A three-dimensional part PMI automatic labeling method based on a graph neural network

Abstract

The application provides a three-dimensional part PMI automatic labeling method based on a graph neural network, belongs to the technical field of aero-engines, and comprises the following steps: initialization configuration based on a UG NX environment; B-Rep topological analysis is performed on an imported three-dimensional part model, and the three-dimensional part model is converted into attribute graph structure data; feature learning and semantic classification are performed on the attribute graph structure data based on a graph neural network, high-dimensional feature fingerprints representing overall engineering semantics of the part and labeling graphs are extracted, and prototype clustering analysis is performed to form a knowledge base; structured recognition is performed on a newly input three-dimensional part model, similar prototype matching and rule calling are performed based on the knowledge base, a parameterized labeling action sequence is generated, the parameterized labeling action sequence is mapped into PMI creation commands, and the labeling of the new model and the establishment of related annotation objects are automatically completed. The scheme realizes the automation of the labeling process and the leap-forward improvement of the efficiency, and improves the labeling accuracy.

Description

Technical Field

[0001] This application relates to the field of aero-engine technology, and in particular to an automatic PMI annotation method for three-dimensional parts based on graph neural networks. Background Technology

[0002] As aero-engines continue to evolve towards higher thrust-to-weight ratios, higher reliability, and longer lifespans, their internal structures are becoming increasingly complex. The aero-engine piping system, as a crucial network transporting fuel, lubricating oil, air, and hydraulic fluids, is known as the engine's "blood vessels," and its design and manufacturing quality directly impacts the overall performance and safety of the engine. Traditional two-dimensional drawings are insufficient to fully define the manufacturing and testing requirements of these piping components with complex spatial orientations. Therefore, Product Manufacturing Information (PMI) annotation technology based on three-dimensional models has become a core component of modern aero-engine digital design and manufacturing systems. PMI directly attaches manufacturing information such as dimensions, tolerances, surface treatments, and material specifications to the three-dimensional model, achieving lossless transfer of design intent and providing a unique and authoritative data source for subsequent process planning, CNC programming, and quality control.

[0003] Currently, in UG NX software, PMI annotation for aero-engine piping models primarily relies on manual operation by designers. This process is quite cumbersome. First, designers need to manually identify the geometric features requiring annotation within the complex 3D assembly and then manually select different PMI annotation commands, such as geometric annotation, geometric tolerance, and surface roughness. After completing the annotation, the position and layer of the annotation need to be manually adjusted. This human-centric process exposes many inherent flaws in the application of complex products like aero-engine piping, becoming a bottleneck for improving design efficiency and data quality.

[0004] First, it is inefficient and time-consuming. Aero-engine piping systems have a large number of parts and complex structures. A complete piping component may contain hundreds of features that need to be labeled. Manually identifying, creating, and placing labels one by one is a repetitive, tedious, and extremely time-consuming process. According to statistics, it takes more than 20 designers half a month to label complex piping components of a certain type of aero-engine, which greatly slows down the product development cycle.

[0005] Secondly, the quality and consistency of annotations are highly dependent on individual experience, making them prone to errors. The accuracy, completeness, and standardization of PMI annotations depend heavily on the operator's skill level and experience. Different engineers may have differing understandings of the same standard, leading to inconsistencies in the annotation methods for similar features. Manual operation is also prone to omissions and errors, which will pass into downstream processes and require subsequent corrections, making the process extremely tedious.

[0006] Furthermore, manual annotation struggles to handle complex geometric and topological relationships. With the continuous upgrading of aero-engines, piping systems are becoming increasingly complex, resulting in a growing number of non-standard geometries. Identifying their key features presents a significant challenge for manual work. Manual identification is prone to overlooking occluded or hidden features and struggles to accurately extract the geometric parameters used for annotation.

