basis for setting the threshold is explained in detail below: Since the flange, sealing, and cylindrical mating surfaces of the main pump and suction pipe are all high-precision machined surfaces at the factory, their surface variation coefficient is extremely small and the normal vector transition is smooth under ideal conditions. However, the rust-preventive coating drips (usually tens to hundreds of micrometers thick) or fine machining marks / scratches present on site can cause drastic changes in geometry within a very small spatial scale. Therefore, in this embodiment, the range of the local curvature discrimination threshold is set to 0.005~0.015; When the curvature feature value of the target data point is greater than the local curvature discrimination threshold, it indicates that the three-dimensional randomness of the area where the target data point is located exceeds the roughness limit of the normal machined surface, and it is highly likely that the edge of the coating accumulation has been hit. The range of the normal vector angle threshold is set to 1.5 to 3.0 degrees. When the spatial angle of a target data point is greater than the normal vector angle threshold, it indicates that the surface orientation of the target data point has undergone abnormal micro-curving or micro-depression, which is a typical geometric manifestation of scratches or uneven coating. After setting the local curvature discrimination threshold and the normal vector angle threshold, an OR logic judgment is performed on each target data point in the original surface point cloud data: If the curvature feature value of the target data point Pi is greater than the local curvature discrimination threshold, or the spatial angle of the target data point Pi is greater than the normal vector angle threshold, as long as either of these conditions is met, the target data point will be judged as a false feature noise point caused by surface coating or machining tool marks. This dual discrimination mechanism of curvature feature value and normal vector angle can ensure zero leakage of micro-defects. All target data points marked as pseudo-feature noise points are stripped from the original surface point cloud data (i.e., removed). After the stripping process, the remaining set of target data points is the purified metal reference point cloud. This purified metal reference point cloud completely eliminates the interference of surface physical state caused by long-distance transportation and on-site storage, and truly restores the metal machining reference surface of the bottom layer of the main pump and suction pipe, thereby cutting off the possibility of small errors being absorbed and amplified by subsequent fitting algorithms from the source. Step 4: The purified metal reference point cloud is fitted with a random sampling consensus algorithm to resist noise. During the sampling iteration process, the local normal vector of the purified metal reference point cloud is used as the directional constraint condition. Outliers that do not conform to the plane parallel constraint or cylinder perpendicular constraint are eliminated to construct a high-precision flange plane and cylinder axis. The purified metal reference point cloud output from step three is received. Since the point cloud has been stripped of pseudo-feature noise points of micro-coating and tool mark noise, it represents the real metal machining surface. Based on the spatial distribution position of each data point in the global measurement coordinate system and the bounding box (the bounding box of the purified metal reference point cloud is generated as a whole), combined with the point cloud geometric region clustering features, the purified metal reference point cloud is automatically divided into a planar point cloud subset for fitting the flange surface and a cylindrical point cloud subset for fitting the cylindrical mating surface. For example, the outer bounding box is an axis-aligned three-dimensional bounding box, which completes the region division based on the differences in the spatial coordinate range of the point clouds in each region; wherein the flange surface structure is generally horizontally distributed, the point cloud spatial height range is narrow, and the normal vector direction is uniformly vertical, forming a planar point cloud subset after clustering; the cylindrical mating surface is spatially distributed in a long strip along the axial direction, and the point cloud extends and is distributed over a large range along the axial direction, forming a cylindrical point cloud subset after clustering, thereby realizing the automatic classification and separation of point clouds of different geometric elements to be fitted; Subsequently, an improved Random Sampling Consensus (RANSAC) algorithm with normal vector direction constraints is introduced to construct the flange plane and cylinder axis with high accuracy: In the subset of the planar point cloud, the Random Sampling Consensus Algorithm (RANSAC) sampling iteration is initiated. In each iteration, three target data points are randomly selected as the minimum sample set, and a candidate flange plane equation model and the theoretical normal vector of the candidate flange plane equation model are calculated. At this point, a plane