Method for identifying boundary features of three-dimensional model of closed cavity of airplane

Abstract

The invention discloses a method for identifying boundary features of a three-dimensional model of a closed cavity of an airplane, and provides a method for automatically and intelligently detecting the boundary features of the closed cavity in order to solve the problems that the extraction precision of the boundary features of the closed cavity is poor and the efficiency is low. Providing a minimum allowable gap recommendation mechanism based on a decision tree, and automatically extracting boundary part information; analyzing and screening an accurate boundary surface through a geometric distance and a normal vector included angle; and constructing a part attribute adjacency graph, and identifying the gap type of the closed cavity based on a GNN enhanced VF3 sub-graph isomorphism algorithm to realize automatic identification and marking of boundary parts and gaps of the closed cavity in the three-dimensional model. The method solves the problems of low manual detection efficiency and large error in a complex assembly environment, remarkably improves the precision and automation level of closed cavity boundary feature recognition, and is suitable for assembly area modeling scenes with few samples and high precision requirements in the field of aeronautical manufacturing.

Description

Technical Field

[0001] The present invention belongs to computer-aided design and manufacturing technology, and in particular relates to a method for identifying boundary features of a three-dimensional model of an aircraft closed cavity. Background Art

[0002] With the continuous development of aviation manufacturing technology, aircraft structural design has become increasingly complex and sophisticated. In particular, under the multiple requirements of lightweight, high safety and high reliability, the degree of structural integration has been significantly improved. In this context, the geometric modeling and precise identification of closed cavities, as common and important spatial units within aviation structures, are of key significance to the performance and structural safety of the entire aircraft. Closed cavities are usually nested between structures such as wings and fuselages, and are enclosed by a large number of components such as skins, ribs, beams, and frames of various shapes. The overall structure is complex, the boundaries are irregular, the definitions are vague, and there are often gaps and clearances between parts. These factors make the automatic identification and digital modeling of closed cavities face severe challenges.

[0003] In actual engineering design, accurately identifying the position, shape, boundary structure, and adjacent relationships of enclosed cavities is of great engineering significance for improving structural space utilization, enhancing local strength, and preventing structural leakage and stress concentration. However, traditional identification of enclosed cavities mainly relies on manual methods, that is, experienced designers perform visual analysis and logical reasoning based on CAD three-dimensional models and design drawings. This type of method is not only inefficient and poorly standardized, but also easily affected by the designer's level of experience. Problems such as incomplete identification and misjudgment often occur, which in turn leads to frequent design modifications, extended manufacturing cycles, and even hidden quality risks.

[0004] To sum up, how to improve recognition efficiency and intelligence while ensuring recognition accuracy is the core issue facing the current research on closed cavity feature modeling and recognition. Summary of the Invention

[0005] Purpose of the Invention: This invention provides a method for identifying boundary features within a three-dimensional aircraft enclosed cavity model, aiming to address the low efficiency and large errors associated with traditional manual inspection. By combining interference detection, geometric topology analysis, and graph and rule-based feature matching, this method enables automated identification of enclosed cavity boundary parts, boundary surfaces, and gaps, significantly improving inspection accuracy and efficiency. This method is suitable for complex assembly scenarios in the aviation manufacturing industry, where small sample sizes and high precision are required.

[0006] Technical solution:

[0007] The present invention proposes a method for identifying boundary features of a three-dimensional model of an aircraft closed cavity, comprising:

[0008] Analyze the 3D aircraft model, including the enclosed cavity, extract the enclosed cavity model and all part models within the assembly area, and identify the boundary features of the enclosed cavity. These boundary features include: the boundary parts that enclose the cavity space, the precise boundary surface formed by the boundary parts, and any gaps in the cavity space.

[0009] An interference detection method is used to calculate the distance between each part model and the closed cavity model. A threshold recommendation mechanism based on a decision tree model is proposed to autonomously set the threshold. Parts with a distance less than the set threshold are marked as boundary parts.

[0010] Extracting a closed cavity patch from the closed cavity model and a boundary part surface patch from the boundary part model, calculating the surface normal vector angle and distance between the closed cavity patch and the boundary part surface patch to screen adjacent patches, and splicing the adjacent patches to obtain an accurate boundary surface;

[0011] Extracting boundary representation data of the boundary part from the accurate boundary surface to construct an attribute adjacency graph, combining the attribute adjacency graph with the gap feature library, and converting the matching between the attribute adjacency graph and the gap feature library into a subgraph isomorphism problem. The subgraph isomorphism problem is solved using a VF3 subgraph isomorphism algorithm enhanced by a graph neural network to obtain the gap of the boundary part;

[0012] The boundary parts, the accurate boundary surfaces, and the gaps of the boundary parts are marked in the three-dimensional aircraft model.

