Part assembly method based on geometric topology fusion

Abstract

The invention discloses a geometric topology fusion-based part assembly method, which comprises the following steps of: acquiring boundary representation models of at least two to-be-assembled CAD parts, and converting the boundary representation model of each CAD part into a structured heterogeneous geometric topology graph; performing feature coding on each node in the heterogeneous geometric topological graph to form node feature representation; inputting the node feature representation into a graph attention reasoning module, and obtaining a node embedding representation representing a geometrical relationship and a topological relationship in the CAD part through multi-layer residual graph structure information propagation and feature aggregation; based on the node embedding representations of the different parts, calculating association scores of node pairs among the different parts; and performing supervised training on the feature coding module and the graph attention reasoning module by using the labeled assembly constraint sample data. According to the method, fine geometric features can be extracted from a boundary representation B-Rep model, and a complex topological dependency relationship is inferred, so that robust assembly constraint inference is realized.

Description

Technical Field

[0001] This invention relates to the field of automated parts assembly technology, and in particular to a parts assembly method based on geometric topology fusion. Background Technology

[0002] The current automated modeling of CAD parts assembly mainly relies on the understanding and reasoning of the geometric and topological relationships of the parts. Existing methods related to or that can be referenced for parts assembly can be roughly divided into the following categories: (1) Traditional rule-driven method: Based on manually defined geometric features (such as holes, shafts, and planes) and heuristic rules (such as nearest neighbor matching and coaxiality detection) to infer the assembly constraints (such as concentricity and fit) between parts. This method heavily relies on clear, standard geometric features and complete feature annotations. (2) Deep learning method based on point cloud or triangular mesh: Represent the parts as point clouds or meshes, and use deep learning models to predict the degrees of freedom of the parts or to reassemble fragments to achieve assembly. (3) Graph neural network method based on boundary representation: Transform the inherent B-Rep structure of CAD parts into graph data, use graph convolutional networks (GCN) or graph attention networks (GAT) to learn node (face, edge) features, and predict the assembly relationships between nodes.

[0003] The above-mentioned existing technologies have the following defects: (1) Traditional rule-based methods and existing B-Rep methods rely on manually designed geometric descriptors (such as area and curvature), which cannot effectively capture the subtle features of complex geometry such as free-form surfaces and weakly defined interfaces, resulting in a significant performance drop in non-hole and non-standard assembly scenarios. (2) Insufficient topological relationship modeling: The existing graph neural network structure based on B-Rep is relatively shallow and has a limited receptive field, making it difficult to model the complex long-range topological dependencies within and between parts, resulting in poor inference performance. (4) Feature representation is disconnected from design intent: Point cloud or mesh-based methods lose the explicit geometric entities (such as precise surfaces and boundaries) and parametric semantics in B-Rep, making it difficult to reliably infer the assembly constraint types with clear engineering semantics. Summary of the Invention

[0004] The purpose of this invention is to provide a part assembly method based on geometric topology fusion. This method can extract fine geometric features from the boundary representation model B-Rep and infer complex topological dependencies, thereby achieving robust assembly constraint inference.

[0005] The objective of this invention is achieved through the following technical solution: A part assembly method based on geometric topology fusion, the method comprising: Step 1: Obtain the boundary representation models of at least two CAD parts to be assembled, and convert the boundary representation model of each CAD part into a structured heterogeneous geometric topology graph; wherein, the heterogeneous geometric topology graph includes face nodes representing geometric surfaces, edge nodes representing geometric edges, and graph edges used to characterize the topological connection relationship between face nodes and edge nodes. Step 2: Perform feature encoding on each node in the heterogeneous geometric topology graph obtained in Step 1, and perform gated adaptive fusion of the encoded geometric features and attribute features to form a node feature representation; wherein, the feature encoding includes sampling the geometric entity corresponding to the node based on geometric parameterization to obtain geometric features, and encoding the type attribute of the geometric entity corresponding to the node to obtain attribute features. Step 3: Input the node feature representation obtained in Step 2 into the graph attention inference module. Through multi-layer residual graph structure information propagation and feature aggregation, obtain the node embedding representation that represents the internal geometric and topological relationships of the CAD part. Step 4: Based on the node embedding representation of different parts, calculate the association score of node pairs between different parts to predict whether there is an assembly constraint relationship between the geometric entities corresponding to the node pairs. Step 5: Using the labeled assembly constraint sample data, supervise the training of the feature encoding module and the graph attention inference module to optimize the assembly constraint prediction results.