parallel constraint is introduced: the local normal vectors of these three target data points are extracted, and the angles between them and the theoretical normal vector Nplane are calculated respectively. A plane parallel angle tolerance threshold is set. Since the purified metal reference point cloud has been purified, the tolerance requirement here is extremely high. In this embodiment, the plane parallel angle tolerance threshold is set to 0.5-1.0 degrees. If the angle between the local normal vector of any of the three target data points and the theoretical normal vector Nplane is greater than the plane parallel angle tolerance threshold, the minimum sample set is directly determined to be invalid and the current iteration is skipped. This step greatly avoids the algorithm from establishing incorrect candidate models based on micro-tilt points. If the angle between the local normal vector and the theoretical normal vector of any one of the three target data points is less than or equal to the plane parallel angle tolerance threshold, then the minimum sample set is deemed valid. If the minimum sample set is valid, then traverse all remaining target data points in the planar point cloud subset, calculate the orthogonal distance d from the remaining target data point to the candidate flange plane, and the angle between the local normal vector and the theoretical normal vector of the remaining target data point. Set a distance tolerance threshold (e.g., 0.02mm-0.05mm). When the orthogonal distance is less than the distance tolerance threshold, and the angle between the local normal vector and the theoretical normal vector Nplane is less than the plane parallel angle tolerance threshold, the target data point is determined to be an inlier; otherwise, outlier target data points whose normal vector direction does not conform to the plane parallel constraint are removed. After multiple iterations, the candidate flange plane equation model with the most interior points is retained, and the least squares method is used to perform the final plane fitting on all interior points corresponding to the candidate flange plane equation model to output a high-precision flange plane. In the cylindrical point cloud subset, the Random Sampling Consensus Algorithm (RANSAC) sampling iteration is initiated. In each iteration, three target data points are randomly selected as the minimum sample set to construct a candidate cylindrical model and obtain the candidate cylindrical axis vector of the candidate cylindrical model. At this point, a cylinder vertical constraint is introduced: according to the principles of spatial geometry, the normal vector of any point on the surface of a real cylinder must be perpendicular to the central axis of the cylinder. The local normal vectors of these three target data points are extracted, and the angles between them and the candidate cylinder axis vector are calculated. A cylinder vertical angle tolerance threshold range is set. Since the purified metal reference point cloud has been purified, the tolerance requirement here is extremely high. In this embodiment, the cylinder vertical angle tolerance threshold range is set to 89.5–91 degrees. If the angle between the local normal vector of any of the three target data points and the candidate cylinder axis vector is not within the cylinder vertical angle tolerance threshold range, then the minimum sample set is directly determined to be invalid, and the current iteration is skipped. This step greatly avoids the algorithm establishing incorrect candidate cylinder models based on microscopic anomalies; otherwise, the minimum sample set is determined to be valid. If the minimum sample set is valid, then iterate through all remaining target data points in the cylindrical point cloud subset, calculate the distance error from the remaining target data point to the candidate cylindrical model, and the angle between the local normal vector of the remaining target data point and the candidate cylindrical axis vector. Set a distance tolerance threshold (e.g., 0.02mm-0.05mm). When the distance error is less than the distance tolerance threshold and the angle between the local normal vector and the candidate cylindrical axis vector satisfies the perpendicular relationship, the target data point is determined as an inlier; otherwise, outlier target data points whose normal vector direction does not conform to the cylindrical perpendicular constraint are removed. After multiple iterations, the candidate cylinder model with the most interior points is retained, and the least squares method is used to perform the final cylinder fitting on all interior points corresponding to the candidate cylinder model, outputting a high-precision cylinder axis. Through the improved fitting process described above, the local normal vector serves as a strong geometric direction constraint, eliminating the "small tilt of the model as a whole" phenomenon that is easily generated when fitting based solely on distance tolerance in the traditional method. This ensures that the constructed flange plane is never warped and the cylindrical axis is never skewed, thus eliminating the error leverage amplification effect caused by the main pump's height of several meters from the bottom of the mathematical algorithm.

