CN-121999176-A - White car body parameterized grid generation system integrating cross-modal mapping and constant-variation characterization

CN121999176ACN 121999176 ACN121999176 ACN 121999176ACN-121999176-A

Abstract

The invention belongs to the technical field of computer aided design, and relates to a white bodywork parameterized grid generating system integrating cross-modal mapping and isomorphism characterization, which comprises an instruction analyzing and packaging module, a graph operation instruction packet, a data processing module and a data processing module, wherein the instruction analyzing and packaging module is used for generating a graph operation instruction packet containing hash information based on a cross-modal instruction and a reference topological graph; the system comprises a parameter injection and verification module, a constraint guiding isomorphism generation module, an auxiliary data generation module, a guided deterministic decoding module and a guided deterministic decoding module, wherein the parameter injection and verification module is used for injecting a topological graph copy into an initial graph after verification is passed, the constraint guiding isomorphism generation module is used for injecting engineering constraint correction signals in interlayer iteration of an isomorphism graph neural network to generate an isomorphism representation, the auxiliary data generation module is used for generating a geometric validity mask and a local differential attribute tensor in parallel according to the isomorphism representation, and the guided deterministic decoding module is used for constructing an isosurface by using the mask and guiding grid division and mapping according to the tensor. The method solves the problems of long design iteration period and data incoherence caused by decoupling of design intent and geometric generation and engineering constraint verification hysteresis in the prior art.

Inventors

QIN JUN
MAO XIANXIN
LEI JIAN
ZHANG JINGRU

Assignees

北京领翼工软科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260119

Claims (10)