[0006] As can be seen from the technical solution provided by the present invention, the above method can extract fine geometric features from the boundary representation B-Rep model and infer complex topological dependencies, thereby achieving robust assembly constraint inference. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of a part assembly method based on geometric topology fusion provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of the B-Rep model and its corresponding heterogeneous geometric topology as described in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments, and do not constitute a limitation of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0011] like Figure 1 The diagram shows a flowchart of a part assembly method based on geometric topology fusion provided in an embodiment of the present invention. The method includes: Step 1: Obtain the boundary representation model (B-Rep) of at least two CAD parts to be assembled, and convert the boundary representation model of each CAD part into a structured heterogeneous geometric topology diagram. The heterogeneous geometric topology graph includes face nodes representing geometric surfaces, edge nodes representing geometric edges, and graph edges used to characterize the topological connection relationship between face nodes and edge nodes. In this step, firstly, based on the boundary representation model data of the input CAD part, all face and edge entities are extracted as a candidate set of graph nodes; Then, a heterogeneous geometric topology graph is constructed, extracting the faces and edges of the CAD parts and their connections, and representing them as a graph structure. Both face nodes and edge nodes are treated as nodes in the graph; where, Let be the set of nodes in the graph. For a set of face nodes, A set of edge nodes; edge set Used to encode the topological relationships between B-Rep elements, i.e., the boundary represents the connection between faces and edges in the model.

[0012] For example, if a face is defined by a boundary, then an undirected edge is created between the corresponding face node and edge node in the graph. Figure 2 As shown, a B-Rep model and its corresponding heterogeneous geometric topology are illustrated.

[0013] Step 2: Encode the features of each node in the heterogeneous geometric topology graph obtained in Step 1, and perform gating adaptive fusion of the encoded geometric features and attribute features to form a node feature representation; The feature encoding includes sampling the geometric entity corresponding to the node based on geometric parameterization to obtain geometric features, and encoding the type attributes of the geometric entity corresponding to the node to obtain attribute features. In this step, for each face node in the heterogeneous geometric topology graph, a uniform mesh (e.g., 10x10) is sampled in the two-dimensional UV parameter domain. U and V refer to two parameters in the parametric surface; in CAD, a surface parameter domain is typically defined by two parameters (U, V) (similar to a two-dimensional coordinate system). Then, two feature meshes (UV) containing coordinates, normal vectors, and trimming masks are generated. , The sampled area represents the real number field, and n represents the number of sampling points; its geometric features are extracted using a two-dimensional convolutional network in UV_Net. Simultaneously, the basic properties of the surface, including surface type and area, are encoded into attribute features using a multilayer perceptron (MLP). ; i represents the i-th node; Geometric features and attribute features are combined to form a unified feature for surface nodes. , represented as: (1) Then, element-wise gating of the geometric features is calculated using linear projection and the sigmoid function. and element-wise gating of attribute features , represented as: , (2) in, This represents the weight matrix used to generate the geometric feature gating weights for the face nodes. This represents the corresponding bias vector; This represents the weight matrix used to generate the gating weights for the attribute features of the face nodes. This represents the corresponding bias vector; This represents the sigmoid activation function; Then, by gating each modality and splicing them together, the fused face node embedding is obtained. , represented as: (3) in This represents a vector concatenation operation; This represents element-wise multiplication. For each edge node in the heterogeneous geometric topology graph, uniform sampling is performed in the one-dimensional parameter domain to generate features containing coordinates and tangent vectors. Geometric features are extracted using a one-dimensional convolutional network in UV_Net. Simultaneously, the basic attributes of the edges, including curve type and length, are encoded into attribute features using a multilayer perceptron (MLP). Geometric features and attribute features are combined to form a unified feature for edge nodes. , represented as: (4) Then, element-wise gating of the geometric features is calculated using linear projection and the sigmoid function. and element-wise gating of attribute features , represented as: , (5) in, This represents the weight matrix used to generate the geometric feature gating weights of the edge nodes. This represents the corresponding bias vector; This represents the weight matrix used to generate the gating weights for edge node attribute features. This represents the corresponding bias vector; Then, by gating each modality and concatenating them, the fused edge node embedding is obtained. , represented as: (6) By sequentially concatenating the feature embeddings of all face nodes and edge nodes, we obtain: (7) in The number of face nodes; This represents the number of edge nodes; Indicates a splicing operation; Finally, the node feature matrix of the heterogeneous geometric topology diagram of the parts is obtained. ,in This represents the total number of face nodes and boundary points, with each node having a feature dimension of 448.

