A part space geometry feature and process knowledge automatic identification and reasoning system

Abstract

The application relates to the technical field of automatic feature recognition, in particular to a part space geometric feature and process knowledge automatic recognition reasoning system, which comprises a model acquisition module, a curved surface segmentation module, a curved surface evaluation module and a feature recognition and process reasoning module.The model acquisition module acquires a B-Rep model of a part, and obtains triangular facets in each face in the B-Rep model.The curved surface segmentation module calculates a region attribution degree, and divides the face into multiple sub-regions.The curved surface evaluation module calculates a discriminant coefficient, and distinguishes all the faces into process faces and auxiliary faces.The feature recognition and process reasoning module determines the sampling precision of each face, discretizes the face and the edge, constructs an attribute adjacency graph of the coded B-Rep model information, and processes the part based on the machining feature recognition result.The application improves the machining process feature recognition precision of the part.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of automatic feature recognition, and in particular to a part space geometric feature and process knowledge automatic recognition and reasoning system. BACKGROUND

[0002] In the intelligent manufacturing industry, a computer-aided design (CAD) model is a core carrier for describing geometric information of a part, and a boundary representation model can accurately define the geometric shape and topological structure of the part. Automatically identifying machining features with specific manufacturing significance from the CAD model of the part and reasoning a reasonable machining process therefrom is a key link for realizing computer-aided process planning (CAPP) automation and intelligentization.

[0003] A traditional learning-based machining feature recognition method usually constructs an attribute adjacency graph and adopts a graph neural network structure to perform feature segmentation and recognition on a part model. Since the intersection relationship of multiple machining features on the part is often complex, the machining features are coupled and interfered with each other, which may cause the feature topology to be destroyed, making it difficult for the graph neural network to accurately learn, resulting in feature misrecognition or missed recognition. In addition, when discretizing different surfaces, a fixed sampling accuracy is usually adopted. This rough strategy cannot adapt to the huge geometric complexity difference between different surfaces of the part, resulting in a large amount of redundant calculation for surfaces with simple process features, and insufficient sampling for surfaces with complex process features, which cannot accurately capture subtle but critical geometric features, causing feature boundary blur and geometric information loss, affecting the recognition accuracy of the machining features, and further leading to part feature misjudgment and process knowledge reasoning error. SUMMARY

[0004] To solve the above technical problems, a part space geometric feature and process knowledge automatic recognition and reasoning system is provided to solve the existing problems.

[0005] The technical problem of the application is solved by providing a part space geometric feature and process knowledge automatic recognition and reasoning system, which comprises:

[0006] A model acquisition module acquires a B-Rep model of a part, performs grid processing on the B-Rep model, and obtains triangular facets located in each surface in the B-Rep model;

[0007] A surface segmentation module analyzes the angle deviation between two adjacent triangular facets in each surface and the bending difference of the vertex connected by the common edge in the local range, calculates the region attribution degree of the two adjacent triangular facets, constructs an undirected weighted graph of each surface, and divides the surface into multiple sub-regions;

[0008] a curved surface evaluation module, configured to calculate a discriminant coefficient of each face by the discreteness of the bending features of the vertices of the triangular patches within each sub-region of the face in a local range and the number of the divided sub-regions, and to distinguish all the faces into process faces and auxiliary faces; to evaluate the continuity of the concave-convex variation of the edges of the triangular patches located on the boundary of different sub-regions in each process face, to calculate a continuity evaluation value of each process face, and to obtain the structural complexity of each process face in combination with the number of the sub-regions in the process face and the discreteness of the bending features thereof;

[0009] a feature recognition and process inference module, configured to determine the sampling precision of each face by the structural complexity, to discretize the faces and edges in the B-Rep model, to construct an attribute adjacency graph of the encoded B-Rep model information, and to process the part based on the machining feature recognition result thereof.

