Three-dimensional digital model generation method and system applied to automobile parts

Abstract

The invention relates to the technical field of automobile part digital models, in particular to a three-dimensional digital model generation method and system applied to automobile parts. The method comprises the following steps: collecting original morphology data of an automobile part, and forming an original point cloud set; inputting the original point cloud set into a pre-trained geometric feature recognition network, and extracting feature mapping data of the surface of the automobile part; identifying hardware structure features of the automobile part based on the feature mapping data; and according to the identified hardware structure features, constructing a topological information matrix of the automobile parts, extracting an assembly relationship between the topological information matrix and adjacent parts, and generating assembly semantic data. According to the method, through fusion of geometric feature recognition, assembly semantic construction and space registration technologies, full-process automatic modeling of the automobile part from point cloud collection to semantic reconstruction and three-dimensional rendering is realized, and the precision, stability and assembly analysis capability of a digital model are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of digital modeling technology for automotive parts, and in particular to a method and system for generating three-dimensional digital models of automotive parts. Background Technology

[0002] 3D modeling technology has become a key supporting means for component design, assembly verification, and virtual simulation. Early 3D models mainly relied on CAD geometric modeling and manual feature annotation, a cumbersome process with limited accuracy in reproducing the morphology of complex curved surfaces and welded structures. Subsequently, the introduction of point cloud scanning and reverse engineering technologies enabled the direct digitization of component surface geometric data. However, traditional point cloud modeling methods only remain at the geometric level, lacking automatic recognition of structural semantics such as hole positions, ribs, and bosses, as well as the expression of assembly relationships, making it difficult for the generated models to participate in structural analysis and assembly simulation. In addition, existing point cloud registration algorithms mostly rely on manual point selection or global optimization, resulting in registration accuracy being greatly affected by noise and unstable attitude calibration, making it difficult to form a holistic digital model with assembly semantics. Summary of the Invention

[0003] Therefore, the present invention needs to provide a method and system for generating three-dimensional digital models of automotive parts to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a method for generating three-dimensional digital models of automotive parts includes the following steps: Step S1: Collect the original morphological data of the automotive parts and construct the original point cloud; input the original point cloud into the pre-trained geometric feature recognition network to extract the feature mapping data of the automotive parts surface. Step S2: Identify the hardware structure features of automotive parts based on feature mapping data; construct the topology information matrix of automotive parts based on the identified hardware structure features, extract their assembly relationship with adjacent components, and generate assembly semantic data. Step S3: Using feature mapping data and assembly semantic data, combined with the spatial constraint relationship of hardware structural features, establish multi-dimensional registration matrix data; Step S4: Based on the multi-dimensional registration matrix data, perform 3D rendering and solid reconstruction of the geometric contours and hardware structural features of the automotive parts to generate a 3D digital model containing complete assembly semantic information.

[0005] Preferably, the present invention also provides a three-dimensional digital model generation system for automotive parts, used to execute the above-described three-dimensional digital model generation method for automotive parts, wherein the three-dimensional digital model generation system for automotive parts includes: The original shape acquisition and point cloud construction module is used to acquire the original shape data of automotive parts and construct the original point cloud set; the original point cloud set is input into a pre-trained geometric feature recognition network to extract the feature mapping data of the automotive part surface. The geometric feature recognition and feature mapping generation module is used to identify the hardware structure features of automotive parts based on feature mapping data; based on the identified hardware structure features, it constructs a topological information matrix of the automotive parts and extracts their assembly relationship with adjacent components to generate assembly semantic data. The structural semantic parsing and topological semantic modeling module is used to establish multi-dimensional registration matrix data by utilizing feature mapping data and assembly semantic data, combined with the spatial constraint relationship of hardware structural features. The spatial registration and 3D reconstruction rendering module is used to perform 3D rendering and solid reconstruction of the geometric contours and hardware structural features of automotive parts based on multi-dimensional registration matrix data, generating a 3D digital model containing complete assembly semantic information.

[0006] This invention significantly improves the intelligence and accuracy of 3D digital modeling of automotive parts by integrating geometric feature recognition, assembly semantic construction, and multi-dimensional spatial registration. The method introduces a high-fidelity feature recognition mechanism into the processing of raw point cloud data, automatically extracting structural features of key geometric areas such as holes, ribs, and bosses, achieving accurate restoration of complex surfaces and subtle features, thus effectively eliminating the shape distortion problems caused by traditional CAD modeling and manual annotation. By establishing a topological information matrix and assembly semantic data, the mating relationships, force constraints, and assembly levels between parts are expressed in a data-driven form, solving the deficiency of traditional point cloud models in participating in assembly analysis and virtual verification. The multi-dimensional registration matrix formed by spatial constraint relationships can automatically correct the attitude offset and positional errors of each component in the global coordinate system, significantly improving registration stability and assembly consistency. This method eliminates the uncertainties of manual point selection and global iteration, controlling registration errors within the micrometer range and enhancing the structural computability and physical consistency of the digital model. In the 3D rendering stage, by integrating geometric, texture and assembly semantic information, the accurate mapping of surface lighting, normal direction and reflection characteristics is achieved, so that the model has both morphological realism and structural semantic integrity in visualization and simulation. Attached Figure Description

[0007] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the steps of a method for generating a three-dimensional digital model of an automotive part according to the present invention; Figure 2 This is a schematic diagram comparing the adaptive sampling density in different regions in the embodiment; Figure 3 This is a schematic diagram of an automotive component in the embodiment. Detailed Implementation

[0008] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0009] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0010] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0011] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for generating three-dimensional digital models of automotive parts, the method comprising the following steps: Step S1: Collect the original morphological data of the automotive parts and construct the original point cloud; input the original point cloud into the pre-trained geometric feature recognition network to extract the feature mapping data of the automotive parts surface. Step S2: Identify the hardware structure features of automotive parts based on feature mapping data; construct the topology information matrix of automotive parts based on the identified hardware structure features, extract their assembly relationship with adjacent components, and generate assembly semantic data. Step S3: Using feature mapping data and assembly semantic data, combined with the spatial constraint relationship of hardware structural features, establish multi-dimensional registration matrix data; Step S4: Based on the multi-dimensional registration matrix data, perform 3D rendering and solid reconstruction of the geometric contours and hardware structural features of the automotive parts to generate a 3D digital model containing complete assembly semantic information.

[0012] Preferably, step S1 includes the following steps: The original morphological data of the automotive parts surface is collected by a 3D scanning device. The original morphological data includes at least the edge contour of the mounting hole, the vertical height of the reinforcing rib, and the surface undulation of the welding boss. Redundant point cloud removal and noise filtering are performed on the original topographic data to generate a high-fidelity original point cloud set. The high-fidelity original point cloud is input into a pre-trained geometric feature recognition network, and the local curvature texture information and hardware structure boundary of the automotive part surface are extracted through the convolutional coding layer of the geometric feature recognition network. Based on the extracted local curvature texture information, the roundness error of the mounting hole, the root transition area of ​​the reinforcing rib, and the boundary contour of the welding boss are identified and corrected. The output is the corrected feature map data, which contains the spatial coordinates and geometric properties of the hardware structure features.

[0013] In this embodiment of the invention, the original morphology acquisition and feature extraction steps are completed using an industrial-grade 3D scanning system. The system includes an optical structured light scanner (model: ATOS Triple Scan) and a high-precision rotating platform. The scanner resolution is set to 0.02 mm, and the measurement interval is controlled within 0.05 mm to ensure the sampling accuracy of key features on the surface of automotive parts.

[0014] Before scanning, the automotive part is fixed to a positioning fixture, ensuring that the center line of the mounting hole coincides with the axis of the rotating platform. The system sequentially acquires the original morphological data of the automotive part's surface within a 360° range, forming an initial point cloud dataset. The acquired data includes at least the complete edge contour of the mounting hole, the height curve of the reinforcing rib facade, and the surface undulation data of the top surface of the welding boss.