[0014] Step 3: Input the node feature representation obtained in Step 2 into the graph attention inference module. Through multi-layer residual graph structure information propagation and feature aggregation, obtain the node embedding representation that represents the internal geometric and topological relationships of the CAD part. In this step, the graph attention reasoning module is composed of The residual graph attention network is stacked in layers, which is composed of GATv2. For the first layer... The layer is calculated as follows: (8) Formula (8) indicates that the output of the previous layer is used as the input of this layer, where l = 0, 1, 2, 3, 4, and , This is the node feature matrix obtained in step 2; Let be the edge set of the graph; ELU is the activation function; residual connections ensure training stability and preservation of low-level features; For two parts to be assembled and It uses the same graph attention network stack with shared parameters for processing, and outputs node embedding representations with high-level semantic enhancement. and This refers to the geometric-topological dependencies within CAD parts, ensuring the consistency of the feature semantic space.

[0015] Step 4: Based on the node embedding representation of different parts, calculate the association score of node pairs between different parts to predict whether there is an assembly constraint relationship between the geometric entities corresponding to the node pairs. In this step, for the parts Nodes in and parts Nodes in Embedding nodes to represent and The data is concatenated, and the result is input into a multilayer perceptron (MLP) to obtain the association score of the node pairs. (9) in Indicates parts Nodes in and parts Nodes in The correlation score between them; Construct an association score matrix from the association scores of all node pairs. , represented as: (10) in and Parts and parts The higher the number of nodes, the better. The greater the likelihood that there is a valid assembly constraint between node i and node j.

[0016] Step 5: Using the labeled assembly constraint sample data, supervise the training of the feature encoding module and the graph attention inference module to optimize the assembly constraint prediction results.

[0017] In this step, the constructed loss function includes maximum likelihood estimation loss and symmetry loss, where: Maximum likelihood estimation loss Represented as: (11) (12) in Let k be the total number of candidate node pairs; For the first Association score of each candidate node pair; For the first Normalized true probability of each candidate node pair; Symmetry loss Represented as: (13) in It is a correlation score matrix The transpose of the constraint connection matrix is ​​used to reflect the two-way physical characteristics of the assembly relationship. Total training loss Represented as: (14) Where B represents the batch during training, and the number of training samples.

[0018] It is worth noting that the contents not described in detail in the embodiments of the present invention belong to the prior art known to those skilled in the art.

[0019] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method.

[0020] This invention also provides a computer storage medium storing a plurality of instructions adapted for loading and executing the method by a processor.