[0025] Step 5: Establish a coordinate system based on the high-precision flange plane and cylindrical axis, and perform simulated installation and interference calculation of the main pump and suction pipe.

[0026] After obtaining the high-precision flange plane and high-precision cylindrical axis of the main pump and suction pipe in step four, the final spatial pose calculation and virtual assembly stage is entered. Since the previous steps have completely eliminated the false feature noise points caused by surface coating and micro-tooling marks, the geometric elements at this time represent the most realistic physical reference, which can ensure extremely high assembly calculation fidelity. First, a coordinate system is established based on the high-precision flange plane and cylindrical axis. Specifically, a local reference coordinate system for the suction conduit and a local assembly coordinate system for the main pump are established respectively. For the inhalation catheter, its high-precision cylindrical axis is defined as the reference Z1 axis; the spatial intersection of its high-precision flange plane and the Z1 axis is defined as the origin O1 of the reference coordinate system; the high-precision flange plane is used as the X1-Y1 reference plane, thereby establishing the local reference coordinate system O1-X1Y1Z1 for the inhalation catheter. Similarly, for the main pump, its high-precision cylindrical axis is defined as the assembly Z2 axis; the spatial intersection of its high-precision flange plane and the Z2 axis is defined as the origin O2 of the assembly coordinate system; the high-precision flange plane is used as the X2-Y2 assembly plane, thereby establishing the local assembly coordinate system O2-X2Y2Z2 of the main pump. Subsequently, since a unified global measurement coordinate system has been established in step one using a laser tracker, the two local coordinate systems mentioned above are mapped to this global measurement coordinate system using a rigid body transformation matrix (including rotation matrix and translation vector) to simulate the installation of the main pump and suction duct. In the three-dimensional measurement software, the three-dimensional digital model driving the main pump approaches the suction duct model along the predetermined lifting installation trajectory until the origins (O1 and O2) of both reach the designed assembly elevation in the Z-axis direction (or until the two high-precision flange planes are theoretically aligned). Finally, under simulated installation conditions, interference calculations and spatial deviation assessments are performed, and the following core assembly parameters are automatically extracted and calculated: Concentricity deviation calculation: Calculate the radial distance in space between the main pump assembly Z2 axis and the suction pipe reference Z1 axis. Since the main pump is several meters high, calculate the radial distance at the flange elevation (bottom) and the impeller elevation (top) respectively, and take the maximum value as the final concentricity deviation value. Parallelism deviation calculation: Calculate the spatial dihedral angle between the high-precision flange plane of the main pump and the high-precision flange plane of the suction pipe, and convert it into the maximum opening gap at the flange edge as the parallelism deviation value. Interference determination logic: retrieve the limit gap threshold set in the nuclear power plant main pump installation process specification (for example, the upper limit of the concentricity deviation set in this embodiment is 0.19mm), and compare the calculated concentricity deviation value with the limit gap threshold. If the concentricity deviation value is greater than or equal to the limit gap threshold, it is determined that during the actual physical lifting process, the outer cylindrical mating surface of the main pump impeller will inevitably interfere and scrape with the inner cylindrical mating surface of the suction pipe. Then, an interference warning and three-dimensional attitude adjustment amount (i.e., translational correction amount and angular leveling amount in the X / Y direction) will be output. If the concentricity deviation value is less than the limit gap threshold and the parallelism deviation value meets the sealing requirements, then the simulated installation is deemed to have passed, and a construction command allowing physical jacking is output. By eliminating microscopic pseudo-feature errors at the level of 0.05mm-0.1mm from the underlying algorithm, the leverage amplification effect of this tiny error on the Z-axis at a height of several meters is avoided, thereby making the final output interference calculation result have absolute engineering guidance reliability and completely ensuring the safety of main pump installation in extremely narrow spaces.