[0013] Furthermore, the interference detection method sets the minimum value of the thickness of the thinnest part of the part and the minimum distance between the part and the cavity model as the minimum allowable gap of the interference detection method.

[0014] Furthermore, the threshold recommendation mechanism based on the decision tree model includes:

[0015] Extracting characteristic parameters of the three-dimensional model as input variables of a decision tree model, wherein the characteristic parameters include a minimum wall thickness of a part and a minimum distance between any two parts;

[0016] The extracted feature parameters are input into a decision tree model trained with prior knowledge to obtain the optimal minimum allowable gap value of the current three-dimensional model, which is set as the threshold for boundary part judgment.

[0017] Furthermore, the filtering out of adjacent facets includes:

[0018] Store the closed cavity facets as set A and the part surface facets as set B, and set the minimum distance threshold and normal vector angle threshold between adjacent facets;

[0019] Calculate the normal vectors of all faces in set A and set B, the distance between each face in set A and each face in set B, and the normal vector angle, and filter out the faces in set B that meet the preset threshold and store them in set C.

[0020] Furthermore, the stitching of the adjacent facets obtained by screening into an accurate boundary surface is to perform topological stitching, merge common edges, and generate an accurate boundary surface of a closed cavity.

[0021] Furthermore, the step of extracting the boundary representation data of the part includes:

[0022] Read the B-rep data structure of the part, extract the detailed properties of each entity element of the part, assign a globally unique number to each entity element, and construct a property set, including:

[0023] (1) Face attribute set:

[0024] F i =(t i ,c i ,E i ,G i ,N i )

[0025] Among them, F i is the face number; t i is the surface type, including plane, cylinder, sphere, and free-form surface; c i It is the surface convexity attribute, which is determined by the sign of the curvature at the centroid: positive for convex, negative for concave, and zero for flat; E i is the edge set associated with the face, G i is the geometric centroid position of the patch, N i is the normal vector of the patch at the centroid;

[0026] (2) Edge attribute set:

[0027] E j =(t j ,c j ,F aj ,F bj ,P s ,P e ,P m ,D j )

[0028] Among them, E j is the edge number; t j is the edge type, including straight line, arc, and free curve; c j is the edge concavity, which is determined by analyzing the angle between the two side normals; F aj 、F bj are the numbers of the two adjacent faces connecting the edge, Ps is the starting point of the edge, P m is the edge endpoint, P s is the midpoint of the side; D j is the edge direction vector, from P s Point to P e .

[0029] Furthermore, the construction of the attribute adjacency graph is represented as G=(V, E), including:

[0030] Define all the facets in the corresponding part as nodes to form a node set V; each node attribute includes the facet type t i 、Concavity c i , center of mass coordinate G i , normal vector N i , and edge set E i ;

[0031] The common edge between any two faces is defined as an edge, forming an edge set E; the attributes of each edge include edge type t j 、Concavity c j , direction vector D j ;

[0032] The connection relationship matrix A is defined as:

[0033]

[0034] Furthermore, the predefined gap feature library includes:

[0035] For typical gap morphology, standard gap features are predefined and organized and stored in the form of attribute adjacency subgraphs. Each type of gap feature is represented as a subgraph template G b =(V b ,E b ), V b Node representing the gap feature patch, E b Represents the topological connection relationship between gap feature patch nodes.

[0036] Furthermore, matching the attribute adjacency graph with the gap feature library is transformed into a subgraph isomorphism problem, including:

[0037] There is an injection mapping M:V b →V, so that the following two conditions are met:

[0038] (1) Node attribute consistency: For any v∈V b , satisfying Attr(v)=Attr(M(v)), where Attr(v) represents the attribute set of node v, including surface type and concavity;

[0039] (2) Edge attribute consistency: For any (u,v)∈E b , satisfying (M(u),M(v))∈E and attr(u,v)=Attr(M(u),M(v)), where u and v are two nodes in the gap feature subgraph, and (u,v) represents an edge in the gap feature subgraph.