[0010] Preferably, the calculation of the region attribution degree of the two adjacent triangular patches comprises:

[0011] calculating the dihedral angle between the two adjacent triangular patches and taking the difference between the dihedral angle and a flat angle as a relative deviation;

[0012] calculating the similarity degree between the normal vectors of the two adjacent triangular patches and performing a positive mapping thereon; taking the ratio between the relative deviation and the result of the positive mapping as a direction difference degree of the two adjacent triangular patches;

[0013] taking all the vertices of the triangular patches connected with any vertex in the B-Rep model as adjacent points; performing a curved surface fitting on the any vertex and all the adjacent points thereof, and calculating the curvature of the fitted curved surface at the any vertex;

[0014] taking the difference between the curvatures of the two vertices connected by the common edge of the two adjacent triangular patches as a curvature difference, and performing a normalization processing on the curvature difference;

[0015] the region attribution degree is a result of a negative mapping of the product of the normalized result and the direction difference degree.

[0016] Preferably, the construction of the undirected weighted graph of each face comprises:

[0017] abstracting each triangular patch on each face in the B-Rep model as a graph node, connecting a graph node corresponding to each triangular patch with a graph node corresponding to a directly adjacent triangular patch with an edge, taking the region attribution degree of the two adjacent triangular patches as an edge weight of the edge, and constructing an undirected weighted graph;

[0018] performing a spectral clustering algorithm on the triangular patches corresponding to all the graph nodes in the weighted undirected graph, and merging all the triangular patches in each clustering cluster as each sub-region of each face in the B-Rep model.

[0019] Preferably, the calculating the discriminant coefficient of each face comprises:

[0020] The average of the corresponding curvatures at the vertices of all triangular facets in each sub-region is denoted as the regional curvature; the dispersion degree of the regional curvatures of all sub-regions in each face of the B-Rep model is calculated; the number of all sub-regions of each face of the B-Rep model is counted;

[0021] The discriminant coefficient is the product of the number and the dispersion degree.

[0022] Preferably, the distinguishing all faces into process faces and auxiliary faces comprises: obtaining a segmentation threshold of the discriminant coefficients of all faces of the B-Rep model; taking the faces with the discriminant coefficients greater than or equal to the segmentation threshold as the process faces, and taking the faces with the discriminant coefficients less than the segmentation threshold as the auxiliary faces.

[0023] Preferably, the calculating the continuity evaluation value of each process face comprises:

[0024] For each sub-region in each process face, all edges located on the boundary thereof are extracted as boundary edges.

[0025] For each boundary edge, the concave-convex property of the edge is judged; if each boundary edge is a convex edge, the concave-convex factor of the boundary edge is assigned as 1; if each boundary edge is a concave edge, the concave-convex factor of the boundary edge is assigned as -1; if each boundary edge is a smooth edge, the concave-convex factor of the boundary edge is assigned as 0.

[0026] The concave-convex factors of all boundary edges on the boundary of each sub-region are arranged in a clockwise sequence to form a concave-convex sequence; if the concave-convex factor of each boundary edge and the concave-convex factor of the next boundary edge in the concave-convex sequence are opposite numbers, the continuity weight value of each boundary edge is assigned as a preset first numerical value; if the concave-convex factor of each boundary edge and the concave-convex factor of the next boundary edge are the same, the continuity weight value of each boundary edge is assigned as a preset third numerical value; if there is only one 1 in the concave-convex factors of each boundary edge and the next boundary edge, the continuity weight value of each boundary edge is assigned as a preset second numerical value; wherein the preset first numerical value is less than the preset second numerical value, and the preset second numerical value is less than the preset third numerical value.

[0027] The continuity evaluation value is the sum of the continuity weight values of all boundary edges of all sub-regions in each process face.

[0028] Preferably, the obtaining the structural complexity of each process face comprises: taking the number of all sub-regions, the dispersion degree, and the continuity evaluation value of each process face as a feature vector; performing a positive processing on all components in the feature vector; using a comprehensive evaluation algorithm to comprehensively evaluate the feature vectors of all process faces to obtain a comprehensive score of each process face as the structural complexity of each process face.