[0021] In summary, the method described in the embodiments of the present invention has the following advantages: (1) By introducing a geometry encoding module based on UV parameterization, the ability to perceive local surface details of parts (especially freeform surfaces) is enhanced, thus solving the problem of insufficient geometric representation ability caused by the reliance on manual features in existing methods; (2) By designing a graph attention reasoning module, the model's ability to model long-range topological dependencies within and between parts is enhanced, thus solving the problem of weak global reasoning ability of existing shallow graph networks; (3) Through the collaborative design of geometric coding and graph reasoning, the adaptive fusion of geometric information and topological information is realized, which significantly improves the assembly accuracy and generalization ability in complex scenarios such as "non-hole assembly", and solves the problem of performance degradation of existing methods in such scenarios; (4) Experiments on publicly available benchmark datasets show that our method outperforms state-of-the-art methods in several key metrics, demonstrating its ability to handle complex geometric conditions in real industrial scenarios. This enables it to be directly integrated into modern CAD systems as a reliable intelligent assembly recommendation engine.

[0022] Furthermore, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0023] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims. The information disclosed in the background section is intended only to enhance the understanding of the overall background technology of the present invention and should not be construed as an admission or implication in any way that such information constitutes prior art known to those skilled in the art.

Claims

1. A part assembly method based on geometric topology fusion, characterized in that, The method includes: Step 1: Obtain the boundary representation models of at least two CAD parts to be assembled, and convert the boundary representation model of each CAD part into a structured heterogeneous geometric topology graph; wherein, the heterogeneous geometric topology graph includes face nodes representing geometric surfaces, edge nodes representing geometric edges, and graph edges used to characterize the topological connection relationship between face nodes and edge nodes. Step 2: Perform feature encoding on each node in the heterogeneous geometric topology graph obtained in Step 1, and perform gated adaptive fusion of the encoded geometric features and attribute features to form a node feature representation; wherein, the feature encoding includes sampling the geometric entity corresponding to the node based on geometric parameterization to obtain geometric features, and encoding the type attribute of the geometric entity corresponding to the node to obtain attribute features. Step 3: Input the node feature representation obtained in Step 2 into the graph attention inference module. Through multi-layer residual graph structure information propagation and feature aggregation, obtain the node embedding representation that represents the internal geometric and topological relationships of the CAD part. Step 4: Based on the node embedding representation of different parts, calculate the association score of node pairs between different parts to predict whether there is an assembly constraint relationship between the geometric entities corresponding to the node pairs. Step 5: Using the labeled assembly constraint sample data, supervise the training of the feature encoding module and the graph attention inference module to optimize the assembly constraint prediction results.

2. The part assembly method based on geometric topology fusion according to claim 1, characterized in that, In step 1, all face and edge entities are extracted as a candidate set of graph nodes based on the boundary representation model data of the input CAD part. Then, a heterogeneous geometric topology graph is constructed, extracting the faces and edges of the CAD parts and their connections, and representing them as a graph structure. Both face nodes and edge nodes are treated as nodes in the graph; where, Let be the set of nodes in the graph. For a set of face nodes, A set of edge nodes; edge set Used to encode the topological relationships between B-Rep elements, i.e., the boundary represents the connection between faces and edges in the model.