[0015] Redundant point removal and noise filtering are performed on the initial point cloud data. Redundant point removal is based on density detection results with a point-to-point distance threshold of 0.03 mm. When the density of points in a certain neighborhood exceeds 1.5 times the average density, the system automatically deletes points exceeding the density limit. Noise filtering uses a combination of median filtering and minimum curvature smoothing: with each sampling point as the center, the average deviation of neighboring points from the plane fitting surface is calculated within a 0.1 mm radius. If the deviation exceeds 0.05 mm, the point is replaced with the coordinates of the point with the smallest deviation in the neighborhood. After processing, a high-fidelity original point cloud is generated.

[0016] The high-fidelity raw point cloud is normalized and then input into the convolutional coding layer of the geometric feature recognition network. The convolutional coding layer consists of multiple layers of three-dimensional convolutional units. Each layer scans the point cloud spatial region with a 5×5×5 cube window, calculating the local curvature value κ and the surface normal distribution vector N. The system transforms the combined information of the local curvature value and the normal gradient into a surface texture matrix, where the matrix elements represent the geometric undulation features of each region of the automotive part.

[0017] Based on the surface texture matrix, the system identifies the geometric features of different structural regions.

[0018] For the installation hole area, the roundness error is calculated using the roundness determination formula. When the error is greater than 0.02mm, the system refits the hole boundary center and radius using the least squares method to achieve hole roundness correction.

[0019] For the area at the root of the stiffener rib, the rate of change of the facade normal is calculated. When the rate of change of the facade normal exceeds 0.15, the geometry of the transition surface is corrected by local curvature continuity constraints to ensure a smooth transition at the rib root.

[0020] For the welded boss area, boundary identification conditions are established based on the surface height difference (difference between upper and lower extreme values) and the average curvature of the neighborhood. Points with a surface height difference greater than 0.5 mm and an average curvature of the neighborhood greater than 0.02 are defined as boss boundary points. The system reconstructs the boundary contour line through spline interpolation and removes high-frequency noise caused by weld reflection.

[0021] After the above processing, the corrected feature mapping data is output. The feature mapping data contains the spatial coordinates (x, y, z) and geometric attributes (hole diameter, rib height, boss curvature, etc.) of each hardware structural feature of the automotive part, and is stored in a feature description file in a unified format.

[0022] Preferably, step S2 includes the following steps: Based on feature mapping data, the hardware structural features of automotive parts are identified, including mounting holes, reinforcing ribs, welding bosses, and mating stops. Calculate the connection boundary between adjacent feature regions in the hardware structure features, and generate geometric association data containing the center coordinates of the mounting hole, the direction vector of the stiffening rib, and the height information of the welding boss; The geometric correlation data is grouped according to the boundary features of automotive parts, and a mapping structure is constructed with mounting holes as nodes and reinforcing ribs as connecting edges, forming a topological information matrix that includes the spatial relationships of the hardware structure. Based on the topological information matrix, the assembly constraint conditions between the mounting hole and the mating stop are calculated, and constraint relationship data including radial clearance and axial preload are generated. Based on the constraint relationship data, the assembly type of the welding boss and adjacent components is identified, and assembly relationship data including the pressure distribution of the contact surface and the torque transmission path is formed. Based on preset automotive part naming rules and installation sequence identifiers, assembly relationship data is transformed into assembly semantic data.

[0023] In this embodiment of the invention, the feature mapping data comes from a standardized feature description file, which contains the three-dimensional coordinates (x, y, z), local curvature, normal vector, and surface roughness of each feature point.

[0024] First, the hardware structural features in automotive parts are identified through a geometric feature classification unit.

[0025] For mounting hole location identification, spatial neighborhood analysis is performed on all feature point sets to establish point group regions with a neighborhood radius of 0.5 mm. Regions are identified as mounting hole locations when the rate of change of the normal vector of points within a region is less than 0.05 and the mean curvature is between 0.01 and 0.02. The center coordinates and radius parameters are calculated using a circular fitting equation, and the center point of the mounting hole is defined using these coordinates.

[0026] To enhance rib identification, based on the consistency of the direction of the normal vectors in the point cloud, point sequences distributed along a single principal direction and with a curvature gradient less than 0.03 are identified as rib regions. Least-squares linear fitting is then performed along the direction of the point sequence to obtain the rib orientation vector.

[0027] Welding boss identification is achieved by calculating the surface height difference and local average curvature. When the surface height difference is between 0.4mm and 1.2mm and the local average curvature is greater than 0.015, it is identified as a welding boss area.

[0028] For mating stop identification, a planar region adjacent to the hole position and with an angle of less than 10° between the surface normal and the hole axis is selected. The outer contour boundary of this region is defined as the mating stop feature.

[0029] After recognition is completed, a feature index table containing the above four types of features is generated. Each row records the feature type, spatial center coordinates, and main geometric parameters.

[0030] Subsequently, the spatial connection boundaries between each feature are calculated to form geometric association data.

[0031] The connection boundary between the mounting hole and the rib is determined by measuring the shortest distance from the circumferential edge of the hole to the surface of the rib. When the shortest distance is less than 0.8 mm, it is considered an adjacent feature. The connection boundary between the rib and the welding boss is determined based on the height difference and the projection overlap rate. When the height difference is less than 0.3 mm and the projection overlap rate is greater than 60%, it is considered a connection relationship.

[0032] After determining the connection relationship, the system extracts the center coordinates of the mounting hole, the direction vector of the rib, and the height value of the welding boss, and generates a geometric association data table.

[0033] Using the mounting holes as core nodes, a chain structure is formed by connecting them sequentially according to the direction of the rib vectors; using the welding boss area as branch nodes, connection paths are established through spatial projection directions. Based on this, a mapping structure with "holes as nodes and ribs as connecting edges" is constructed, and the connection strength (calculated based on spatial distance and included angle) is converted into a weighted value to form a weighted topology graph.

[0034] The weighted topology graph is expressed in the form of an adjacency matrix, where the matrix element M_ij represents the spatial association weight between feature i and feature j, and finally generates a topology information matrix.

[0035] Based on the topological information matrix, the system calculates the assembly constraints between the mounting holes and the mating stops.

[0036] The radial clearance is calculated using the difference between the bore diameter D1 and the outer diameter of the stop D2, and is maintained within the range of 0.05mm to 0.15mm. The axial preload is obtained using a pressure formula based on the contact pressure P and contact area A of the mating surfaces, and is limited to the range of 80N to 120N. The calculation results are compiled into a constraint relationship data table, recording the clearance value and preload parameters for each mating point.

[0037] Finally, when the inclination angle of the mating surface is less than 5° and the contact pressure is evenly distributed, it is defined as a surface contact assembly; when the contact area is annular and the angle between the central torque transmission direction and the normal is less than 10°, it is defined as a toroidal assembly; when the welding boss and the adjacent part only contact at the circumferential boundary, it is defined as a line contact assembly.

[0038] Based on the pressure distribution and torque direction of the contact surfaces for each assembly type, assembly relationship data including pressure distribution curves and torque transmission paths is generated.

[0039] Assembly relationship data, combined with automotive part naming rules and installation sequence identifiers, is automatically converted into an assembly semantic data file. The file records assembly levels, part names, assembly sequence numbers, and constraint parameters in JSON format.

[0040] Of particular importance is that the connection boundaries between adjacent feature regions in the computational hardware structure features include: Circumferential sampling is performed on the point cloud at the edge of the mounting hole to extract the sequence of normal vectors for the hole wall; Longitudinal slices were made of the point cloud of the stiffening rib facade to calculate the rib thickness distribution; Region segmentation is performed on the point cloud of the welded boss surface to identify the boss boundary contour; Based on the hole wall normal vector sequence, rib thickness distribution, and boss boundary contour, connection boundary data including mounting hole roundness error, reinforcing rib transition curvature, and welding boss flatness are generated.