[0007] To address the aforementioned issues, the field of automated 3D PMI annotation for aero-engine piping urgently requires a technology capable of deeply understanding the design semantics of 3D models, automatically identifying complex geometric features, and intelligently generating standardized PMI annotations. This technology aims to overcome the bottleneck of manual operation and improve design efficiency and data quality. This invention is proposed against this backdrop. Summary of the Invention

[0008] In view of this, the present application provides an automatic PMI annotation method for three-dimensional parts based on graph neural networks, which at least partially solves the problems of heavy reliance on manual labor, low efficiency and poor consistency in the three-dimensional PMI annotation of aero-engine pipelines in the prior art.

[0009] This application provides an automatic PMI annotation method for 3D parts based on graph neural networks, including:

[0010] Based on the UG NX environment, after receiving instructions from the user to import a 3D part model, select a processing mode, or initiate automatic annotation, the system performs initial configuration. The imported 3D part model is subjected to B-Rep topology analysis to convert the 3D part model into attribute graph structure data; Based on graph neural networks, feature learning and semantic classification are performed on attribute graph structure data to extract high-dimensional feature fingerprints that represent the overall engineering semantics and annotation intentions of parts, and prototype clustering analysis is performed to form a knowledge base. The new input 3D part model is subjected to structured recognition. Based on the structured recognition results and knowledge base, similar prototype matching and rule calling are performed to generate a parameterized annotation action sequence. The generated parametric annotation action sequence is mapped to PMI creation commands in the UG NX environment, automatically completing the annotation of the newly input 3D part model and the creation of related annotation objects.

[0011] According to a specific implementation of an embodiment of this application, the initialization configuration includes: Model loading, interface status detection, and preparation of basic parameters required for subsequent processes.

[0012] According to a specific implementation of an embodiment of this application, the step of performing B-Rep topology analysis on the imported 3D part model to convert the 3D part model into attribute graph structure data includes: Extract geometric data from the imported 3D part model to obtain the hierarchical relationships, adjacency relationships, and geometric constraints of geometric elements, including face elements, edge elements, and vertex elements; For face elements, face features are calculated, including face type, area, curvature, normal vector, center position, number of boundaries, and number of adjacent faces; for edge elements, edge features are calculated, including edge type, length, curvature, endpoint position, and relationship with adjacent faces; for vertex elements, vertex features are calculated, including spatial coordinates, number of connecting edges, and number of faces to which it belongs. The face features, edge features, and vertex features are numerically encoded to form the initial feature matrix; A topological relationship graph with attribute information is constructed using parts as the basic unit. The nodes of the topological relationship graph represent geometric elements, and the edges of the topological relationship graph represent the adjacency relationships between geometric elements. Both nodes and edges are attached with an initial feature matrix, and the topological relationship graph is used as the attribute graph structure data.

[0013] According to a specific implementation of an embodiment of this application, the step of performing feature learning and semantic classification on attribute graph structure data based on graph neural networks to extract high-dimensional feature fingerprints representing the overall engineering semantics and annotation intent of the parts includes: For the aero-engine piping component samples annotated with historical standard PMI, the corresponding attribute graph structure data is obtained. The attribute graph structure data includes surface-level semantic attribute graphs. Based on the surface-level semantic attribute graph, a graph neural network is used to perform message passing and feature updating on the surface nodes, and to extract high-dimensional feature fingerprints that represent the overall engineering semantics and annotation intent of the parts.

[0014] According to a specific implementation of an embodiment of this application, the expression of the face-level semantic attribute graph is: , in, For a face-level semantic attribute graph, It is the set of nodes consisting of all B-Rep surfaces of the part; This is the set of engineering-related edges between faces. A connection is established when two faces satisfy the conditions of shared edges, coaxiality, approximate parallelism, or engineering intersection. The node's geometric topological feature matrix; This is the feature matrix of edge relationships; This is the associated label matrix obtained by analyzing samples of aero-engine piping components labeled with historical standard PMI; The expression for the associated label vector of a single face node in the associated label matrix is: , Where the subscript i represents the number of the current face node, For a single face node, there is a companion label vector. For accompanying label mapping functions; as well as These represent the coding results for dimensions, tolerances, datums, process semantics, and annotation action sequences, respectively.