[0040] Furthermore, the VF3 subgraph isomorphism algorithm solves the subgraph isomorphism problem by using the subgraph template G in the feature library in the attribute adjacency graph G = (V, E). b =(V b ,E b ) as matching targets, including:

[0041] (1) According to the node attributes, all node pairs with the same face attributes are stored in the set C0:

[0042] C0={(V b ,V)|Attr(V b )=Attr(V)}

[0043] (2) Add a new node V′ in the template graph b Match to a node V′ in the main graph, for each candidate node pair (V′ b ,V′) constructs its context attribute feature f(V′) b ), f(V′), the context attribute features are the original input features of the node, including face type, normal vector, and concavity;

[0044] h(V′ b )=GNN(f(V′ b ))

[0045] h(V′)=GNN(f(V′))

[0046] h(·) is a fixed-dimensional vector that concatenates the representations of the two nodes and feeds them into a multilayer perceptron classifier for scoring. The result is converted to a probability value between 0 and 1 through Sigmoid activation. A threshold θ is set, and nodes whose output probability value is lower than the threshold θ are pruned.

[0047] (3) Let the currently established match be M k , extended to M k+1 When VF3 Look-Ahead condition is satisfied, select the node pair (V′ b ,V′) join the matching, C k is the candidate set in the k-th iteration; when Attr(V′ b )=Attr(V′)∧CM final (V′ b,V′)>θ, there is M k+1 =M k ∪{(V′ b ,V′)};

[0048] In each expansion step, the node attributes and edge attributes are strictly constrained, including: node type t i 、Concavity c i Exactly the same; the type of the connecting edge is t j 、Concavity c j 、Direction D j consistent;

[0049] If the candidate set is empty or the attribute consistency check fails in a certain step of expansion, the backtracking operation is performed and returns to the previous step M k-1 , try other unexplored paths;

[0050] (4) When all nodes of the attribute adjacency graph have completed searching, the matching is terminated and an output representing the gap of the boundary part is obtained.

[0051] Beneficial Effects: This invention combines contact surface materialization with interference detection technology to accurately identify conductive surfaces, avoiding misjudgments and omissions. It is suitable for modeling assembly areas in the aviation manufacturing field with small sample sizes and high precision requirements. Compared with existing technologies, this invention has the following significant advantages:

[0052] (1) Introducing a machine learning threshold recommendation mechanism to achieve adaptive parameter setting. The present invention automatically analyzes part structural features (such as minimum wall thickness, gap between parts) and historical interference sample data by introducing a threshold recommendation mechanism based on decision tree learning, and constructs a training model to predict the optimal interference judgment threshold in the current scenario. Compared with the traditional method of relying on manual experience to set fixed parameters, this method realizes dynamic parameter adjustment and adaptive matching, improves the intelligence level and robustness of interference detection, and significantly reduces the problem of false alarms or missed alarms caused by improper human parameter selection. At the same time, the present invention can complete the boundary part identification task of complex assembly models within minutes, greatly reducing manual intervention and improving detection efficiency, which is particularly evident in the analysis of closed cavities with multiple modules and multiple interfaces.

[0053] (2) Automatic identification of high-complexity gaps and reduced reliance on experience. Existing methods often have difficulty in identifying polygonal, irregular or oblique gaps in complex cavities and require manual intervention. The present invention uses a rule-driven graph structure feature recognition algorithm, combined with local curvature and boundary continuity analysis, to intelligently identify and classify potential gap areas formed between heterogeneous parts, thereby achieving accurate extraction of "hidden gaps" that are difficult to handle with traditional methods. The present invention also enhances the context pruning mechanism of the VF3 subgraph isomorphism algorithm by introducing a graph neural network (GNN), significantly improving the accuracy and intelligence of matching. Compared with traditional static rule-based pruning strategies, GNN can dynamically predict the extension feasibility of candidate node pairs based on the geometry, topology and local structural properties of the nodes, thereby effectively reducing mis-pruning and improving matching recall and robustness. The method has self-supervised training capabilities and can continuously optimize the pruning discriminant model through historical matching data to achieve a learning-driven matching process and adapt to more diverse gap shapes and assembly structures.

[0054] (3) Strong adaptability, suitable for industrial modeling scenarios with few samples and high complexity. This method can adapt to the 3D model formats of various mainstream modeling platforms and supports stable recognition of closed assembly cavities with uncertain boundary changes, high structural complexity, and scarce sample data. It has good engineering feasibility and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is an overall flow chart of the method of the present invention.