[0041] In this embodiment of the invention, the input data consists of a high-fidelity point cloud and a structural feature index table. All point cloud data are expressed in millimeters and uniformly calibrated in a three-dimensional coordinate system.

[0042] The center coordinates and fitting radius of the mounting hole are read from the feature index table. Using the center point as the center and the fitting radius as the radius, uniform angular interval sampling is performed in the hole wall region, with a sampling angle interval of 5°, forming 72 sampling points. For each sampling point, a local neighborhood is established with a neighborhood radius set to 0.2 mm. The normal vector of this neighborhood point is calculated using the least squares plane fitting method. The normal vectors of all sampling points are arranged sequentially to form a hole wall normal vector sequence. The normal deviation sequence is calculated using the angle between adjacent normals to evaluate the hole wall roundness deviation. The hole wall roundness error is obtained using the roundness error calculation formula.

[0043] Statistical analysis of the normal deviation sequence identifies the concentrated distribution area of ​​hole wall roundness error, which serves as the local deformation area at the hole position boundary.

[0044] Extract the starting coordinates and orientation vector of the reinforcing rib region from the feature index table; slice longitudinally along the orientation vector at fixed intervals of 0.5 mm, with each slice having a thickness of 0.1 mm; within each slice plane, extract the boundary points of the point cloud on both sides of the rib and calculate the average distance difference between the two points and the center line; arrange the average thickness values ​​of each slice in order of position to form a thickness distribution sequence; perform linear fitting on the thickness distribution sequence and calculate the thickness change rate; when the change rate is greater than 0.03, the region is determined to be a thickness abrupt change region; in the abrupt change region, calculate the local radius of curvature to describe the geometric continuity of the rib transition fillet.

[0045] The thickness distribution sequence and curvature information are combined to form transition curvature data of the stiffener facade area, which is used to describe the spatial geometric changes of the stiffener boundary.

[0046] Point cloud data of the welding boss area is read from the feature index table to determine the bounding box size of the area. Elevation stratification is performed with a surface height threshold of 0.3 mm, dividing the point cloud into three groups along the Z-axis: high, middle, and low. Density clustering is performed on the middle-layer point set with a cluster radius of 0.4 mm, extracting the density-continuous region as the main body of the boss. The normal variation rate of the boss boundary region is calculated, and points with a normal variation rate greater than 0.1 are marked as boundary points. Curve fitting is performed according to the spatial position order of the boundary points to form the welding boss boundary contour line. Based on the height difference and normal consistency between the points inside and outside the boundary contour line, surface flatness parameters are calculated using the flatness evaluation formula.

[0047] The obtained welding boss boundary contour data records the boss height variation and surface smoothness, providing basic data for determining the integrity of the connection boundary.

[0048] The three types of feature data mentioned above are correlated and matched; the peak point of the hole wall normal deviation corresponding to the abrupt change point of the rib thickness is taken as a set of connection boundary features; when the spatial distance of this set is less than 1mm and the normal angle is less than 15°, it is marked as a valid connection boundary; at the same time, the overlapping part of the weld boss boundary line in this spatial range is retrieved. If the overlap length exceeds 20% of the total boundary length, the area is defined as a "composite connection boundary area".

[0049] The system stores all parameters (roundness error, transition curvature, flatness) of the connected boundary regions as a connected boundary dataset.

[0050] Preferably, constructing a mapping structure with mounting holes as nodes and reinforcing ribs as connecting edges includes: Use the center coordinates of the mounting hole as a topological node to record its three-dimensional spatial position; The direction vector of the stiffener is used as a topological edge to record its connection relationship and mechanical transmission characteristics; A weighted topology graph is constructed based on topology nodes and topology edges, including the distribution density of installation holes and the connectivity of the reinforcing rib network. The weighted topology graph is transformed by adjacency matrix transformation to form a topology information matrix.

[0051] In this embodiment of the invention, the input data comes from a geometric association data table, which contains the center coordinates of the mounting hole, the direction vector of the reinforcing rib, and the spatial distance information between adjacent features.

[0052] Extract the center coordinates of all mounting holes from the geometric correlation data table. Each coordinate point is recorded in millimeters with a coordinate accuracy of ±0.01mm. The center coordinates of each mounting hole are defined as a topology node. The number of nodes is denoted as n, and the node number is incremented according to the hole identification order. Calculate the three-dimensional spatial distribution range of all nodes and determine the overall boundary range using the minimum bounding box.

[0053] Extract the orientation vector of each stiffener from the data table; calculate the node numbers of the two holes it connects to based on the spatial coordinates of the stiffener's start and end points; record the following attributes for each topological edge: orientation vector, stiffener length (unit: mm), and average cross-sectional thickness (unit: mm); for each connection, calculate the mechanical transfer weight value of the stiffener, which is calculated using the mechanical transfer weight calculation formula and is determined by the stiffener cross-sectional thickness, Young's modulus of the material, and length; establish a topological edge data table containing node numbers, connection node numbers, and weight values.

[0054] Import the node set and edge set into the 3D geometry analysis module to form a topology graph G=(N,E); calculate the node distribution density by counting the number of neighboring nodes in a neighborhood with a radius of 5mm for each node, where node distribution density = number of neighboring nodes / neighborhood volume; calculate the connectivity of the rib network by counting the number of topological edges directly connected to the nodes; assign a weighting coefficient to each topological edge; construct a weighted topology graph, where the weight values ​​of each node and its connected edges together describe the spatial and mechanical characteristics of the automotive component structure.

[0055] An adjacency matrix is ​​constructed in the order of node numbers, where the matrix element A_ij represents the connection weight between nodes N_i and N_j. When there is a connecting edge E_ij between nodes, A_ij takes its corresponding weighted value w_ij; if there is no connection, A_ij = 0. Through matrix standardization, the elements in each row of matrix A are normalized to the interval of 0~1 according to the maximum weight to eliminate size differences. Finally, the topology information matrix is ​​output, and the matrix file is output in standard CSV format.

[0056] In another embodiment, the weighted accuracy of the topology graph is enhanced by the node density distribution.

[0057] All hole nodes are projected onto the XY plane to establish a two-dimensional density map. The average number of nodes in the neighborhood (radius 5mm) of each node is calculated. The node density is used to correct the edge weights. The direction angle of the rib direction vector V_j is calculated. When the direction angle is close to the main force direction, the edge weight is increased by 10%. The weighted edge value is calculated by combining the above parameters. The generated weighted topology map is transformed into a topology information matrix through matrix transformation. The matrix is ​​stored with double-precision floating-point number precision (64 bits).

[0058] The final generated topology information matrix records the connection relationship between each installation hole node and the rib, the spatial distance, the connectivity weight, and the mechanical transmission characteristics.

[0059] Preferably, step S3 includes: Extract the feature center coordinates of the mounting holes, the intersection nodes of the reinforcing ribs, and the positioning reference points of the welding bosses from the feature mapping data to generate feature extraction results; Extract semantic location data of each component from the assembly semantic data to generate assembly extraction results; A spatial anchor point set containing key features of the hardware structure is constructed using the feature extraction results and assembly extraction results. Components with spatial anchor points greater than the assembly constraint threshold are selected as global reference components. The overall assembly coordinate frame is calculated with the center of the mounting hole array as the origin and the main direction of the reinforcing ribs as the coordinate axis direction, and registration reference data is generated. Based on the registration reference data, the geometric corresponding points of the mounting holes in the matching feature mapping data are calculated, the position offset and attitude rotation angle between adjacent components are calculated, and attitude matching data is generated. The attitude matching data is fused with the mating stop constraints in the assembly semantic data to correct the position offset and attitude rotation angle, thus forming alignment correction data. Based on the alignment correction data, the translation vector and rotation matrix of each automotive component in the global coordinate system are calculated and combined to generate multi-dimensional registration matrix data.