1. The white body parameterized grid generation system integrating cross-modal mapping and constant-change characterization is characterized by comprising the following steps: The instruction analysis and encapsulation module is used for acquiring a cross-modal design instruction and generating a parameterized graph operation instruction packet containing hash check information and an instruction sequence based on a preset white bodywork reference topological graph; The parameter injection and verification module is used for receiving the parameterized graph operation instruction packet, and injecting parameters in the instruction sequence into a copy of the white bodywork reference topological graph after the hash verification information and the instruction sequence are verified, so as to generate a parameterized initial graph; The constraint-guided isomorphism generation module is used for inputting the parameterized initial graph into the isomorphism graph neural network, injecting a correction signal based on engineering constraint in the interlayer iteration process, and generating a parameterized isomorphism representation carrying the engineering constraint; The auxiliary data generation module is used for generating branches and differential attribute calculation branches through masks based on parameterization isomorphism characterization carrying engineering constraints, and respectively generating a geometric validity mask for defining boundaries and a local differential attribute tensor for characterizing curved surface features; and the guided deterministic decoding module is used for constructing an entity equivalent surface by utilizing the geometric effectiveness mask, guiding grid division and node coordinate mapping according to the local differential attribute tensor and generating a white bodywork parameterized grid.
2. The system for generating a parameterized grid for body in white with fused cross-modal mapping and isomorphism characterization of claim 1, wherein the instruction parsing and packaging module is configured to perform the following operation steps: analyzing the cross-modal design instruction, and extracting geometric targets, parameter types and parameter values containing design intents; loading a white bodywork reference topological graph and calculating a hash value of the white bodywork reference topological graph as hash check information; mapping the extracted geometric targets into topological graph node identifiers, and packaging the node identifiers, parameter types, parameter values and influence weights into four-tuple instructions; and combining the hash check information with an instruction sequence formed by a plurality of tetrad instructions to construct a parameterized graph operation instruction packet.
3. The system for generating a parameterized grid for body in white with fusion of cross-modal mapping and invariant features of claim 1, wherein the parameter injection and verification module is configured to perform the following steps: analyzing a parameterized graph operation instruction packet, and comparing hash check information in the packet with a real-time hash value of a currently loaded white body reference topological graph; After Ha Xibi pairs are consistent, verifying whether node identifiers in the instruction sequence exist in the white bodywork reference topological graph one by one; after all the verification passes, creating a copy of the white bodywork reference topological graph, and positioning the nodes according to the node identifiers in the instruction sequences; And locking the corresponding dimension of the node feature vector according to the parameter type in the instruction sequence, and injecting the product of the parameter value and the influence weight into the corresponding dimension to form a parameterized initial graph.
4. The system for generating a parameterized grid of body in white with fusion of cross-modal mapping and invariant features of claim 1, wherein the constraint-guided invariant generation module is configured to perform the following steps: Forward propagation is carried out on the parameterized initial graph and the isomorphic graph neural network, and scalar and vector features of neighborhood nodes are aggregated through message passing in each layer of iteration; Inputting the intermediate geometric representation generated by the current layer into a preset engineering constraint agent model, and predicting the risk probability distribution of each node violating engineering constraint; calculating a correction vector pointing to a risk reduction direction according to the risk probability distribution, and combining to form a correction force field signal; And (3) injecting the corrected force field signal as a modulation item into the node vector feature updating process of the next layer iteration until all the layer iterations are completed, and outputting parameterized isomorphism characterization carrying engineering constraints.
5. The system for generating a parameterized grid for body in white with fusion of cross-modal mapping and invariant characterizations according to claim 1, wherein the assistance data generating module is configured to perform the following steps: generating branches by parameterized isomorphism representation input masks carrying engineering constraints, predicting the entity existence probability of each voxel in a voxel grid through a three-dimensional convolution network, and generating a binarized geometric effectiveness mask after threshold processing; Inputting parameterized invariant features carrying engineering constraints into differential attribute calculation branches, and determining local neighborhood point clouds of each node based on a K neighbor algorithm; performing principal component analysis or differential operation on the local neighborhood point cloud, and calculating a normal vector and a curvature value of the local fitting plane; And (5) collecting normal vectors and curvature values of all nodes to construct a local differential attribute tensor.
6. The system for generating a parameterized mesh for body-in-white with fused cross-modal mapping and isomorphism characterization of claim 1, wherein the guided deterministic decoding module is configured to perform the following steps: Processing the geometric effectiveness mask by using an isosurface extraction algorithm, and constructing triangular patches in the voxel units to form closed entity isosurfaces; dividing grid cells in the entity isosurface, and dynamically adjusting the size and arrangement direction of local grid cells according to curvature values in local differential attribute tensors; Establishing a mapping relation between node coordinates and newly generated grid vertices in parameterized isomorphism representation carrying engineering constraints by adopting a radial basis function interpolation method; And calculating the final three-dimensional coordinates of the grid vertices based on the mapping relation to generate the body-in-white parameterized grid.
7. The system for generating the parameterized grid of the body-in-white fused cross-modal mapping and isomorphism representation according to claim 2, wherein if a plurality of instructions are modified for the same parameter type of the same geometric target, the influence weight is calculated according to a preset constraint intensity priority strategy, or the parameter value of a high priority instruction is directly adopted when the priority difference exceeds a threshold value.
8. The system for generating a parameterized grid for body-in-white fused cross-modal mapping and isomorphism representation of claim 4, wherein the process of computing the corrected force field signal comprises setting a risk threshold, computing a gradient direction of a risk probability value of a node with a value exceeding the risk threshold with respect to a node vector feature input to the engineering constraint proxy model, taking the gradient direction as a correction vector, and setting the correction vector of the node with a value lower than the risk threshold in the risk probability distribution as a zero vector.
9. The system for generating a parameterized grid for a body in white by fusing cross-modal mapping and invariant feature of claim 5, wherein calculating the curvature value comprises constructing a covariance matrix of the local neighborhood point cloud, calculating all eigenvalues of the covariance matrix, and taking a ratio of the eigenvalue with the smallest value to a sum of all eigenvalues as the estimated curvature value.
10. The system for generating a parameterized grid for a body in white by fusing cross-modal mapping and isomorphism representation as recited in claim 6, wherein dynamically adjusting the size of the local grid cells comprises setting a predetermined curvature threshold, comparing a curvature value in the local differential attribute tensor with the predetermined curvature threshold, generating grid cells having a size smaller than a predetermined reference size in a region where the curvature value is greater than the predetermined curvature threshold, and generating grid cells having a size greater than the predetermined reference size in a region where the curvature value is less than the predetermined curvature threshold.