[0056] Figure 2 It is a schematic diagram of the parts and cavity models in the assembly area;

[0057] Figure 3 It is a flow chart of a threshold parameter adaptive recommendation method based on a decision tree;

[0058] Figure 4 Schematic diagram of the notch feature in this embodiment;

[0059] Figure 5 This is a flowchart of subgraph isomorphism matching and pruning optimization based on the GNN-enhanced VF3 algorithm. DETAILED DESCRIPTION

[0060] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0061] like Figure 1 As shown, a method for identifying boundary features of a three-dimensional model of an aircraft closed cavity includes the following steps:

[0062] S1: Analyze the 3D model file and extract the geometric data and topological structure of the parts in the selected assembly area. The boundary parts and closed cavity models in the assembly area are as follows: Figure 2 As shown in the figure, an interference detection algorithm is used to calculate the distance between the part and the closed cavity pair by pair. If the distance is less than a set threshold, it is identified as an interference area and the corresponding part combination is marked. Finally, a graphical report is generated containing the location of the interference point, the name of the boundary part, and the geometric characteristics. To improve detection accuracy and adapt to the part complexity of different assembly scenarios, a decision tree-based threshold parameter adaptive recommendation method is proposed to intelligently set the minimum allowable clearance parameters and avoid false detection and missed detection caused by manual experience.

[0063] Specifically, the method includes the following steps:

[0064] S1.1: Read 3D model files and parse geometric data and topological structures.

[0065] Using the standard STEP (STP) format as an intermediate exchange model, the OpenCASCADE Technology (OCC) library is parsed and processed using the Python programming language to construct the structured part data model required for subsequent analysis. CAD part models from various sources (CATIA, SolidWorks, Creo, etc.) are uniformly converted into the ISO 10303-compliant .stp file format. This process is automated using the batch export functionality or API interface of commercial CAD software.

[0066] Use OCC.STEPControl.STEPControl_Reader to load the STP file and parse it to generate the TopoDS_Shape data structure as a unified representation of the geometry; construct a hierarchical index graph of topological elements such as faces, edges, and vertices for fast access and subsequent processing. The geometric information extracted from it includes:

[0067] Face attributes: extract the geometric parameters corresponding to each TopoDS_Face (such as curvature, normal vector, area);

[0068] Edge attributes: extract boundary type (straight line, arc, spline), edge length, and endpoint coordinates;

[0069] Vertex: records spatial position coordinates and is used to construct a mesh connection graph;

[0070] Planes, cylinders, cones, free-form surfaces, etc. are determined through the GeomAdaptor_Surface interface.

[0071] Based on the above geometric information, a topological adjacency graph of the part is constructed to record the association relationship between each topological unit, including face-edge adjacency, edge-vertex connection, and connection type (common edge) between adjacent faces.

[0072] S1.2: Set the interference detection standard. In traditional methods, the minimum allowable gap needs to be specified based on manual experience or preset parameters. In order to improve the level of automation, the present invention introduces a threshold recommendation mechanism based on decision tree learning. By inputting features such as the minimum wall thickness of parts, the minimum gap between parts, and historical interference statistics, the classification model is trained to predict the optimal minimum allowable gap value in the current scenario, realize dynamic parameter setting, and enhance the adaptability and versatility of the algorithm. Figure 3 The figure shows the process of the threshold recommendation mechanism, including:

[0073] S1.2.1: Feature Parameter Collection. Extract multiple representative geometric and structural feature parameters from the current 3D assembly model as model input variables. These feature parameters include: minimum part wall thickness, minimum spacing between any two parts in the assembly, and the interference ratio of parts in historical assembly scenarios (i.e., the ratio of the number of historical interference occurrences to the total number of occurrences). These feature parameters are input into the decision tree model as vectors.

[0074] S1.2.2: Model Training and Construction. A decision tree regression model is trained based on multiple annotated historical assembly samples, where the training samples contain feature parameters and their corresponding optimal minimum allowable gap values. Modeling is performed using supervised learning, with mean squared error (MSE) used as the evaluation metric. Hyperparameters such as decision tree depth and minimum leaf node sample count are optimized through cross-validation to construct a generalizable prediction model.

[0075] S1.2.3: Online recommendation. Before performing interference detection, input the characteristic parameters of the current assembly model into the trained decision tree model to obtain the optimal minimum allowable clearance value θ under the assembly scenario. rec This threshold will be used as the criterion for subsequent interference judgment.

[0076] S1.2.4: Parameter injection. The recommended minimum allowable gap value θ rec The automatic injection interference detection module replaces manual parameter setting, so that the minimum distance between the part and the cavity will be marked as interference when it is less than this value.

[0077] S1.3: Combine the adaptive threshold θ obtained in the previous step rec ,The collision detection algorithm is used to calculate the minimum distance between the part and the cavity pair by pair. If the distance is less than the minimum allowable gap, it is marked as a boundary part.