[0060] In this embodiment of the invention, the set of coordinates of the center of the mounting hole feature, the set of intersection nodes of the reinforcing ribs, and the set of positioning reference points of the welding boss are read from the feature mapping data. The feature coordinate accuracy is limited to ±0.02mm.

[0061] The semantic positioning data of each component is read from the assembly semantic data, including component identification, target assembly center coordinates, assembly sequence number and mating stop constraint values ​​(radial clearance, axial contact height, etc.); the assembly positioning coordinate accuracy is limited to ±0.05mm, and the mating clearance value is set at a resolution of 0.01mm.

[0062] The set of center coordinates of the mounting hole features, the set of intersection nodes of the reinforcing ribs, and the set of positioning reference points of the welding boss are merged into a spatial anchor point set; the number of neighborhood points (neighborhood radius = 5mm) is calculated for each anchor point, and the number of neighborhood points is used to characterize the importance of the anchor point, denoted as d_i (integer); the anchor point importance threshold is set to 3; all anchor points whose d_i is greater than or equal to the anchor point importance threshold are denoted as candidate global references.

[0063] Among the candidate anchor points, based on the assembly constraint strength (sorted by the absolute value of the constraint values ​​in the assembly semantic data), several strongly constrained components are selected as global reference components, and at least three non-collinear hole positions are selected; the geometric fitting center point of the mounting hole array (solved by least squares fitting) is used as the origin of the coordinate system; the direction vector of the main reinforcing ribs is obtained by vector averaging and used as the X-axis direction; a right-handed rectangular coordinate system is constructed for X, Y, and Z, and the accuracy of the coordinate system is that the angle error is less than or equal to 0.1°.

[0064] For each component, calculate the initial translation vector and initial rotation angle from local coordinates to global coordinates (the rotation matrix is ​​obtained by solving the rotation matrix formula); the initial translation accuracy is less than or equal to 0.1 mm, and the initial angle error is less than or equal to 0.5°.

[0065] For each pair of adjacent components A and B, establish a nearest point correspondence between the set of mounting holes of A and the set of corresponding mounting holes of B. The nearest point is defined as the correspondence with the smallest Euclidean distance. Calculate the rigid transformation parameters (translation vector and rotation matrix) of the corresponding point set and solve them iteratively through a least-squares point-to-point registration process. The iteration terminates when the position increment is less than 0.01 mm and the angle increment is less than 0.02°, or when the number of iterations reaches 50. Record the obtained translation vector and rotation matrix as the attitude matching data of the adjacent component pair.

[0066] Read the assembly semantic data for mating stop constraints (radial clearance range, axial preload height); for each registration result, calculate the actual clearance (obtained through the geometric relationship between the hole diameter and the difference in the stop outer diameter); if the actual clearance... The radial clearance range is adjusted by correcting the radial component of the translation vector, and the rotation matrix is ​​recalculated after correction; the upper limit of the correction iteration is 5 times; in the axial direction, the position offset is converted into a force value and compared with the preload height according to the relationship between axial preload and contact stiffness.

[0067] The correction results form an alignment correction data record table, with fields including: component pair identifier, translation vector before correction, rotation matrix before correction, translation vector after correction, rotation matrix after correction, and number of corrections.

[0068] For each component, the final corrected translation vector and rotation matrix are combined into a transformation matrix; the transformation matrices of all components are output in the order of component identification as a multi-dimensional registration matrix data set, with data precision of double-precision floating-point numbers, translation component precision less than or equal to 0.01mm, and rotation matrix element precision less than or equal to 1e-6.

[0069] In another embodiment, the features and anchor points are constructed in the same way as above, and at least four non-coplanar hole positions are selected as the basis points of the constraint equation.

[0070] Point-to-point distance constraint equations are used to fix the relative distance between key holes; point-to-surface normal consistency constraints are used to ensure that the direction of ribs is consistent with the coordinate axis definition after registration; and inter-surface clearance constraint equations are used to limit the radial clearance range of the mating stop.

[0071] The unknown translation and rotation quantities are solved step by step through the sparse linear equation solution process. The solution process adopts an incremental update method. The incremental termination condition is that the absolute value of the change of the unknown quantity is less than the given threshold (translation threshold 0.01mm, angle threshold 0.02°). The maximum number of iterations is 100.

[0072] After the solution is completed, a consistency check is performed, verifying the residual values ​​of all constraint equations one by one. The maximum residual value is limited to 0.02 mm. If the residual exceeds the limit, the component is marked as an abnormal assembly component, and the abnormal item is recorded for manual review.

[0073] Output the final multi-dimensional registration matrix data, in the same format as above.

[0074] Preferably, constructing a spatial anchor point set containing key features of the hardware structure includes: Identify the spatial distribution pattern of the installation hole group and select the center of the area with the highest distribution density as the main positioning anchor point; Key intersection nodes in the reinforcing rib network are extracted and used as auxiliary positioning anchor points; Collect the coordinates of the highest and lowest points of the welding boss to construct a height reference anchor point; The primary positioning anchor point, auxiliary positioning anchor point, and height reference anchor point are sorted by priority to form a hierarchical set of spatial anchor points.

[0075] In this embodiment of the invention, during the construction of the spatial anchor point set, the spatial distribution pattern of the mounting hole group is first identified. A density clustering method is used, with the center point of each mounting hole as the reference, and a search radius of 5mm is set to count the number of neighboring points within an 8mm radius of each point. Regions with more than 6 neighboring points are selected as high-density regions, and the geometric center coordinates of these high-density regions are calculated to determine the primary positioning anchor points.

[0076] When extracting key intersection nodes in the reinforcing rib network, the following steps are taken: collect the centerline point cloud data of all reinforcing ribs and set the intersection angle threshold to 45 degrees; when the included angle of three or more reinforcing rib centerlines is less than the threshold, the location is determined to be a key intersection node; calculate the average three-dimensional coordinates of these intersection nodes and use them as auxiliary positioning anchor points.

[0077] When collecting the coordinates of the highest and lowest points of the welding boss, the following method is used: establish a local coordinate system in the welding boss area, and perform layer sampling at 0.1mm intervals along the vertical direction; record the coordinates of the highest and lowest points of each layer, and take the points with the maximum and minimum Z coordinate values ​​in all sampling layers to construct height reference anchor points.

[0078] The main positioning anchor points, auxiliary positioning anchor points, and height reference anchor points are sorted according to assembly priority: the main positioning anchor point has priority 1, the auxiliary positioning anchor point has priority 2, and the height reference anchor point has priority 3. These anchor points are then stored as an ordered set according to priority, forming a hierarchical spatial anchor point set.

[0079] In another embodiment, a coordinate measuring machine (CMM) is used to acquire the precise coordinates of the mounting holes. The measuring probe moves along the inner wall of the mounting hole at a speed of 0.5 mm / s, acquiring data at one point every 0.2 mm. The center coordinates of each mounting hole are obtained by least squares fitting, and the spatial distribution of the center points of all mounting holes is calculated.

[0080] For the identification of key intersection nodes in the reinforcing rib network, the following specific method is adopted: A line laser scanner is used to acquire surface data of the reinforcing ribs at a resolution of 0.05 mm, and the centerline path of the reinforcing ribs is determined through curvature analysis. When multiple centerlines are detected to intersect within a 3 mm range, and the intersection angle is within the range of 30-60 degrees, the center of this intersection area is determined as an auxiliary positioning anchor point.

[0081] The height reference data for the welding boss was acquired using a contact probe. The probe applied a contact force of 0.1N to perform gridded measurements on the boss surface, with a grid spacing of 0.5mm. The coordinates corresponding to the maximum and minimum Z values ​​in the measurement data were recorded as the height reference anchor points.