Description

White car body parameterized grid generation system integrating cross-modal mapping and constant-variation characterization Technical Field The invention belongs to the technical field of computer aided design, and relates to a white body parameterized grid generation system integrating cross-modal mapping and constant-variation characterization. Background In the development process of white automobile body, a computer aided design system is a core tool for geometric modeling and structural design. Designers typically utilize such software to construct three-dimensional digital models of body-in-white by manipulating geometric primitives such as control points, curves, surfaces, and the like. For parameterized modifications, constraint relationships between geometries and driving parameters are generally predefined, and the local or global deformation of the model is subsequently achieved by modifying the values of these parameters. After the geometric model of the design scheme is initially completed, the model needs to be converted into a grid model suitable for finite element analysis, and then the grid model is imported into computer aided engineering software to perform simulation analysis on the performances such as structural rigidity, collision safety and vibration noise. In the existing design and analysis process, a designer needs to manually convert a high-level design intent such as enhancing the strength of a specific component into a series of geometric operations of the bottom layer, such as selecting a specific curved surface, increasing the thickness or adjusting the angle. This parameterized modeling approach relies primarily on pre-fixed logical relationships based on script or historical features. Meanwhile, the simulation of engineering performance is usually executed as an independent and serial step after geometric design, namely, after the geometric model is changed, the grid is regenerated and complete simulation calculation is performed, so that performance feedback can be obtained. However, the serial and manual translation-dependent processing described above has technical limitations in coping with complex design requirements. On one hand, the lack of formal mapping mechanism between the design intent and the underlying geometric operation leads to semantic gap between the design intent and the underlying geometric operation, and on the other hand, the parameterized model based on preset logic has limited flexibility, is difficult to adapt to design changes beyond the preset parameterized category or involving topological structure changes, and often faces the requirement of model reconstruction. In addition, the data isolation among the geometric design, the parameterized script and the simulation analysis and the off-line simulation verification mode lead to engineering performance feedback lag, so that potential defects of the design scheme are easily found at the tail end of the design flow, and the cost of repeated verification and the whole development period are increased. Disclosure of Invention In order to solve the problems, the invention provides a white body parameterized grid generation system integrating cross-modal mapping and invariant characterization. A body-in-white parameterized grid generation system that fuses cross-modal mapping with invariant characterizations, comprising: The instruction analysis and encapsulation module is used for acquiring a cross-modal design instruction and generating a parameterized graph operation instruction packet containing hash check information and an instruction sequence based on a preset white bodywork reference topological graph; The parameter injection and verification module is used for receiving the parameterized graph operation instruction packet, and injecting parameters in the instruction sequence into a copy of the white bodywork reference topological graph after the hash verification information and the instruction sequence are verified, so as to generate a parameterized initial graph; The constraint-guided isomorphism generation module is used for inputting the parameterized initial graph into the isomorphism graph neural network, injecting a correction signal based on engineering constraint in the interlayer iteration process, and generating a parameterized isomorphism representation carrying the engineering constraint; The auxiliary data generation module is used for generating branches and differential attribute calculation branches through masks based on parameterization isomorphism characterization carrying engineering constraints, and respectively generating a geometric validity mask for defining boundaries and a local differential attribute tensor for characterizing curved surface features; and the guided deterministic decoding module is used for constructing an entity equivalent surface by utilizing the geometric effectiveness mask, guiding grid division and node coordinate mapping according to the local