[0015] According to a specific implementation of an embodiment of this application, the expression for updating the face node is: , Wherein, the subscript j represents the number of the adjacent face node that is connected to the current face node i. This represents the feature representation of adjacent face node j in the l-th layer. This represents the feature representation of the current face node i in the l-th layer. This represents the feature representation of the current face node i in the (l+1)th layer. Features of engineering relationships between surfaces It is a non-linear activation function. For its own feature weight matrix, This is the relation feature weight matrix. Let N(i) be the relation weight of adjacent face node j to the current face node i, and N(i) be the set of face nodes.

[0016] According to a specific implementation of an embodiment of this application, the step of performing prototype clustering analysis to form a knowledge base includes: Within the same macro category, for every two mechanical structures of a part, a comprehensive similarity matrix is ​​constructed based on high-dimensional feature fingerprints, dimensions, benchmarks, and tolerance distributions; Feature classification is performed based on the comprehensive similarity matrix to form multiple labeled intent prototype clusters, and the prototype center of each labeled intent prototype cluster is calculated. The annotation action template, baseline priority template, and tolerance range template corresponding to each annotation intent prototype cluster are accumulated to form a knowledge base.

[0017] According to a specific implementation of an embodiment of this application, the expression for the comprehensive similarity matrix is: , Among them, S ab For the comprehensive similarity matrix, λ1, λ2, and λ3 are the first, second, and third coefficients, respectively, and z a For the high-dimensional feature fingerprint of the a-th mechanical structure, z b For the high-dimensional feature fingerprint of the b-th mechanical structure, σ zσ is the scale parameter for the distance of high-dimensional feature fingerprints. p P is the scale parameter for the similarity of dimensional or tolerance parameters. a Let P be the vector of key dimensional parameters for the a-th mechanical structure. b Let d be the vector of key dimensional parameters for the b-th mechanical structure. a Let d be the candidate reference priority vector for the a-th mechanical structure. b Let b be the candidate reference priority vector for the b-th mechanical structure. It is the Euclidean norm; The expression for the prototype center is: , Among them, c u C is the prototype center of the u-th labeled prototype cluster. u Let be the set of mechanical structure samples contained in the u-th labeled prototype cluster.

[0018] According to a specific implementation of an embodiment of this application, the step of performing structured recognition on the newly input 3D part model, and based on the structured recognition results and a knowledge base, performing similar prototype matching and rule invocation to generate a parameterized annotation action sequence includes: The new input 3D part model is subjected to structured recognition, and a new surface-level semantic attribute map corresponding to the new input 3D part model is extracted. Based on the new surface-level semantic attribute map, a new high-dimensional feature fingerprint is obtained, and key dimension parameter vectors and candidate benchmark priority vectors are extracted. Based on the new high-dimensional feature fingerprint, key size parameter vector, candidate benchmark priority vector and prototype center, calculate the comprehensive matching score between the new high-dimensional feature fingerprint and each labeled intent prototype cluster; The optimal prototype cluster index is obtained based on the comprehensive matching score; Generate the corresponding parameterized annotation action sequence based on the optimal prototype cluster index.

[0019] According to a specific implementation of an embodiment of this application, the expression for the comprehensive matching score is: , in, To calculate the overall matching score, For new high-dimensional feature fingerprints, For the key dimension parameter vector, Let P be the candidate benchmark priority vector. u Let be the center vector of the key dimensions and tolerance parameters corresponding to the u-th annotation intent prototype cluster. In the weight matrix Distance norm under constraints Let be the baseline priority template vector corresponding to the u-th labeled intent prototype cluster. Let ω1, ω2, ω3, and ω4 be the rule confidence or compatibility correction term for the u-th prototype cluster, and let ω1, ω2, ω3, and ω4 be the fourth, fifth, sixth, and seventh coefficients, respectively. The expression for the parameterized annotation action sequence is: , in, For parametric annotation of action sequences, A function for generating parameterized labeled action sequences. For the optimal prototype cluster index u The corresponding knowledge base storage unit, For the T-th annotation action, each annotation action includes annotation type, associated surface identifier, annotation view, annotation value, tolerance content, datum content, and layer attributes.