[0082] In another embodiment, point cloud data of the automotive parts is acquired using a 3D scanner, with a point cloud density of 25 points per square millimeter. A cylindrical fitting method is used for the mounting hole area: at least 50 point cloud data are selected around each mounting hole, and a cylindrical model is fitted using an iterative nearest-point method. The angular deviation between the cylinder axis and the reference plane is controlled within 2 degrees, and the center point of the cylinder is selected as the candidate point for the main positioning anchor.

[0083] The extraction of the intersection nodes of the reinforcing ribs adopts the following steps: the skeleton of the reinforcing rib point cloud is extracted, and the skeleton line width threshold is 3mm; the intersection points of the skeleton lines are detected. When three or more skeleton lines are detected to intersect in a spherical area with a diameter of 4mm, the weighted center position of these intersection points is calculated. The weight is determined according to the width of the skeleton line. The weight coefficient of the skeleton line with a larger width is 1.2, and that of the skeleton line with a smaller width is 0.8.

[0084] The height reference of the welding boss is determined by elevation analysis: a 5mm×5mm grid is established in the welding boss area, and the extreme value of the Z coordinate of the point cloud in each grid is calculated; the average value of the top 5% of points with the largest Z coordinate in all grids is selected as the highest reference point, and the average value of the top 5% of points with the smallest Z coordinate is selected as the lowest reference point.

[0085] All anchor point data are hierarchically categorized according to their importance in the assembly process: the main positioning anchor points for mounting holes are at level 1, the auxiliary positioning anchor points for reinforcing ribs are at level 2, and the reference anchor points for welding boss heights are at level 3. The final spatial anchor point set includes the three-dimensional coordinates, level number, and feature type information of each anchor point.

[0086] Preferably, correcting the position offset and attitude rotation angle includes: Calculate the vertical displacement compensation based on the difference between the actual height and the theoretical height of the welding boss; Based on the stiffness distribution characteristics of the reinforcing ribs, the attitude rotation angle is elastically deformed and corrected. Adjust the horizontal offset based on the clearance requirements of the mating stop; The various compensation values ​​are superimposed on the original offset and rotation angle to output the alignment correction parameters.

[0087] In this embodiment of the invention, during the correction of position offset and attitude rotation angle, the actual height of the welding boss is first measured. Using a coordinate measuring machine with an accuracy of ±0.01mm, nine measurement points are evenly selected on the surface of the welding boss, arranged in a 3×3 grid with a grid spacing of 2mm. The average Z-coordinate of the nine measurement points is calculated and used as the actual height of the welding boss. The difference between the actual height and the theoretical height is calculated to obtain the vertical displacement compensation. If the theoretical height is 12.50mm and the actual height is 12.45mm, the displacement compensation is +0.05mm.

[0088] When correcting elastic deformation based on the stiffness distribution characteristics of the stiffening ribs, the following method is used: The elastic modulus data of the stiffening ribs is obtained using a material testing machine; the elastic modulus value is 210 GPa. Strain gauges are arranged on the surface of the stiffening ribs with a spacing of 10 mm. After applying a pressure of 0.5 MPa, the deformation of each strain gauge is recorded. The curvature change of the stiffening ribs is calculated based on the deformation, and the curvature change range is controlled within... The curvature change is converted into a posture rotation angle correction value. The conversion formula is: Correction angle = Curvature change × Rib length / 2. If the rib length is 50mm, the curvature change is... The attitude rotation angle correction value is 0.075 degrees.

[0089] When adjusting the horizontal offset based on the clearance requirements of the mating stop, a feeler gauge is used to measure the actual clearance of the mating stop. The feeler gauge has an accuracy of 0.02mm, and the clearance value is measured evenly at 8 positions along the circumference of the mating stop. If the measured clearance value exceeds the design value by 0.1mm, the horizontal offset adjustment is calculated. The adjustment calculation formula is: Offset adjustment = (Measured clearance - Design clearance) / 2. When the design clearance is 0.2mm and the measured clearance is 0.25mm, the horizontal offset adjustment is -0.025mm.

[0090] When adding the compensation values ​​to the original offset and rotation angle, a linear superposition method is used. The original offsets are X=0.1mm, Y=0.05mm, Z=0.02mm, and the original rotation angles are Rx=0.2 degrees, Ry=0.1 degrees, Rz=0.05 degrees. After superposition calculation, the corrected offsets are X=0.075mm, Y=0.025mm, Z=0.07mm, and the corrected rotation angles are Rx=0.125 degrees, Ry=0.175 degrees, Rz=0.05 degrees.

[0091] In another embodiment, a white light scanner with a resolution of 0.02 mm is used to measure the height of the welding boss. A dense scan of the welding boss area is performed, with a point cloud density of 40 points per square millimeter. The average height of the welding boss is calculated using a plane fitting method, with the angle deviation between the fitted plane and the reference plane controlled within 0.5 degrees. The difference between the actual height and the theoretical height is calculated as a vertical displacement compensation after three arithmetic averages.

[0092] The elastic deformation of the stiffener ribs was measured using a laser displacement sensor with an accuracy of 0.001 mm. Twenty measurement points were arranged on the surface of the stiffener ribs, with a spacing of 5 mm. After applying a 1 MPa load, the displacement change at each measurement point was recorded. Based on the displacement change data, the curvature distribution of the stiffener ribs was calculated using a differential method. The curvature calculation formula is as follows: When converting curvature distribution data into attitude rotation angle correction values, an integral calculation method is used with an integration step size of 5 mm.

[0093] A capacitive displacement sensor is used for detecting the gap between the mating surfaces. The sensor has a measurement range of 0-1mm and an accuracy of 0.005mm. Twelve sensors are installed at the mating surfaces, evenly spaced. The readings of each sensor are monitored in real time. When the reading exceeds the design value ±0.02mm, a horizontal offset adjustment is triggered. The adjustment amount is calculated based on the standard deviation of the sensor readings, using the formula: Adjustment Amount = Standard Deviation × Correction Factor, where the correction factor is 0.8.

[0094] The compensation values ​​are superimposed using a sequential superposition method. First, the vertical displacement compensation is superimposed, then the attitude rotation angle correction value is superimposed, and finally the horizontal offset adjustment value is superimposed. A range check is performed after each superposition to ensure the correction parameters are within the allowable range. The allowable range for vertical displacement compensation is ±0.2mm, for attitude rotation angle correction is ±0.3 degrees, and for horizontal offset adjustment is ±0.1mm. This sequential superposition process outputs alignment correction parameters that meet the requirements.

[0095] In another embodiment, a laser tracker is used to measure the height of the welding boss, with a measurement accuracy of ±0.005 mm / m. Five target ball seats are arranged on the surface of the welding boss in a cross shape, with one in the center and one at each of the four corners. The three-dimensional coordinates of each target ball seat are measured, and the minimum value of the Z-coordinate is taken as the actual height of the welding boss. The difference between this actual height and the theoretical height is directly used as the vertical displacement compensation.

[0096] Digital image correlation (DIC) was used to measure the stiffness distribution characteristics of the stiffening ribs. A speckle pattern with a speckle diameter of 0.1 mm was sprayed onto the surface of the stiffening ribs. Deformation images were acquired using two 5-megapixel industrial cameras at a rate of 30 frames per second. The displacement field at each point on the stiffening rib was calculated by comparing the changes in the speckle pattern before and after loading. The displacement field measurement accuracy was 0.01 pixels. The strain distribution of the stiffening rib was calculated based on the displacement field data, with a strain measurement range of 0.001%–0.1%. The strain distribution data was then input into an elasticity formula to calculate the attitude rotation angle correction value.