[0027] Example 2

[0028] like Figure 2 As shown, based on the specific implementation process of Embodiment 1, the present invention provides a three-dimensional modeling and measurement system for a main pump with point cloud curvature discrimination and noise-resistant fitting, comprising: Point cloud scanning acquisition module: Used with a laser tracker and a spatial scanning probe to perform high-density continuous scanning of the flange surface, sealing surface and cylindrical mating surface of the main pump and suction pipe to acquire raw surface point cloud data; Topology construction and calculation module: used to perform meshing processing on the original surface point cloud data to construct a spatial topology structure. Based on the spatial topology structure, it calculates the curvature feature values ​​of each target data point and the local neighborhood of each target data point in the original surface point cloud data, and extracts the local normal vector of each target data point. Point cloud purification module: used to calculate the spatial angle between the local normal vector of each target data point and the average normal vector of its local neighborhood, set the local curvature discrimination threshold and the normal vector angle threshold, and determine the target data points whose curvature feature value exceeds the local curvature discrimination threshold or whose spatial angle exceeds the normal vector angle threshold as pseudo-feature noise points, and remove them from the original surface point cloud data to obtain purified metal reference point cloud; Noise-resistant fitting module: used to perform noise-resistant fitting on the purified metal reference point cloud using a random sampling consensus algorithm, and to construct a high-precision flange plane and cylinder axis by using the local normal vector of the purified metal reference point cloud as a directional constraint. Assembly interference simulation calculation module: used to establish a coordinate system based on the high-precision flange plane and cylindrical axis, and to simulate the installation and interference of the main pump and suction pipe.

[0029] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for three-dimensional modeling and measurement of a main pump using point cloud curvature discrimination and noise-resistant fitting, characterized in that: include: Using a laser tracker in conjunction with a spatial scanning probe, high-density continuous scanning is performed on the flange surface, sealing surface, and cylindrical mating surface of the main pump and suction duct to obtain raw surface point cloud data; The original surface point cloud data is meshed to construct a spatial topology. Based on the spatial topology, the curvature feature values ​​of each target data point and the local neighborhood of each target data point in the original surface point cloud data are calculated, and the local normal vector of each target data point is extracted. Calculate the spatial angle between the local normal vector of each target data point and the average normal vector of its local neighborhood. Set a local curvature discrimination threshold and a normal vector angle threshold. Target data points whose curvature feature value exceeds the local curvature discrimination threshold or whose spatial angle exceeds the normal vector angle threshold are identified as pseudo-feature noise points and removed from the original surface point cloud data to obtain a purified metal reference point cloud. The purified metal reference point cloud is fitted with a random sampling consensus algorithm to resist noise, and the local normal vector of the purified metal reference point cloud is used as a directional constraint to construct a high-precision flange plane and cylinder axis. A coordinate system is established based on the high-precision flange plane and cylindrical axis to simulate the installation of the main pump and the suction pipe and calculate the interference.

2. The method for three-dimensional modeling and measurement of a main pump with point cloud curvature discrimination and noise-resistant fitting according to claim 1, characterized in that, The specific process for obtaining the original surface point cloud data is as follows: By using pre-deployed spatial control points, a global measurement coordinate system is established for the laser tracker, and the laser tracker and the spatial scanning probe are unified into the global measurement coordinate system; The spatial scanning probe scans the flange surface, sealing surface, impeller outer cylindrical mating surface, and suction duct flange surface, inner cylindrical mating surface, and outer cylindrical mating surface of the main pump body at a set scanning frame rate and scanning point spacing to obtain raw surface point cloud data.

3. The method for three-dimensional modeling and measurement of a main pump with point cloud curvature discrimination and noise-resistant fitting according to claim 1, characterized in that, The specific calculation process for the curvature characteristic value is as follows: The spatial topology of the original surface point cloud data is constructed using the Kd-tree algorithm or the Delaunay triangular mesh partitioning algorithm. In the spatial topology, a search radius for the local neighborhood is set. For any target data point in the original surface point cloud data, all neighboring points within the search radius are extracted to form a local neighborhood point set. The geometric center of the local neighborhood point set is calculated, and a 3×3 covariance matrix of the local neighborhood point set is constructed. Eigenvalue decomposition is performed on the 3×3 covariance matrix to obtain three non-negative eigenvalues ​​and eigenvectors. Finally, the curvature eigenvalue of the target data point is calculated. λ1, λ2, and λ3 are three non-negative eigenvalues, and λ1 ≥ λ2 ≥ λ3.