[0020] Beneficial effects: The graph neural network-based automatic PMI annotation method for 3D parts in this application embodiment achieves automation and a significant improvement in efficiency during the annotation process. By fully automating the geometric feature recognition and PMI annotation generation process, it replaces the traditional manual operation mode. Users can complete batch and rapid annotation of complex pipeline models simply by clicking a button, reducing the annotation work that originally took hours or even days to minutes, greatly liberating designers. The intelligent recognition method based on graph neural networks (GNNs) can accurately and stably identify complex spatial features, avoiding omissions and misjudgments that may occur with manual recognition. At the same time, the system's built-in standardized annotation knowledge base ensures that features of the same type are always labeled according to a unified standard, completely eliminating problems such as mislabeling and omissions caused by personal reasons, and significantly improving the authority and reliability of 3D digital models as the basis for manufacturing. By using graph neural networks to perform deep learning on the topology and geometric relationships of a model, it is possible to effectively understand and process complex features that are difficult for humans to quickly and accurately identify, and automatically extract key annotation parameters, thus overcoming the bottleneck of the difficulty for humans to identify complex geometric structures. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of an automatic PMI annotation method for three-dimensional parts based on a graph neural network according to an embodiment of the present invention. Detailed Implementation

[0023] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0024] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this application, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0026] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The illustrations only show the components related to this application and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0027] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0028] In one embodiment, a method for automatic PMI annotation of 3D parts based on graph neural networks is provided, referring to... Figure 1 Specifically, it includes the following steps: Step 1: Based on the UG NX environment, after receiving the user's instruction to import a 3D part model, select a processing mode, or initiate automatic annotation, perform initial configuration; Step 2: Perform B-Rep topology analysis on the imported 3D part model to convert the 3D part model into attribute graph structure data; B-Rep is a boundary representation method, which is a 3D solid representation method in CAD software. It uses boundary elements such as points, edges, and faces and their connection relationships to describe the 3D shape of the part. Step 3: Based on graph neural networks, perform feature learning and semantic classification on the attribute graph structure data, extract high-dimensional feature fingerprints that represent the overall engineering semantics and annotation intent of the parts, and perform prototype clustering analysis to form a knowledge base; Step 4: Perform structured recognition on the newly input 3D part model. Based on the structured recognition results and the knowledge base, perform similar prototype matching and rule calling to generate a parametric annotation action sequence. Step 5: Map the generated parametric annotation action sequence to the PMI creation command in the UG NX environment to automatically complete the annotation of the newly input 3D part model and the creation of related annotation objects.

[0029] In practice, the plugin is integrated into the UG NX software environment as a menu button or function area icon. Users can start the entire automatic annotation process by clicking the button; in response to user commands, it receives the selected aero-engine piping assembly in the UG NX workpiece as the processing target and loads its complete geometric and topological data.

[0030] Furthermore, the initialization configuration includes: Model loading, interface status detection, and preparation of basic parameters required for subsequent processes.

[0031] Specifically, after receiving a user's command to import parts, select a processing mode, or initiate automatic annotation, the system completes the initialization configuration of the current session, including model loading, interface status detection, and preparation of basic parameters required for subsequent processes. Through the initialization module, a connection can be established between the user's operating terminal and the backend intelligent recognition and annotation execution process.

[0032] Furthermore, the step of performing B-Rep topology analysis on the imported 3D part model to convert the 3D part model into attribute graph structure data includes: Extract geometric data from the imported 3D part model to obtain the hierarchical relationships, adjacency relationships, and geometric constraints of geometric elements, including face elements, edge elements, and vertex elements; For face elements, face features are calculated, including face type, area, curvature, normal vector, center position, number of boundaries, and number of adjacent faces; for edge elements, edge features are calculated, including edge type, length, curvature, endpoint position, and relationship with adjacent faces; for vertex elements, vertex features are calculated, including spatial coordinates, number of connecting edges, and number of faces to which it belongs. The face features, edge features, and vertex features are numerically encoded to form the initial feature matrix; A topological relationship graph with attribute information is constructed using parts as the basic unit. The nodes of the topological relationship graph represent geometric elements, and the edges of the topological relationship graph represent the adjacency relationships between geometric elements. Both nodes and edges are attached with an initial feature matrix, and the topological relationship graph is used as the attribute graph structure data.