[0097] The clearance of the mating stop is measured using a pneumatic gauge with a measurement accuracy of 0.001 mm. Eight measuring points are set at 45-degree intervals along the circumference of the mating stop. Each measuring point is measured three times consecutively, and the average value is taken as the clearance value for that point. If the clearance value at any measuring point deviates from the design value by more than 0.05 mm, the horizontal offset adjustment procedure is initiated. The adjustment amount is calculated based on the average clearance deviation of all measuring points, using the following formula: .

[0098] The compensation value superposition process uses a weighted average method. The weight of the vertical displacement compensation is 0.6, the weight of the attitude rotation angle correction is 0.3, and the weight of the horizontal offset adjustment is 0.1. The weighted calculation formula is: Final correction value = Vertical compensation × 0.6 + Rotation correction × 0.3 + Horizontal adjustment × 0.1. Through this weighted calculation, the final alignment correction parameters are output.

[0099] Preferably, calculating the translation vector and rotation matrix of each automotive component in the global coordinate system based on the alignment correction data includes: Independent transformation parameters for each hardware structural feature are calculated based on alignment correction data in the global coordinate system. The position offset vector of the overall centroid of the automotive part is calculated using the fitting center of the mounting hole array as the translation reference. Using the main directional axis of the reinforcing rib as the rotation reference, calculate the attitude rotation angles of the automotive part around the X, Y, and Z axes, and solve for the corresponding rotation matrix; The position offset vector is combined with the rotation matrix to generate rigid transformation parameters; The rigid transformation parameters are iteratively optimized to output the translation vector and rotation matrix.

[0100] In this embodiment of the invention, when calculating the independent transformation parameters of each hardware structural feature based on alignment correction data in the global coordinate system, the following method is adopted: A coordinate measuring machine is used to measure the actual center coordinates of the mounting holes, with a measurement accuracy of ±0.01 mm. For each mounting hole, the coordinates of 24 points evenly distributed on its circumference are collected, and the actual center position is obtained by fitting using the least squares method. The actual center coordinates are compared with the theoretical coordinates, and the independent offset of each mounting hole in the X, Y, and Z directions is calculated.

[0101] When calculating the position offset vector of the overall centroid of an automotive part using the fitted center of the mounting hole array as the translation reference, the following steps are performed: Select the actual center coordinates of all mounting holes on the automotive part, calculate the arithmetic mean of these coordinates to obtain the fitted center of the mounting hole array. Compare this fitted center with the theoretical design position to obtain the position offset vector. The position offset vector is calculated using the vector subtraction formula: Offset vector = Actual fitted center coordinates - Theoretical fitted center coordinates.

[0102] When calculating the attitude rotation angles of an automotive component around the X, Y, and Z axes using the main directional axis of the reinforcing rib as the rotation reference, the following process is adopted: The three-dimensional coordinates of multiple feature points on the reinforcing rib are measured using a laser tracker, with a spacing of 10 mm between the feature points. Principal component analysis is used to determine the direction vector of the main directional axis of the reinforcing rib. This direction vector is compared with the theoretical direction vector, and the rotation angles around each coordinate axis are calculated using vector cross product and dot product operations. The rotation matrix is ​​solved using the Euler angle transformation formula, with the angle values ​​in the transformation formula accurate to 0.001 degrees.

[0103] When combining the position offset vector with the rotation matrix, a 4×4 homogeneous coordinate transformation matrix is ​​used. The position offset vector serves as the translation component of the transformation matrix, and the rotation matrix serves as the rotation component.

[0104] When iteratively optimizing the rigid transformation parameters, the optimization objectives are set as follows: the coaxiality error of the mounting holes is less than 0.05 mm, the coplanarity deviation of the reinforcing ribs is less than 0.1 mm, and the flatness error of the welded boss is less than 0.08 mm. The gradient descent method is used for optimization, adjusting the transformation parameter values ​​in each iteration with a step size of 0.01 mm. Iteration stops when all error indicators simultaneously meet the requirements, and the final translation vector and rotation matrix are output.

[0105] In another embodiment, a laser scanner is used to acquire point cloud data of the automotive part surface, with a point cloud density of 100 points per square millimeter. When calculating the independent transformation parameters of each hardware structural feature based on the alignment correction data, edge point clouds are extracted from the mounting hole area, and the actual axial direction of the mounting hole is obtained through a cylindrical fitting method. The cylindrical fitting uses the least squares method, and the fitting residual is controlled within 0.02 mm.

[0106] When calculating the fitting center of the mounting hole array, the actual center coordinates of all mounting holes are selected, and a weighted average method is used to calculate the fitting center position. The weights are determined according to the diameter of the mounting holes; the weight coefficient for mounting holes with larger diameters is 1.2, and the weight coefficient for mounting holes with smaller diameters is 0.8.

[0107] To determine the principal orientation axis of the stiffener rib, the following steps are taken: Point cloud data of the stiffener rib surface is extracted, and the principal orientation of the point cloud distribution is calculated using eigenvalue decomposition. The eigenvector corresponding to the largest eigenvalue obtained from eigenvalue decomposition is the principal orientation axis direction vector. This direction vector is compared with the theoretical direction vector, and the rotation matrix is ​​calculated using the Rodrigues rotation formula.

[0108] The rigid transformation parameters are combined using a step-by-step combination method. First, the translation transformation is combined, and then the rotation transformation is combined. The multiplication order of the transformation matrices is fixed as translation first, then rotation, to ensure the determinism of the transformation.

[0109] The iterative optimization process is set to a maximum of 100 iterations and a convergence threshold of 0.001 mm. After each iteration, the coaxiality error of the mounting holes, the coplanarity deviation of the reinforcing ribs, and the flatness error of the welded boss are calculated. If the errors do not meet the requirements, the transformation parameters are adjusted according to the magnitude and direction of the errors, with the adjustment amount being 0.1 times the current error value. When the maximum number of iterations is reached or all errors are less than the convergence threshold, the optimized translation vector and rotation matrix are output.

[0110] In another embodiment, a digital photogrammetry system is used to acquire the three-dimensional coordinates of feature points on the surface of the automotive part, with a measurement accuracy of ±0.02 mm. When calculating independent transformation parameters based on alignment correction data, coded marker points with a diameter of 2 mm are placed at the edges of the mounting holes. The three-dimensional coordinates of the marker points are acquired using the photogrammetry system, and the actual position and orientation of each mounting hole are calculated.

[0111] The geometric median method was used to calculate the fitting center of the mounting hole array, and the geometric median was obtained through iterative calculation.

[0112] The main orientation axis of the stiffener was determined using the overall least squares method. Measurement points were arranged on the surface of the stiffener, with a spacing of 5 mm. The equation of the straight line was fitted using the overall least squares method, and the direction vector of the line is the direction of the main orientation axis. The sum of squared residuals from the overall least squares method was controlled within a certain range. Within.

[0113] The rotation matrix is ​​calculated using the quaternion method. The measured principal axis direction vector is converted into quaternion form, where the real part of the quaternion is the cosine of the rotation angle, and the imaginary part is the direction vector of the rotation axis. The rotation matrix is ​​obtained through quaternion multiplication, ensuring its orthogonality.

[0114] The Levenberg-Marquardt method was used for iterative optimization of the rigid transformation parameters. The damping factor was initially set to 0.01, and adjusted in each iteration based on the error change. When the error decreased, the damping factor was multiplied by 0.1; when the error increased, the damping factor was multiplied by 10. The condition number was monitored during optimization to ensure numerical stability. The optimization process terminated when the error decrease rate was less than 0.0001 or the maximum number of iterations (50) was reached, and the final translation vector and rotation matrix were output.