4. The method for three-dimensional modeling and measurement of a main pump with point cloud curvature discrimination and noise-resistant fitting according to claim 1, characterized in that, The specific process for extracting the local normal vectors of each target data point is as follows: Extract the eigenvector corresponding to the smallest eigenvalue, and define it as the local normal vector of the target data point.

5. The method for three-dimensional modeling and measurement of a main pump with point cloud curvature discrimination and noise-resistant fitting according to claim 1, characterized in that, The specific process for determining a point as a pseudo-feature noise point is as follows: The local normal vectors of all neighboring points within the local neighborhood of the target data point are summed and normalized to obtain the average normal vector of the local neighborhood. Calculate the spatial angle between the local normal vector and the average normal vector of the target data point; set a local curvature discrimination threshold and a normal vector angle threshold; If the curvature feature value of the target data point is greater than the local curvature discrimination threshold, or the spatial angle is greater than the normal vector angle threshold, then the target data point is determined to be a pseudo-feature noise point.

6. The method for three-dimensional modeling and measurement of a main pump with point cloud curvature discrimination and noise-resistant fitting according to claim 1, characterized in that, The specific process for obtaining the purified metal reference point cloud is as follows: All target data points marked as pseudo-feature noise points are stripped from the original surface point cloud data, and the remaining set of target data points is the purified metal reference point cloud.

7. The method for three-dimensional modeling and measurement of a main pump with point cloud curvature discrimination and noise-resistant fitting according to claim 1, characterized in that, The specific process for constructing a high-precision flange plane is as follows: The purified metal reference point cloud is divided into a planar point cloud subset. In the planar point cloud subset, the Random Sampling Consensus Algorithm (RANSAC) is used for sampling iteration. In each iteration, three target data points are randomly selected as the minimum sample set to calculate a candidate flange plane equation model and the theoretical normal vector of the candidate flange plane equation model. Calculate the angle between the local normal vector and the theoretical normal vector for each of the three target data points, and set a tolerance threshold for the plane parallel angle. If the angle between the local normal vector and the theoretical normal vector of any target data point is less than or equal to the plane parallel angle tolerance threshold, the minimum sample set is determined to be valid. Then, the orthogonal distance from the target data point to the candidate flange plane and the angle between the local normal vector and the theoretical normal vector of the target data point are calculated, and the distance tolerance threshold is set. When the orthogonal distance is less than the distance tolerance threshold and the angle between the local normal vector and the theoretical normal vector is less than the plane parallel angle tolerance threshold, the target data point is determined as an interior point. The candidate flange plane equation model with the most interior points is retained, and the least squares method is used to fit the plane to all interior points to output a high-precision flange plane.

8. The method for three-dimensional modeling and measurement of a main pump with point cloud curvature discrimination and noise-resistant fitting according to claim 1, characterized in that, The specific process for constructing a high-precision cylindrical axis is as follows: The purified metal reference point cloud is segmented into a cylindrical point cloud subset; In the cylindrical point cloud subset, the Random Sampling Consensus Algorithm (RANSAC) is used for sampling iteration. In each iteration, the minimum number of target data points required to calculate the cylinder are randomly selected as the minimum sample set to construct a candidate cylindrical model and obtain the candidate cylindrical axis vector of the candidate cylindrical model. Calculate the angle between the local normal vector of each target data point in the minimum sample set and the axis vector of the candidate cylinder, and set the tolerance threshold range for the vertical angle of the cylinder; If the angle between the local normal vector of all target data points and the candidate cylinder axis vector is within the cylinder vertical angle tolerance threshold range, then the minimum sample set is determined to be valid. If the minimum sample set is valid, calculate the distance error of other target data points to the candidate cylindrical model, and calculate the angle between its local normal vector and the axis vector of the candidate cylindrical model, and set the distance tolerance threshold. When the distance error is less than the distance tolerance threshold, and the angle between the local normal vector and the candidate cylinder axis vector satisfies a perpendicular relationship, the target data point is determined to be an interior point; The candidate cylindrical model with the most interior points is retained, and the least squares method is used to fit the cylinder to all interior points of the model, outputting a high-precision cylinder axis.