3. The part assembly method based on geometric topology fusion according to claim 1, characterized in that, The process of step 2 is as follows: For each face node in the heterogeneous geometric topology graph, uniform mesh sampling is performed on the two-dimensional UV parameter domain, where U and V refer to two parameters in the parametric surface; then, two feature meshes containing coordinates, normal vectors, and trimmed masks are generated (UV). , Let n represent the real number field, and n represent the number of sampling points; Extracting geometric features using a 2D convolutional network in UV_Net Simultaneously, the basic properties of the surface, including surface type and area, are encoded into attribute features using a multilayer perceptron (MLP). ; i represents the i-th node; Geometric features and attribute features are combined to form a unified feature for surface nodes. , represented as: （1） Then, element-wise gating of the geometric features is calculated using linear projection and the sigmoid function. and element-wise gating of attribute features , represented as: , （2） in, This represents the weight matrix used to generate the geometric feature gating weights for the face nodes. This represents the corresponding bias vector; This represents the weight matrix used to generate the gating weights for the attribute features of the face nodes. This represents the corresponding bias vector; This represents the sigmoid activation function; Then, by gating each modality and splicing them together, the fused face node embedding is obtained. , represented as: （3） in This represents a vector concatenation operation; This represents element-wise multiplication. For each edge node in the heterogeneous geometric topology graph, uniform sampling is performed in the one-dimensional parameter domain to generate features containing coordinates and tangent vectors. Geometric features are extracted using a one-dimensional convolutional network in UV_Net. Simultaneously, the basic attributes of the edges, including curve type and length, are encoded into attribute features using a multilayer perceptron (MLP). Geometric features and attribute features are combined to form a unified feature for edge nodes. , represented as: （4） Then, element-wise gating of the geometric features is calculated using linear projection and the sigmoid function. and element-wise gating of attribute features , represented as: , （5） in, This represents the weight matrix used to generate the geometric feature gating weights of the edge nodes. This represents the corresponding bias vector; This represents the weight matrix used to generate the gating weights for edge node attribute features. This represents the corresponding bias vector; Then, by gating each modality and concatenating them, the fused edge node embedding is obtained. , represented as: （6） By sequentially concatenating the feature embeddings of all face nodes and edge nodes, we obtain: （7） in The number of face nodes; This represents the number of edge nodes; Indicates a splicing operation; Finally, the node feature matrix of the heterogeneous geometric topology diagram of the parts is obtained. ,in This represents the total number of face nodes and boundary points, with each node having a feature dimension of 448.

4. The part assembly method based on geometric topology fusion according to claim 3, characterized in that, In step 3, the graph attention reasoning module is composed of The residual graph attention network is stacked in layers, which is composed of GATv2. For the first layer... The layer is calculated as follows: （8） Formula (8) indicates that the output of the previous layer is used as the input of this layer, where l = 0, 1, 2, 3, 4, and , This is the node feature matrix obtained in step 2; Let be the edge set of the graph; ELU is the activation function; residual connections ensure training stability and preservation of low-level features; For two parts to be assembled and It uses the same graph attention network stack with shared parameters for processing, and outputs node embedding representations with high-level semantic enhancement. and This refers to the geometric-topological dependencies within CAD parts, ensuring the consistency of the feature semantic space.

5. The part assembly method based on geometric topology fusion according to claim 4, characterized in that, In step 4, for the parts Nodes in and parts Nodes in Embedding nodes to represent and The data is concatenated, and the result is input into a multilayer perceptron (MLP) to obtain the association score of the node pairs. （9） in Indicates parts Nodes in and parts Nodes in The correlation score between them; Construct an association score matrix from the association scores of all node pairs. , represented as: （10） in and Parts and parts The higher the number of nodes, the better. The greater the likelihood that there is a valid assembly constraint between node i and node j.

6. The part assembly method based on geometric topology fusion according to claim 5, characterized in that, In step 5, the constructed loss function includes maximum likelihood estimation loss and symmetry loss, where: Maximum likelihood estimation loss Represented as: （11） （12） in Let k be the total number of candidate node pairs; For the first Association score of each candidate node pair; For the first Normalized true probability of each candidate node pair; Symmetry loss Represented as: （13） in It is a correlation score matrix The transpose of the constraint connection matrix is ​​used to reflect the two-way physical characteristics of the assembly relationship. Total training loss Represented as: （14） Where B represents the batch during training, and the number of training samples.

7. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 6.

8. A computer storage medium, characterized in that, The computer storage medium stores a plurality of instructions adapted for loading by a processor and executing the method of any one of claims 1 to 6.

Images