[0078] S1.4: Generate a graphical report, marking the interference point coordinates and corresponding part information.

[0079] S2: Extract mesh data from boundary components and closed cavity models, calculate the angle between the face normal vectors of each face, and select matching adjacent faces through distance filtering and normal angle analysis. Boundary components include beams, trusses, and ribs. Using a topological stitching algorithm, these selected faces are stitched together into a complete and accurate boundary surface, ensuring surface closure and geometric continuity.

[0080] S2.1: Input the surface patch sets of the closed cavity and boundary parts, which are stored as set A (cavity patches) and set B (part patches) respectively;

[0081] S2.2: Calculate the normal vector of each facet and set the minimum distance threshold and normal vector angle threshold between adjacent faces.

[0082] The numerical calculation method of the normal vector angle between adjacent facets is as follows:

[0083] The general formula for a three-dimensional surface equation is f(x,y,z)=0, where (x,y,z) are the rectangular coordinate variables in three-dimensional space. The normal vector of a three-dimensional surface f(x,y,z)=0 at point f(x0,y0,z0) is as follows:

[0084]

[0085] in It is the maximum direction of the directional derivative of the surface at the point (x0, y0, z0), that is, the direction of the local normal vector, which can be used as the approximate normal vector at the centroid of the patch.

[0086] Patch normal angle The calculation method is as follows:

[0087] Given two facet normal vectors

[0088]

[0089] First calculate the dot product of the two vectors:

[0090]

[0091] Then calculate the modulus of the two vectors:

[0092]

[0093] Then calculate the cosine angle between the two vectors. To prevent the floating-point error from exceeding the domain of definition, the cosθ value needs to be limited to [-1, 1]:

[0094]

[0095] Convert radians to degrees, and finally get the angle value between the normal vectors of adjacent facets:

[0096]

[0097] S2.3: Loop through the distance and normal angle between each face in set A and all faces in set B, and select the boundary surfaces that meet the preset threshold and store them in set C.

[0098] S2.4: Perform topological stitching on the faces in set C, merge common edges, and generate a closed cavity-accurate boundary surface.

[0099] S3: Extract the part's boundary representation data to construct an attribute adjacency graph, defining the geometric properties and connectivity of the facets. Combined with a predefined gap feature library, this algorithm employs subgraph isomorphism matching and pruning optimization based on the GNN-enhanced VF3 algorithm to achieve high-precision gap recognition.

[0100] S3.1: For each input part, first read its complete B-rep data structure, extract the detailed properties of the face and edge that constitute the part in turn, and assign a globally unique number to each entity element to form a structured attribute description set.

[0101] The attribute set of a face is defined as:

[0102] F i =(t i ,c i ,E i ,G i ,N i )

[0103] in:

[0104] F i : Surface number, globally unique number;

[0105] t i : Surface type, such as plane, cylinder, sphere, free-form surface, etc.;

[0106] c i : Surface concavity attribute, determined by the sign of the curvature at the centroid, positive for convex, negative for concave, and zero for flat;

[0107] E i : The edge set associated with the face, a set consisting of edge numbers;

[0108] G i : The geometric centroid position of the patch, three-dimensional coordinates (x, y, z);

[0109] N i : Normal vector of the face at the centroid, three-dimensional direction vector.

[0110] The attribute set of an edge is defined as:

[0111] E j =(t j ,c j ,F aj ,F bj ,P s ,P e ,P m ,D j )

[0112] in:

[0113] E j : edge number, globally unique number;

[0114] t j : Edge type, straight line, arc, free curve;

[0115] c j : The edge concavity is determined by analyzing the angle between the two side normals;

[0116] F aj : The numbers of the two adjacent faces connecting the edge, F a noodle;

[0117] F bj : The numbers of the two adjacent faces connecting the edge, F b noodle;

[0118] P s ,P e ,P m : The three-dimensional coordinates of the starting point, end point and midpoint of the edge;

[0119] D j : Edge direction vector, from P s Point to P e .

[0120] S3.2: Attribute adjacency graph construction. After extracting the complete face and edge attributes, further organize the local structural relationships of the part and construct the attribute adjacency graph. The attribute adjacency graph G = (V, E) is defined as follows:

[0121] Node set V: corresponds to all the faces in the part;

[0122] Edge set E: indicates that any two faces are connected by a common edge.

[0123] The connection relationship matrix A is defined as:

[0124]

[0125] Node attribute definition: Each node attribute includes the type of patch t i 、Concavity ci , center of mass coordinate G i , normal vector N i , and edge set E i .