9. The method for three-dimensional modeling and measurement of a main pump with point cloud curvature discrimination and noise-resistant fitting according to claim 1, characterized in that, The specific process for simulating the installation and calculating the interference of the main pump and the suction duct is as follows: For the inhalation catheter, the high-precision cylindrical axis is defined as the reference Z1 axis, the spatial intersection of the high-precision flange plane and the Z1 axis is defined as the origin O1 of the reference coordinate system, and the high-precision flange plane is used as the X1-Y1 reference plane, thereby establishing the local reference coordinate system O1-X1Y1Z1 for the inhalation catheter. For the main pump, the high-precision cylindrical axis is defined as the assembly Z2 axis, and the spatial intersection of the high-precision flange plane and the Z2 axis is defined as the origin O2 of the assembly coordinate system. The high-precision flange plane is used as the X2-Y2 assembly plane, thereby establishing the local assembly coordinate system O2-X2Y2Z2 of the main pump. Using a rigid body transformation matrix, the local reference coordinate system O1-X1Y1Z1 of the inhalation tubing and the local assembly coordinate system O2-X2Y2Z2 of the main pump are transformed into the global measurement coordinate system; Calculate the radial distance in space between the main pump assembly Z2 axis and the suction pipe reference Z1 axis, and take the maximum value as the final concentricity deviation value. Calculate the spatial dihedral angle between the high-precision flange plane of the main pump and the high-precision flange plane of the suction pipe, and convert it into the maximum opening gap at the flange edge as the parallelism deviation value. The calculated concentricity deviation value is compared with the limit gap threshold. If the concentricity deviation value is greater than or equal to the limit clearance threshold, it is determined that the outer cylindrical mating surface of the main pump impeller will cause metal interference and scraping with the inner cylindrical mating surface of the suction duct. If the concentricity deviation is less than the limit gap threshold and the parallelism deviation meets the sealing requirements, then the installation is considered successful.

10. A three-dimensional modeling and measurement system for a main pump, used for point cloud curvature discrimination and noise-resistant fitting, for performing the method described in any one of claims 1-9, characterized in that, include: Point cloud scanning acquisition module: Used with a laser tracker and a spatial scanning probe to perform high-density continuous scanning of the flange surface, sealing surface and cylindrical mating surface of the main pump and suction pipe to acquire raw surface point cloud data; Topology construction and calculation module: used to perform meshing processing on the original surface point cloud data to construct a spatial topology structure. Based on the spatial topology structure, it calculates the curvature feature values ​​of each target data point and the local neighborhood of each target data point in the original surface point cloud data, and extracts the local normal vector of each target data point. Point cloud purification module: used to calculate the spatial angle between the local normal vector of each target data point and the average normal vector of its local neighborhood, set a local curvature discrimination threshold and a normal vector angle threshold, and determine the target data points whose curvature feature value exceeds the local curvature discrimination threshold or whose spatial angle exceeds the normal vector angle threshold as pseudo-feature noise points, and remove them from the original surface point cloud data to obtain a purified metal reference point cloud; Noise-resistant fitting module: used to perform noise-resistant fitting on the purified metal reference point cloud using a random sampling consensus algorithm, and to construct a high-precision flange plane and cylinder axis by using the local normal vector of the purified metal reference point cloud as a directional constraint. Assembly interference simulation calculation module: used to establish a coordinate system based on the high-precision flange plane and cylindrical axis, and to simulate the installation and interference of the main pump and suction pipe.

Images