[0033] In this embodiment, the conversion of the 3D part model is implemented in the 3D model data extraction and graph structure construction module. The extracted features provide a structured data foundation for the subsequent input of the graph neural network. Through this module, traditional CAD geometric models can be converted into a graph representation suitable for machine learning processing.

[0034] Furthermore, the step of performing feature learning and semantic classification on attribute graph structure data based on graph neural networks to extract high-dimensional feature fingerprints representing the overall engineering semantics and annotation intent of the parts includes: For the aero-engine piping component samples annotated with historical standard PMI, the corresponding attribute graph structure data is obtained. The attribute graph structure data includes surface-level semantic attribute graphs. Based on the surface-level semantic attribute graph, a graph neural network is used to perform message passing and feature updating on the surface nodes, and to extract high-dimensional feature fingerprints that represent the overall engineering semantics and annotation intent of the parts.

[0035] Specifically, the expression for the face-level semantic attribute graph is: , in, For a face-level semantic attribute graph, It is the set of nodes consisting of all B-Rep surfaces of the part; This is the set of engineering-related edges between faces. A connection is established when two faces satisfy the conditions of shared edges, coaxiality, approximate parallelism, or engineering intersection. The node's geometric topological feature matrix; This is the feature matrix of edge relationships; This is the associated label matrix obtained by analyzing samples of aero-engine piping components labeled with historical standard PMI; The expression for the associated label vector of a single face node in the associated label matrix is: , Where the subscript i represents the number of the current face node, For a single face node, there is a companion label vector. This is a companion label mapping function used to convert PMI annotation information associated with geometric surfaces in historical samples into digital label codes. PMI annotation information includes dimensions, tolerances, datum, roughness, process notes, and annotation action types. as well as These represent the coding results for dimensions, tolerances, datums, process semantics, and annotation action sequences, respectively.

[0036] Furthermore, the expression for updating the face nodes is: , Wherein, the subscript j represents the number of the adjacent face node that is connected to the current face node i. This represents the feature representation of adjacent face node j in the l-th layer. This represents the feature representation of the current face node i in the l-th layer. This represents the feature representation of the current face node i in the (l+1)th layer. Features of engineering relationships between surfaces It is a non-linear activation function. For its own feature weight matrix, This is the relation feature weight matrix. Let N(i) be the relation weight of adjacent face node j to the current face node i, and N(i) be the set of face nodes.

[0037] Furthermore, high-dimensional feature fingerprints of the parts are obtained through multi-layer transfer: , z (m) For high-dimensional feature fingerprints, N m The total number of face nodes is given by Pool(·), which is a pooling operation used to pool the m-th part. The feature representations of each face node are aggregated to generate a high-dimensional feature fingerprint at the part level.

[0038] Furthermore, the step of performing prototype clustering analysis to form a knowledge base includes: Within the same macro category, for every two mechanical structures of a part, a comprehensive similarity matrix is ​​constructed based on high-dimensional feature fingerprints, dimensions, benchmarks, and tolerance distributions; Feature classification is performed based on the comprehensive similarity matrix to form multiple labeled intent prototype clusters, and the prototype center of each labeled intent prototype cluster is calculated. The annotation action template, baseline priority template, and tolerance range template corresponding to each annotation intent prototype cluster are accumulated to form a knowledge base.