[0126] Edge attribute definition: Each edge attribute includes the common edge attributes corresponding to the two connected faces, including edge type t j 、Concavity c j , direction vector D j .

[0127] S3.3: Gap feature library construction. For typical gap morphology, pre-defined standard gap features are organized and stored in the form of attribute adjacency subgraphs. Each type of gap feature is represented as a subgraph template G b =(V b ,E b ) for subsequent matching and identification. b Node representing the gap feature patch, E b Represents the topological connection relationship between the gap feature patch nodes. Figure 4 Shown is a notch feature in this embodiment.

[0128] The gap feature subgraph mainly includes:

[0129] (1) Chamfer notch feature subgraph: It contains an outer cylindrical surface node and its two adjacent plane nodes; the surfaces are connected by smooth connections, that is, the angle between the connecting edges is close to 180°;

[0130] (2) Bend notch feature subgraph: It contains two surface nodes, namely the outer cylindrical surface and the inner cylindrical surface, which satisfy the following geometric relationship:

[0131] The central axis is in the same direction (normal vectors are parallel): |N1×N2|≈0;

[0132] The distance between the centroids of the cylindrical surfaces satisfies the given sheet metal thickness d: Distance(G1,G2)≈d;

[0133] The central angles are equal and the curvatures have opposite signs (convex and concave).

[0134] S3.4 Subgraph isomorphism matching and pruning optimization based on GNN-enhanced VF3 algorithm. In the attribute adjacency graph G = (V, E) of the part, the subgraph template G in the feature library is used. b =(V b ,E b ) is used as the matching target, and the VF3 subgraph isomorphism algorithm is applied to find whether there is a gap feature isomorphic to it in the local area.

[0135] Compared with VF2 commonly used in the prior art, VF3 originally relied on static context constraint rules for early pruning in the process of searching for candidate expansion node pairs. In order to improve pruning accuracy and matching efficiency, the present invention upgrades the static rules to a dynamic learning context scoring model CM. final (V′ b ,V′), introduces the context pruning mechanism enhanced by graph neural network (GNN), and modifies the context matrix CM(V′) in traditional VF3. b ,V′) to perform intelligent prediction and correction, which greatly reduces erroneous pruning behaviors and improves the overall recall rate and operation efficiency.

[0136] The subgraph isomorphism problem is formally defined as: there exists an injection mapping M:V b →V, so that the following two conditions are met:

[0137] (1) Node attribute consistency: For any v∈V b , satisfying Attr(v)=Attr(M(v)), where Attr(v) represents the attribute set of node v, including surface type, concavity and convexity and other features.

[0138] (2) Edge attribute consistency: For any (u,v)∈E b , satisfying (M(u),M(v))∈E and Attr(u,v)=Attr(M(u),M(v)), where: u,v are two nodes in the gap feature subgraph, and (u,v) represents an edge in the gap feature subgraph.

[0139] like Figure 5 The following is the basic process of the VF3 enhancement algorithm in the present invention, including:

[0140] S3.4.1 Initialization. According to the node attributes, find all the node pairs with the same face attributes and put them into the candidate set C0, with C0 = {(V b ,V)|Attr(V b )=Attr(V)}.

[0141] S3.4.2GNN prediction (context pruning score). Consider a new node V′ in the template graph b Match to a node V′ in the main graph, for each candidate node pair (V′ b ,V′) constructs its context attribute feature f(V′) b ), f(V′), and sent to the GNN model for expansion feasibility scoring:

[0142] h(V′ b )=GNN(f(V′ b ))

[0143] h(V′)=GNN(f(V′))

[0144] The contextual attribute features are the original input features of the node, including face type, normal vector, and concavity. h(·) is a fixed-dimensional vector representing the contextual structural features of the node. The representations of the two nodes are concatenated and fed into a multi-layer perceptron classifier for scoring. The result is converted to a probability value between 0 and 1 through Sigmoid activation:

[0145] CM′(V′ b ,V′)=σ(MLP([h(V′ b )||h(V′)]))

[0146] The final context matrix is ​​generated by fusing the original rule matrix with the GNN output:

[0147] CM final (V′ b ,V′)=CM(V′ b ,V′)·CM′(V′ b ,V′)

[0148] If the rule judgment output is 0 (no match), prune directly; if the rule judgment output is 1, GNN will give further probability score; set the threshold θ (take 0.5) and judge:

[0149] CM final (V′ b ,V′)>θ

[0150] If so, the node is considered promising for expansion and is retained; otherwise, it is considered to have little potential and is pruned in advance to save subsequent search overhead.