[0115] Preferably, step S4 includes the following steps: Extract the spatial pose parameters and assembly level information of each automotive component from the registration matrix data to form a three-dimensional pose matrix table; Based on the spatial coordinate point set in the three-dimensional pose matrix table, the surface topography features of the automotive parts are resampled using multi-resolution vertex resampling, and a polygon connection structure is established to form a mesh topology table. The mesh topology table is fused with the surface texture vectors in the feature mapping data to construct a three-dimensional surface mapping set; Based on the three-dimensional surface mapping set, the normal distribution, reflectivity and roughness parameters of each region are extracted. The optical characteristic parameters of the inner wall of the mounting hole, the side of the reinforcing rib and the top surface of the welding boss are calculated respectively to generate the illumination parameter set. By superimposing a preset global registration reference coordinate in a three-dimensional coordinate space using a set of lighting parameters, the lighting distribution and shadow mapping of automotive parts are determined based on the spatial distribution of the preset hardware structure, thus forming a three-dimensional rendering scene table. Load material maps, surface reflection models, and ambient light parameters into the 3D rendering scene table, and output a 3D rendering set; The 3D rendering set and registration matrix data are combined and converted into a standardized 3D file format to output a 3D digital model.

[0116] In this embodiment of the invention, when extracting the spatial pose parameters and assembly hierarchy information of each automotive component from the registration matrix data, a structured query method is used to read the transformation matrix from the registration database. The spatial pose parameters of each automotive component include one translation vector and one rotation matrix. The assembly hierarchy information is stored in a tree structure with a depth of no more than 5 levels. These parameters are then organized into a 4×4 homogeneous transformation matrix to form a three-dimensional pose matrix table containing the pose information of all components.

[0117] When performing multi-resolution vertex resampling of automotive part surface topography features based on the spatial coordinate point set in the 3D pose matrix table, a distance-weighted sampling method is used. A sampling density of 50 points per millimeter is set in the mounting hole edge region, 30 points per millimeter in the stiffening rib elevation region, and 40 points per millimeter on the welded boss surface region. During sampling, the vertex density at feature boundaries is maintained at three times that of flat areas. A polygonal connection structure is established using the Delaunay triangulation method, with the maximum side length of triangles limited to 2 mm and the minimum interior angle limited to 15 degrees, forming a complete mesh topology table.

[0118] When fusing the mesh topology table with the surface texture vectors in the feature mapping data, a UV coordinate mapping method is used. A unique texture coordinate is assigned to each vertex, with values ​​ranging from [0,1]. The surface texture vectors are mapped to the mesh surface using bilinear interpolation, with a texture resolution set to 2048×2048 pixels, constructing a 3D surface mapping set containing both geometric and texture information.

[0119] When extracting optical property parameters for each region based on the 3D surface mapping set, a hemispherical reflection measurement method was used in the inner wall region of the mounting hole, with a measurement point spacing of 1 mm, to obtain normal distribution data; reflectivity was measured at a 60-degree incident angle in the side region of the reinforcing rib, with a measurement accuracy of ±0.01; and micromorphology was measured using a surface roughness meter in the top surface region of the welding boss, with a sampling length of 0.8 mm. The generated illumination parameter set contains 256 normal vectors, 128 sets of reflectivity data, and 64 roughness parameters.

[0120] When overlaying global registration reference coordinates onto a lighting parameter set in 3D coordinate space, a bounding box containing all vehicle parts is established with the world coordinate system origin as the reference. The bounding box dimensions are 2000mm × 1500mm × 1000mm. Based on the spatial distribution of the preset hardware structure, ray tracing is used to calculate the lighting distribution and shadow mapping. The light source is set to a parallel light with an intensity of 1000 lux, and the shadow map resolution is set to 4096 × 4096 pixels, forming a 3D rendering scene table.

[0121] When loading material maps into the 3D rendering scene table, a physically-based rendering workflow is used. GGX reflection models are loaded onto metallic surfaces, and Lambertian diffuse reflection models are loaded onto rough surfaces. Ambient light parameters are set as follows: ambient light intensity 0.3, direct light intensity 0.7, and reflected light intensity 0.5. The output is a 3D rendering set containing all optical effects.

[0122] When jointly converting the 3D rendering set and registration matrix data into a standardized 3D file format, the STL file format is used. The number of triangles is controlled to within 500,000, and the file size does not exceed 200MB. All geometric transformation relationships are preserved during the conversion process, and a complete 3D digital model is output.

[0123] Of particular importance is the fusion of the mesh topology table with the surface texture vectors in the feature map data, and the multi-resolution vertex resampling of the surface topography features of automotive parts, including: A high-density sampling strategy is used for the edge area of ​​the mounting hole to preserve the geometric details of the hole wall chamfer. An adaptive sampling algorithm is used for the stiffened rib facade to balance feature preservation and data simplification; A curvature-driven sampling method is used on the surface of the welded boss to accurately reproduce the shape of the boss boundary. Low-density sampling is used in flat areas to optimize the overall grid data volume.

[0124] In this embodiment of the invention, when employing a high-density sampling strategy for the edge region of the mounting hole, a line laser scanner is used to collect point cloud data of the chamfered area of ​​the hole wall at a point spacing of 0.02 mm. The scanning path spirals along the axis of the mounting hole with a pitch of 0.1 mm, collecting 180 data points per revolution. In the chamfered transition region, the point density is increased to 100 points per square millimeter to ensure that the fitting error of the chamfer profile does not exceed 0.005 mm. Adjacent data points are connected using a cubic spline interpolation method to form a continuous chamfer profile line.

[0125] When using adaptive sampling for the facade of the stiffener ribs, the sampling density is dynamically adjusted based on the rate of curvature change. The initial sampling interval is set to 0.5 mm. When the angle between the normal vectors of adjacent points exceeds 5 degrees, supplementary sampling points are inserted in the corresponding area, reducing the sampling interval to 0.1 mm. In the rounded corner area at the root of the stiffener rib, a circular array sampling mode is used, collecting data points at 5-degree intervals with a central angle to ensure complete reproduction of the rounded corner contour.

[0126] When using a curvature-driven sampling method on the surface of a welded boss, the sampling priority is determined by calculating the Gaussian curvature of the surface. The curvature threshold is set to... When the local curvature exceeds the threshold, the sampling density is increased to 80 points per square millimeter. At the boundary contour of the welded boss, an isoparametric sampling method is used to collect data along the boundary normal with a step size of 0.05 mm, and the boundary reconstruction accuracy is controlled within 0.01 mm.

[0127] When using low-density sampling in flat areas, the flatness criterion is set as an angle between the normal vectors of adjacent points being less than 1 degree. A uniform sampling grid with a grid size of 2mm × 2mm is used in this type of area. A boundary matching method is used to ensure a smooth transition between sampling areas of different densities, and the width of the transition area is set to 5mm.

[0128] Please see Figure 3 This is a schematic diagram of an automotive component in this embodiment. The schematic diagram includes automotive component limiting holes and automotive component positioning holes used for automotive structure.

[0129] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is not limited by the foregoing description. Thus, all changes falling within the meaning and scope of the equivalents of the application are intended to be included within the scope of the invention.

[0130] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for generating three-dimensional digital models of automotive parts, characterized in that, Includes the following steps: Step S1: Collect the original morphological data of the automotive parts and construct the original point cloud; input the original point cloud into the pre-trained geometric feature recognition network to extract the feature mapping data of the automotive parts surface. Step S2: Identify the hardware structural features of automotive parts based on feature mapping data; Based on the identified hardware structural features, a topological information matrix of automotive parts is constructed, and their assembly relationships with adjacent parts are extracted to generate assembly semantic data. Step S3: Using feature mapping data and assembly semantic data, combined with the spatial constraint relationship of hardware structural features, establish multi-dimensional registration matrix data; Step S4: Based on the multi-dimensional registration matrix data, perform 3D rendering and solid reconstruction of the geometric contours and hardware structural features of the automotive parts to generate a 3D digital model containing complete assembly semantic information.