[0039] According to a specific implementation of an embodiment of this application, the expression for the comprehensive similarity matrix is: , Among them, S ab For the comprehensive similarity matrix, λ1, λ2, and λ3 are the first, second, and third coefficients, respectively, and z a For the high-dimensional feature fingerprint of the a-th mechanical structure, z b For the high-dimensional feature fingerprint of the b-th mechanical structure, σ z σ is the scale parameter for the distance of high-dimensional feature fingerprints. p P is the scale parameter for the similarity of dimensional or tolerance parameters. a Let P be the vector of key dimensional parameters for the a-th mechanical structure. b Let d be the vector of key dimensional parameters for the b-th mechanical structure. a Let d be the candidate reference priority vector for the a-th mechanical structure. b Let b be the candidate reference priority vector for the b-th mechanical structure. It is the Euclidean norm; The expression for the prototype center is: , Among them, c u C is the prototype center of the u-th labeled prototype cluster. u Let be the set of mechanical structure samples contained in the u-th labeled prototype cluster; The expression for the storage unit of the knowledge base is: , K u Let a be the storage unit of the knowledge base corresponding to the u-th labeled intent prototype cluster. u This serves as the annotation action template for the u-th prototype cluster. This serves as the baseline priority template for the u-th prototype cluster. This is the tolerance range template for the u-th prototype cluster. Let be the set of rule constraints for the u-th prototype cluster.

[0040] In one embodiment, the step of performing structured recognition on the newly input 3D part model, and based on the structured recognition results and a knowledge base, performing similar prototype matching and rule invocation to generate a parameterized annotation action sequence includes: Perform structured recognition on the new input 3D part model and extract the new surface-level semantic attribute map g corresponding to the new input 3D part model. ( ) Based on the new surface-level semantic attribute map, a new high-dimensional feature fingerprint is obtained by using the trained graph neural network (the graph neural network trained by the method in step 3), and key size parameter vectors and candidate benchmark priority vectors are extracted. Based on the new high-dimensional feature fingerprint, key size parameter vector, candidate benchmark priority vector and prototype center, calculate the comprehensive matching score between the new high-dimensional feature fingerprint and each labeled intent prototype cluster; The optimal prototype cluster index is obtained based on the comprehensive matching score; Generate the corresponding parameterized annotation action sequence based on the optimal prototype cluster index.

[0041] According to a specific implementation of an embodiment of this application, the expression for the comprehensive matching score is: , in, To calculate the overall matching score, For new high-dimensional feature fingerprints, For the key dimension parameter vector, Let P be the candidate benchmark priority vector. u Let be the center vector of the key dimensions and tolerance parameters corresponding to the u-th annotation intent prototype cluster. In the weight matrix Distance norm under constraints Let be the baseline priority template vector corresponding to the u-th labeled intent prototype cluster. Let ω1, ω2, ω3, and ω4 be the rule confidence or compatibility correction term for the u-th prototype cluster, and let ω1, ω2, ω3, and ω4 be the fourth, fifth, sixth, and seventh coefficients, respectively. The expression for the optimal prototype cluster index is: , Among them, u For the optimal prototype cluster index, The value of the independent variable u when it reaches its maximum value; The expression for the parameterized annotation action sequence is: , in, For parametric annotation of action sequences, A function for generating parameterized labeled action sequences. For the optimal prototype cluster index u The corresponding knowledge base storage unit, For the T-th annotation action, each annotation action includes annotation type, associated surface identifier, annotation view, annotation value, tolerance content, datum content, and layer attributes.

[0042] Finally, the parametric annotation action sequence was mapped into executable PMI creation commands through the UG NX secondary development interface, automatically completing the creation of dimension annotations, tolerance annotations, datum annotations, and related annotation objects. The generated PMI objects remain associated with the original 3D geometric entities, enabling the annotation information to be updated synchronously when the model is modified subsequently, realizing the automatic reconstruction and parametric output of the engineering semantics of 3D parts.

[0043] The embodiments provided by this invention learn the geometric topological features of 3D parts through graph neural networks, which can fully explore the engineers' annotation intentions contained in historical annotation data. This solves the problems of traditional rules relying on manual sorting of logic and poor generalization ability. It can also stably output PMI annotation results that conform to engineering specifications for 3D parts with varied structures. At the same time, the annotation efficiency is several times higher than manual annotation, which greatly shortens the process preparation cycle of 3D part models and can adapt to the rapid annotation needs of complex parts such as aero-engine pipes.