[0151] S3.4.3 Recursive Expansion and Look-Ahead Mechanism: Let the currently established match be M k , extended to M k+1 When VF3 Look-Ahead condition is satisfied, select the node pair (V′ b ,V′) join the matching, C k is the candidate set in the k-th iteration.

[0152] When Attr(V′ b )=Attr(V′)∧CM final (V′ b ,V′)>θ, there is M k+1 =M k ∪{(V′ b ,V′)}. Among them, Attr(V′ b)=Attr(V′) represents the node V′ b Consistent with the attributes of node V′.

[0153] S3.4.4 Attribute consistency constraints: In each expansion step, the node attributes and edge attributes are strictly constrained, including: node type t i 、Concavity c i Exactly the same; the type of the connecting edge is t j 、Concavity c j 、Direction D j consistent.

[0154] S3.4.5 Backtracking search: If the candidate set is empty or the attribute consistency check fails in a certain step of expansion, a backtracking operation is performed and the query returns to the previous step M. k-1 , try other unexplored paths.

[0155] S3.4.6 Matching Termination Judgment: When all nodes in the part attribute adjacency graph have completed the search, the matching is terminated. The matching result represents the gap in the corresponding boundary part in the attribute adjacency graph.

[0156] S4 marks the boundary parts, accurate boundary surfaces, and gaps identified in the previous steps in the aircraft three-dimensional model to obtain a visual model with closed cavity boundary feature information for use in research and application.

Claims

1. A method for identifying boundary features of a three-dimensional model of an aircraft closed cavity, characterized in that: include: Analyze the 3D aircraft model including the enclosed cavity, extract the enclosed cavity model and all part models within the assembly area, and sequentially identify the boundary features of the enclosed cavity. These boundary features include: boundary parts that enclose the cavity space, the precise boundary surface formed by the boundary parts, and gaps in the cavity space. An interference detection method is used to calculate the distance between each part model and the closed cavity model. A threshold recommendation mechanism based on a decision tree model is proposed to autonomously set the threshold. Parts with a distance less than the set threshold are marked as boundary parts. Extracting a closed cavity patch from the closed cavity model and a boundary part surface patch from the boundary part model, calculating the surface normal vector angle and distance between the closed cavity patch and the boundary part surface patch to screen adjacent patches, and splicing the adjacent patches to obtain an accurate boundary surface; Extracting boundary representation data of the boundary part from the accurate boundary surface to construct an attribute adjacency graph, combining the attribute adjacency graph with the gap feature library, and converting the matching between the attribute adjacency graph and the gap feature library into a subgraph isomorphism problem. The subgraph isomorphism problem is solved using the VF3 subgraph isomorphism algorithm enhanced by a graph neural network to obtain the gap of the boundary part; The boundary parts, accurate boundary surfaces and gaps of boundary parts are marked in the three-dimensional aircraft model.

2. The three-dimensional model boundary feature recognition method according to claim 1, characterized in that: The interference detection method uses the thickness of the thinnest part of the part and the minimum distance between the part and the cavity model, and takes the minimum value of the two values ​​as the minimum allowable gap of the interference detection method.

3. The three-dimensional model boundary feature recognition method according to claim 2, characterized in that: The threshold recommendation mechanism based on the decision tree model includes: Extracting characteristic parameters of the three-dimensional model as input variables of a decision tree model, wherein the characteristic parameters include a minimum wall thickness of a part and a minimum distance between any two parts; The extracted feature parameters are input into a decision tree model trained with prior knowledge to obtain the optimal minimum allowable gap value of the current three-dimensional model, which is set as the threshold for boundary part judgment.

4. The three-dimensional model boundary feature recognition method according to claim 3, characterized in that: The screening out of adjacent patches includes: Store the closed cavity facets as set A and the part surface facets as set B, and set the minimum distance threshold and normal vector angle threshold between adjacent facets; Calculate the normal vectors of all faces in set A and set B, the distance and normal vector angle between each face in set A and each face in set B, and select the faces in set B that meet the preset threshold and store them in adjacent face set C.

5. The three-dimensional model boundary feature recognition method according to claim 4, characterized in that: The stitching of the adjacent facets obtained by screening into an accurate boundary surface is to perform topological stitching, merge common edges, and generate an accurate boundary surface of a closed cavity.