[0029] Preferably, determining the sampling accuracy of each face includes: setting the structural complexity of the auxiliary face to a preset value; if the structural complexity of each face is distributed in... The sampling accuracy of that surface is then determined. for If the structural complexity is distributed in The sampling accuracy of that surface is then determined. for If the structural complexity is distributed in The sampling accuracy of that surface is then determined. for If the structural complexity is distributed in The sampling accuracy of that surface is then determined. for ;in, , , , as well as All are preset values, and .

[0030] Preferably, the construction of the attribute adjacency graph encoding B-Rep model information includes: extracting and encoding the geometric and attribute information of each face and each edge in the B-Rep model; concatenating the geometric and attribute information of each face and assigning it as a node feature to the attribute adjacency graph; concatenating the vectors corresponding to the geometric and attribute information of each edge and assigning them as edge features to the attribute adjacency graph; thereby obtaining the attribute adjacency graph encoding B-Rep model information.

[0031] Preferably, the extraction and encoding of geometric and attribute information for each face and each edge in the B-Rep model includes:

[0032] Obtain the attribute information of each face and each edge in the B-Rep model;

[0033] Convert all faces of the B-Rep model to UV coordinates. In UV coordinates, based on the sampling accuracy... Discretize each surface into The 2D mesh structure is used to obtain the 3D absolute coordinates, 3D absolute surface normal vectors, and clipping masks of all UV mesh points in each face, which serve as the geometric information of each face.

[0034] In UV coordinates, each edge in the B-Rep model is discretized into a 1D mesh structure with a preset size. The 3D absolute coordinates, tangent vector, and normal vector of each UV mesh point on each edge are obtained as the geometric information of each edge.

[0035] This application has at least the following beneficial effects:

[0036] The application calculates the area attribution degree of the two adjacent triangular facets by analyzing the local angle deviation and vertex bending difference between the two adjacent triangular facets, and has the beneficial effect that the change of the two adjacent triangular facets in the geometric shape direction and the case that the two triangular facets are located in different regions with large difference in bending degree are considered, the continuity and mutation of the in-plane geometric properties can be sensitively captured to evaluate the microscopic geometric discontinuous features of the two triangular facets; the undirected weight graph of each face is constructed, and the spectral clustering algorithm is used to divide the face into multiple sub-regions, and has the beneficial effect that the triangular facets with similar geometric features are divided into a sub-region by classifying the triangular facets, so that the macroscopic face is decomposed into sub-regions represented by different geometric features, which can effectively distinguish the sub-regions with different geometric structures and process semantics in each face of the B-Rep model, greatly reduce the interference caused by the coupling of complex features, and effectively reduce the topological damage and misjudgment; the discriminant coefficient of each face is calculated, and all the faces are divided into process faces and auxiliary faces, and has the beneficial effect that the geometric complexity and curvature change of each face are considered, and the process face bearing the main machining features and the auxiliary face mainly serving as the connection and support are accurately identified, wherein the process face usually has high geometric complexity and the auxiliary face usually has simple and gentle geometric features, and through the division, different sampling strategies are adopted for different types of faces in the subsequent process; the continuity evaluation value of each process face is calculated, and has the beneficial effect that the continuity of the concave-convex change on the boundary of the process face is considered, and the regularity of the process feature boundary is reflected; the structural complexity of each process face is obtained, and has the beneficial effect that the machining complexity of the process feature on the process face is comprehensively evaluated from the geometric complexity and boundary regularity; the sampling accuracy of each face is determined, and has the beneficial effect that the sampling accuracy is dynamically determined according to the machining complexity of each face, so that a smaller step length is adopted for the face with high complexity to capture more geometric structure details of the face, and a lower sampling rate is adopted for the face with low complexity to reduce the calculation amount, avoid the waste of computing power on the face with low complexity and the loss of features on the face with high complexity, so that the generated attribute adjacency graph is lightweight and information-rich, and the efficiency and accuracy of the subsequent graph neural network processing are greatly improved.The face and edge in the B-Rep model are discretized, the geometric information and attribute information of each face and each edge are extracted and coded, an attribute adjacency graph of the coded B-Rep model information is constructed, and a processing feature is recognized by using a graph neural network, based on the recognition result, a pre-defined process knowledge base is matched to generate a processing path and a process parameter, so that the part is processed, and the beneficial effect is that by fusing the attribute information and the geometric information of the face and the line, more rich feature descriptors are provided, the influence of misjudgment caused by the destruction or loss of topological information due to feature coupling interference is avoided, by establishing the attribute adjacency graph with high-level semantic information, not only the topological connection relationship of the B-Rep model can be accurately reflected, but also the deep processing semantic information is carried through the rich node and edge features, and the processing feature recognition accuracy of the part is improved. BRIEF DESCRIPTION OF DRAWINGS