[0044] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for automatic PMI annotation of 3D parts based on graph neural networks, characterized in that, include: Based on the UG NX environment, after receiving instructions from the user to import a 3D part model, select a processing mode, or initiate automatic annotation, the system performs initial configuration. The imported 3D part model is subjected to B-Rep topology analysis to convert the 3D part model into attribute graph structure data; Based on graph neural networks, feature learning and semantic classification are performed on attribute graph structure data to extract high-dimensional feature fingerprints that represent the overall engineering semantics and annotation intentions of parts, and prototype clustering analysis is performed to form a knowledge base. The process of forming a knowledge base through prototype clustering analysis includes: within the same macro category, for every two mechanical structures of a part, constructing a comprehensive similarity matrix based on high-dimensional feature fingerprints, dimensions, benchmarks, and tolerance distributions; performing feature classification based on the comprehensive similarity matrix to form multiple annotation intent prototype clusters, calculating the prototype center of each annotation intent prototype cluster; and accumulating the annotation action template, benchmark priority template, and tolerance range template corresponding to each annotation intent prototype cluster to form a knowledge base. The new input 3D part model is subjected to structured recognition. Based on the structured recognition results and knowledge base, similar prototype matching and rule calling are performed to generate a parameterized annotation action sequence. The generated parametric annotation action sequence is mapped to PMI creation commands in the UG NX environment, automatically completing the annotation of the newly input 3D part model and the creation of related annotation objects.

2. The automatic PMI annotation method for 3D parts based on graph neural networks according to claim 1, characterized in that, The initialization configuration includes: Model loading, interface status detection, and preparation of basic parameters required for subsequent processes.

3. The automatic PMI annotation method for 3D parts based on graph neural networks according to claim 1, characterized in that, The process of performing B-Rep topology analysis on the imported 3D part model to convert it into attribute graph structure data includes: Extract geometric data from the imported 3D part model to obtain the hierarchical relationships, adjacency relationships, and geometric constraints of geometric elements, including face elements, edge elements, and vertex elements; For face elements, face features are calculated, including face type, area, curvature, normal vector, center position, number of boundaries, and number of adjacent faces; for edge elements, edge features are calculated, including edge type, length, curvature, endpoint position, and relationship with adjacent faces; for vertex elements, vertex features are calculated, including spatial coordinates, number of connecting edges, and number of faces to which it belongs. The face features, edge features, and vertex features are numerically encoded to form the initial feature matrix; A topological relationship graph with attribute information is constructed using parts as the basic unit. The nodes of the topological relationship graph represent geometric elements, and the edges of the topological relationship graph represent the adjacency relationships between geometric elements. Both nodes and edges are attached with an initial feature matrix, and the topological relationship graph is used as the attribute graph structure data.

4. The automatic PMI annotation method for 3D parts based on graph neural networks according to claim 1, characterized in that, The method of performing feature learning and semantic classification on attribute graph structure data based on graph neural networks to extract high-dimensional feature fingerprints representing the overall engineering semantics and annotation intent of parts includes: For the aero-engine piping component samples annotated with historical standard PMI, the corresponding attribute graph structure data is obtained. The attribute graph structure data includes surface-level semantic attribute graphs. Based on the surface-level semantic attribute graph, a graph neural network is used to perform message passing and feature updating on the surface nodes, and to extract high-dimensional feature fingerprints that represent the overall engineering semantics and annotation intent of the parts.

5. The automatic PMI annotation method for 3D parts based on graph neural networks according to claim 4, characterized in that, The expression for the facet-level semantic attribute graph is: ， in, For a face-level semantic attribute graph, It is the set of nodes consisting of all B-Rep surfaces of the part; This is the set of engineering-related edges between faces. A connection is established when two faces satisfy the conditions of shared edges, coaxiality, approximate parallelism, or engineering intersection. The node's geometric topological feature matrix; This is the feature matrix of edge relationships; This is the associated label matrix obtained by analyzing samples of aero-engine piping components labeled with historical standard PMI; The expression for the associated label vector of a single face node in the associated label matrix is: ， Where the subscript i represents the number of the current face node, For a single face node, there is a companion label vector. For accompanying label mapping functions; as well as These represent the coding results for dimensions, tolerances, datums, process semantics, and annotation action sequences, respectively.