6. The three-dimensional model boundary feature recognition method according to claim 5, characterized in that: The process of extracting the boundary representation data of the part, reading the B-rep data structure of the part, extracting the detailed attributes of each entity element of the part, assigning a globally unique number to each entity element, and constructing an attribute set includes: (1) Face attribute set: F i =(t i ,c i ,E i ,G i ,N i ) Among them, F i is the face number; t i is the surface type, including plane, cylinder, sphere, and free-form surface; c i It is the surface convexity attribute, which is determined by the sign of the curvature at the centroid: positive for convex, negative for concave, and zero for flat; E i is the edge set associated with the face, G i is the geometric centroid position of the patch, N i is the normal vector of the patch at the centroid; (2) Edge attribute set: E j =(t j ,c j ,F aj ,F bj ,P s ,P e ,P m ,D j ) Among them, E j is the edge number; t j is the edge type, including straight line, arc, and free curve; c j is the edge concavity, which is determined by analyzing the angle between the two side normals; F aj 、F bj are the numbers of the two adjacent faces connecting the edge, P s is the starting point of the edge, P m is the edge endpoint, P s is the midpoint of the side; D j is the edge direction vector, from P s Point to P e .

7. The three-dimensional model boundary feature recognition method according to claim 6, characterized in that: The constructed attribute adjacency graph is represented as G=(V, E), including: Define all the facets in the corresponding part as nodes to form a node set V; each node attribute includes the facet type t i 、Concavity c i , center of mass coordinate G i , normal vector N i , and edge set E i ; The common edge between any two faces is defined as an edge, forming an edge set E; the attributes of each edge include edge type t j 、Concavity c j , direction vector D j ; The connection relationship matrix A is defined as:

8. The three-dimensional model boundary feature recognition method according to claim 7, characterized in that: The predefined gap feature library includes: For typical gap morphology, standard gap features are predefined and organized and stored in the form of attribute adjacency subgraphs. Each type of gap feature is represented as a subgraph template G b =(V b ,E b ), V b Node representing the gap feature patch, E b Represents the topological connection relationship between gap feature patch nodes.

9. The three-dimensional model boundary feature recognition method according to claim 8, characterized in that: The matching of the attribute adjacency graph and the gap feature library is transformed into a subgraph isomorphism problem, and an injection mapping M:V is set. b →V, so that it satisfies the following conditions: (1) Node attribute consistency: For any v∈V b , satisfying Attr(v)=Attr(M(v)), where Attr(v) represents the attribute set of node v, including surface type and concavity; (2) Edge attribute consistency: For any (u,v)∈E b , satisfying (M(u),M(v))∈E and Attr(u,v)=Attr(M(u),M(v)), where u and v are two nodes in the gap feature subgraph, and (u,v) represents an edge in the gap feature subgraph.

10. The three-dimensional model boundary feature recognition method according to claim 9, characterized in that: The VF3 subgraph isomorphism algorithm solves the subgraph isomorphism problem by using the subgraph template G in the feature library in the attribute adjacency graph G = (V, E). b =(V b ,E b ) as matching targets, including: (1) According to the node attributes, all node pairs with the same face attributes are stored in the set C0: C0={(V b ,V)|Attr(V b )=Attr(V)} (2) Add a new node V′ in the template graph b Match to a node V′ in the main graph, for each candidate node pair (V′ b ,V′) constructs its context attribute feature f(V′) b ), f(V′), the context attribute features are the original input features of the node, including face type, normal vector, and concavity; h(V′ b )=GNN(f(V′ b )) h(V′)=GNN(f(V′)) h(·) is a fixed-dimensional vector that concatenates the representations of the two nodes and feeds them into a multilayer perceptron classifier for scoring. The result is converted to a probability value between 0 and 1 through Sigmoid activation. A threshold θ is set, and nodes whose output probability value is lower than the threshold θ are pruned. (3) Let the currently established match be M k , extended to M k+1 When VF3 Look-Ahead condition is satisfied, select the node pair (V′ b ,V′) join the matching, C k is the candidate set in the k-th iteration; when Attr(V′ b )=Attr(V′)∧CM final (V′ b ,V′)>θ, there is M k+1 =M k ∪{(V′ b ,V′)}; In each expansion step, the node attributes and edge attributes are strictly constrained, including: node type t i 、Concavity c i Exactly the same; the type of the connecting edge is t j 、Concavity c j 、Direction D j consistent; If the candidate set is empty or the attribute consistency check fails in a certain step of expansion, the backtracking operation is performed and returns to the previous step M k-1 , try other unexplored paths; (4) When all nodes of the attribute adjacency graph have completed searching, the matching is terminated and an output representing the gap of the boundary part is obtained.

Images