[0037] The part space geometric feature and process knowledge automatic identification reasoning system provided by the application will be further described in detail below with reference to the drawings.

[0038] Figure 1 The part space geometric feature and process knowledge automatic identification reasoning system provided by the application will be further described in detail below with reference to the drawings.

[0039] Figure 2 The part space geometric feature and process knowledge automatic identification reasoning system provided by the application will be further described in detail below with reference to the drawings. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical scheme and advantages of the application more clear, the part space geometric feature and process knowledge automatic identification reasoning system provided by the application will be further described in detail below with reference to the drawings and the implementation examples. It should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the application.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs.

[0042] Please refer to Figure 1 , which shows a part space geometric feature and process knowledge automatic identification reasoning system block diagram provided by an embodiment of the application, the system comprises: a model acquisition module, a curved surface segmentation module, a curved surface evaluation module, a feature recognition and process reasoning module.

[0043] The model acquisition module acquires the B-Rep model of the part, performs meshing processing on it, and obtains the triangular facets located in each face in the B-Rep model.

[0044] In modern manufacturing, the processing and manufacturing of sheet metal parts is an important part of mechanical processing, sheet metal parts are usually processed by shearing, stamping, bending, forming and other processes from metal sheets, and are widely used in automobile, aerospace, electronic equipment, home appliances and other industries. With the development of automation and intelligentization of manufacturing industry, higher requirements are put forward for the processing efficiency and quality of sheet metal parts.

[0045] In the process of part processing and manufacturing, computer-aided design (CAD) software is used to create three-dimensional models of parts, which are mainly composed of basic geometric elements such as points, lines, and surfaces according to specific topological relationships. In order to accurately store and process this three-dimensional model, the computer generally uses the boundary representation method (B-rep) to describe the three-dimensional model. B-rep describes entities through their geometric boundaries. The boundary representation method generally represents the model in a hierarchical relationship of "volume-surface-loop-edge-point". It can accurately and completely represent the surfaces, edges, points and related relationships that make up the entity. A three-dimensional model of a part can be represented as a closed spatial region formed by the adhesion of a finite number of bounded surfaces. The boundary defines the internal and external points of the object. Specifically, the part is defined by the surface, each surface is defined by the boundary, each edge is defined by two vertices, and finally the vertex is defined by three coordinates. One of the main features of the B-Rep model is that it records the geometric information of the surfaces, edges, and points that make up the geometric body and the topological information of their mutual relationships in detail, and separates the geometric information and topological information of the entity.

[0046] The three-dimensional model created by the design personnel through the CAD software usually only contains design parameter information. Although these information describes the geometric shape and size of the part in detail, it does not directly contain processing-related process information and cannot be directly used by the computer-aided process planning (CAPP) system. In order to convert the CAD model into the process knowledge required by the CAPP system, it is necessary to identify the processing features in the B-Rep model and convert these features into the process information required by the CAPP system.