6. The automatic PMI annotation method for 3D parts based on graph neural networks according to claim 5, characterized in that, The expression for updating the face nodes is: ， Wherein, the subscript j represents the number of the adjacent face node that is connected to the current face node i. This represents the feature representation of adjacent face node j in the l-th layer. This represents the feature representation of the current face node i in the l-th layer. This represents the feature representation of the current face node i in the (l+1)th layer. Features of engineering relationships between surfaces It is a non-linear activation function. For its own feature weight matrix, This is the relation feature weight matrix. Let N(i) be the relation weight of adjacent face node j to the current face node i, and N(i) be the set of face nodes.

7. The automatic PMI annotation method for 3D parts based on graph neural networks according to claim 6, characterized in that, The expression for the comprehensive similarity matrix is: ， Among them, S ab For the comprehensive similarity matrix, λ1, λ2, and λ3 are the first, second, and third coefficients, respectively, and z a For the high-dimensional feature fingerprint of the a-th mechanical structure, z b For the high-dimensional feature fingerprint of the b-th mechanical structure, σ z σ is the scale parameter for the distance of high-dimensional feature fingerprints. p P is the scale parameter for the similarity of dimensional or tolerance parameters. a Let P be the vector of key dimensional parameters for the a-th mechanical structure. b Let d be the vector of key dimensional parameters for the b-th mechanical structure. a Let d be the candidate reference priority vector for the a-th mechanical structure. b Let b be the candidate reference priority vector for the b-th mechanical structure. It is the Euclidean norm; The expression for the prototype center is: ， Among them, c u C is the prototype center of the u-th labeled prototype cluster. u Let be the set of mechanical structure samples contained in the u-th labeled prototype cluster.

8. The automatic PMI annotation method for 3D parts based on graph neural networks according to claim 7, characterized in that, The process involves performing structured recognition on the newly input 3D part model, and based on the structured recognition results and a knowledge base, performing similar prototype matching and rule invocation to generate a parameterized annotation action sequence, including: The new input 3D part model is subjected to structured recognition, and a new surface-level semantic attribute map corresponding to the new input 3D part model is extracted. Based on the new surface-level semantic attribute map, a new high-dimensional feature fingerprint is obtained, and key dimension parameter vectors and candidate benchmark priority vectors are extracted. Based on the new high-dimensional feature fingerprint, key size parameter vector, candidate benchmark priority vector and prototype center, calculate the comprehensive matching score between the new high-dimensional feature fingerprint and each labeled intent prototype cluster; The optimal prototype cluster index is obtained based on the comprehensive matching score; Generate the corresponding parameterized annotation action sequence based on the optimal prototype cluster index.

9. The automatic PMI annotation method for three-dimensional parts based on graph neural networks according to claim 8, characterized in that, The expression for the comprehensive matching score is: ， in, To calculate the overall matching score, For new high-dimensional feature fingerprints, For the key dimension parameter vector, Let P be the candidate benchmark priority vector. u Let be the center vector of the key dimensions and tolerance parameters corresponding to the u-th annotation intent prototype cluster. In the weight matrix Distance norm under constraints Let be the baseline priority template vector corresponding to the u-th labeled intent prototype cluster. Let ω1, ω2, ω3, and ω4 be the rule confidence or compatibility correction term for the u-th prototype cluster, and let ω1, ω2, ω3, and ω4 be the fourth, fifth, sixth, and seventh coefficients, respectively. The expression for the parameterized annotation action sequence is: ， in, For parametric annotation of action sequences, A function for generating parameterized labeled action sequences. For the optimal prototype cluster index u The corresponding knowledge base storage unit, For the T-th annotation action, each annotation action includes annotation type, associated surface identifier, annotation view, annotation value, tolerance content, datum content, and layer attributes.

Images