2. The method for generating three-dimensional digital models of automotive parts according to claim 1, characterized in that, Step S1 includes: The original morphological data of the automotive parts surface is collected by a 3D scanning device. The original morphological data includes at least the edge contour of the mounting hole, the vertical height of the reinforcing rib, and the surface undulation of the welding boss. Redundant point cloud removal and noise filtering are performed on the original topographic data to generate a high-fidelity original point cloud set. The high-fidelity original point cloud is input into a pre-trained geometric feature recognition network, and the local curvature texture information and hardware structure boundary of the automotive part surface are extracted through the convolutional coding layer of the geometric feature recognition network. Based on the extracted local curvature texture information, the roundness error of the mounting hole, the root transition area of ​​the reinforcing rib, and the boundary contour of the welding boss are identified and corrected. The output is the corrected feature map data, which contains the spatial coordinates and geometric properties of the hardware structure features.

3. The method for generating three-dimensional digital models of automotive parts according to claim 1, characterized in that, Step S2 includes: Based on feature mapping data, the hardware structural features of automotive parts are identified, including mounting holes, reinforcing ribs, welding bosses, and mating stops. Calculate the connection boundary between adjacent feature regions in the hardware structure features, and generate geometric association data containing the center coordinates of the mounting hole, the direction vector of the stiffening rib, and the height information of the welding boss; The geometric correlation data is grouped according to the boundary features of automotive parts, and a mapping structure is constructed with mounting holes as nodes and reinforcing ribs as connecting edges, forming a topological information matrix that includes the spatial relationships of the hardware structure. Based on the topological information matrix, the assembly constraint conditions between the mounting hole and the mating stop are calculated, and constraint relationship data including radial clearance and axial preload are generated. Based on the constraint relationship data, the assembly type of the welding boss and adjacent components is identified, and assembly relationship data including the pressure distribution of the contact surface and the torque transmission path is formed. Based on preset automotive part naming rules and installation sequence identifiers, assembly relationship data is transformed into assembly semantic data.

4. The method for generating a three-dimensional digital model of automotive parts according to claim 3, characterized in that, The construction of a mapping structure with mounting holes as nodes and reinforcing ribs as connecting edges includes: Use the center coordinates of the mounting hole as a topological node to record its three-dimensional spatial position; The direction vector of the stiffener is used as a topological edge to record its connection relationship and mechanical transmission characteristics; A weighted topology graph is constructed based on topology nodes and topology edges, including the distribution density of installation holes and the connectivity of the reinforcing rib network. The weighted topology graph is transformed into an adjacency matrix to form a topology information matrix.

5. The method for generating three-dimensional digital models of automotive parts according to claim 1, characterized in that, Step S3 includes: Extract the feature center coordinates of the mounting holes, the intersection nodes of the reinforcing ribs, and the positioning reference points of the welding bosses from the feature mapping data to generate feature extraction results; Extract semantic location data of each component from the assembly semantic data to generate assembly extraction results; A spatial anchor point set containing key features of the hardware structure is constructed using the feature extraction results and assembly extraction results. Components with spatial anchor points greater than the assembly constraint threshold are selected as global reference components. The overall assembly coordinate frame is calculated with the center of the mounting hole array as the origin and the main direction of the reinforcing ribs as the coordinate axis direction, and registration reference data is generated. Based on the registration reference data, the geometric corresponding points of the mounting holes in the matching feature mapping data are calculated, the position offset and attitude rotation angle between adjacent components are calculated, and attitude matching data is generated. The attitude matching data is fused with the mating stop constraints in the assembly semantic data to correct the position offset and attitude rotation angle, thus forming alignment correction data. Based on the alignment correction data, the translation vector and rotation matrix of each automotive component in the global coordinate system are calculated and combined to generate multi-dimensional registration matrix data.

6. The method for generating a three-dimensional digital model of automotive parts according to claim 5, characterized in that, Constructing a spatial anchor set containing key features of the hardware structure includes: Identify the spatial distribution pattern of the installation hole group and select the center of the area with the highest distribution density as the main positioning anchor point; Key intersection nodes in the reinforcing rib network are extracted and used as auxiliary positioning anchor points; Collect the coordinates of the highest and lowest points of the welding boss to construct a height reference anchor point; The primary positioning anchor point, auxiliary positioning anchor point, and height reference anchor point are sorted by priority to form a hierarchical set of spatial anchor points.

7. The method for generating a three-dimensional digital model of automotive parts according to claim 5, characterized in that, Corrections for position offset and attitude rotation angle include: Calculate the vertical displacement compensation based on the difference between the actual height and the theoretical height of the welding boss; Based on the stiffness distribution characteristics of the reinforcing ribs, the attitude rotation angle is elastically deformed and corrected. Adjust the horizontal offset based on the clearance requirements of the mating stop; The various compensation values ​​are superimposed on the original offset and rotation angle to output the alignment correction parameters.

8. The method for generating a three-dimensional digital model of automotive parts according to claim 5, characterized in that, Based on the alignment correction data, the translation vector and rotation matrix of each vehicle component in the global coordinate system are calculated as follows: Independent transformation parameters for each hardware structural feature are calculated based on alignment correction data in the global coordinate system. The position offset vector of the overall centroid of the automotive part is calculated using the fitting center of the mounting hole array as the translation reference. Using the main directional axis of the reinforcing rib as the rotation reference, calculate the attitude rotation angles of the automotive part around the X, Y, and Z axes, and solve for the corresponding rotation matrix; The position offset vector is combined with the rotation matrix to generate rigid transformation parameters; The rigid transformation parameters are iteratively optimized to output the translation vector and rotation matrix.

9. The method for generating a three-dimensional digital model of automotive parts according to claim 1, characterized in that, Step S4 includes the following steps: Extract the spatial pose parameters and assembly level information of each automotive component from the registration matrix data to form a three-dimensional pose matrix table; Based on the spatial coordinate point set in the three-dimensional pose matrix table, the surface topography features of the automotive parts are resampled using multi-resolution vertex resampling, and a polygon connection structure is established to form a mesh topology table. The mesh topology table is fused with the surface texture vectors in the feature mapping data to construct a three-dimensional surface mapping set; Based on the three-dimensional surface mapping set, the normal distribution, reflectivity and roughness parameters of each region are extracted. The optical characteristic parameters of the inner wall of the mounting hole, the side of the reinforcing rib and the top surface of the welding boss are calculated respectively to generate the illumination parameter set. By superimposing a preset global registration reference coordinate in a three-dimensional coordinate space using a set of lighting parameters, the lighting distribution and shadow mapping of automotive parts are determined based on the spatial distribution of the preset hardware structure, thus forming a three-dimensional rendering scene table. Load material maps, surface reflection models, and ambient light parameters into the 3D rendering scene table, and output a 3D rendering set; The 3D rendering set and registration matrix data are combined and converted into a standardized 3D file format to output a 3D digital model.

10. A three-dimensional digital model generation system for automotive parts, characterized in that, For performing the method for generating a three-dimensional digital model of an automotive part as described in claim 1, the system for generating a three-dimensional digital model of an automotive part comprises: The original shape acquisition and point cloud construction module is used to acquire the original shape data of automotive parts and construct the original point cloud set; the original point cloud set is input into a pre-trained geometric feature recognition network to extract the feature mapping data of the automotive part surface. The geometric feature recognition and feature mapping generation module is used to identify the hardware structure features of automotive parts based on feature mapping data; based on the identified hardware structure features, it constructs a topological information matrix of the automotive parts and extracts their assembly relationship with adjacent components to generate assembly semantic data. The structural semantic parsing and topological semantic modeling module is used to establish multi-dimensional registration matrix data by utilizing feature mapping data and assembly semantic data, combined with the spatial constraint relationship of hardware structural features. The spatial registration and 3D reconstruction rendering module is used to perform 3D rendering and solid reconstruction of the geometric contours and hardware structural features of automotive parts based on multi-dimensional registration matrix data, generating a 3D digital model containing complete assembly semantic information.

Images