[0047] Based on the above analysis, the three-dimensional model of the part created by the CAD software, i.e. the CAD model, uses the B-Rep method to obtain the B-Rep model of the part;

[0048] The basic unit constituting the triangular mesh model is a triangular facet, and these triangular facets are all planes. Since the faces between the B-Rep model and the triangular mesh model are one-to-one corresponding, different triangular facets on the triangular mesh model are mapped to the B-Rep model to obtain each triangular facet located in each face on the B-Rep model, wherein each vertex on the triangular facet is a point on the face in the B-Rep model, which can represent the local surface feature and machining feature information of the part space curve;

[0049] In the embodiment, the three-dimensional modeling software adopts Blender software, wherein the Blender software is a known technology and will not be described here.

[0050] Thus, each triangular facet located in each face on the B-Rep model is obtained.

[0051] The curved difference of the connecting vertex of the common edge in the local range is analyzed, and the regional attribution degree of the adjacent two triangular facets is calculated to construct an undirected weighted graph of each face, so as to divide the face into a plurality of sub-regions.

[0052] In the design and manufacturing process of complex parts, the geometric body curve of the part is usually composed of a plurality of triangular facets, and different regions may have different structural features and process semantics. For example, a geometric body curve of a part may have machining features such as through holes, square grooves, chamfers, etc. In order to better process these complex geometric structures and process features, it is necessary to first perform similar merging processing on the triangular facets.

[0053] Secondly, the triangular facets of different regions may have different geometric features such as curvature and direction. For example, one region may be a plane, another region may be a curved surface, and may also contain machining features such as through holes, square grooves, chamfers, etc. These regions not only differ in geometry, but also have different process requirements in the machining process. For example, through holes need to be drilled, square grooves need to be milled, and chamfers need special cutting tools and paths. In actual machining, if each triangular facet is processed separately, the machining path will be complex and inefficient. Through similar merging processing, triangular facets with similar geometric features and process requirements can be merged into a larger region, thereby simplifying the machining path and improving the machining efficiency.

[0054] Based on the above analysis, the direction difference degree is calculated by analyzing the angle between adjacent triangular facets on each face in the B-Rep model, specifically as follows:

[0055] The dihedral angle between the adjacent two triangular facets on each face in the B-Rep model is calculated.

[0056] It should be noted that the dihedral angle calculation is a known technology, which will not be repeated here, wherein the dihedral angle value range is .

[0057] The similarity between the normal vectors of the two adjacent triangular facets is calculated, and a positive mapping is performed thereon;

[0058] In this embodiment, the normal vector acquisition is a known technology, which will not be repeated here; secondly, the similarity is calculated by the cosine similarity between the normal vectors of the two adjacent triangular facets, wherein the cosine similarity calculation is a known technology, which will not be repeated here; secondly, the process of positive mapping is: positive mapping is performed by an exponential function, wherein it is assumed that the similarity is denoted as , the result of is taken as the result of positive mapping, wherein is an exponential function with a natural constant as the base number, and the result of positive mapping is greater than 0 through the process of positive mapping.

[0059] The difference between the dihedral angle and the flat angle is denoted as the relative deviation, and the ratio between the relative deviation and the result of positive mapping is taken as the direction difference degree of the two adjacent triangular facets;

[0060] In this embodiment, the difference between the flat angle and the dihedral angle is denoted as the relative deviation, wherein the flat angle is .

[0061] It should be noted that the smaller the dihedral angle of the two adjacent triangular facets is, the greater the relative deviation is, the more acute the shape feature of the common edge between the two adjacent triangular facets is, which can be a very sharp edge or a deep valley, the greater the direction difference degree between the facet geometrical shapes is, the smaller the similarity is, the smaller the result of positive mapping is, the greater the overall orientation difference of the part space plane corresponding to the two adjacent triangular facets is, and the greater the obtained direction difference degree is, which indicates that the change of the two triangular facets in the geometric shape direction is more drastic.

[0062] Secondly, by analyzing the bending condition of the vertex connected by the common edge between the two adjacent triangular facets in a local range, the curvature difference is calculated to evaluate the geometric discontinuity between the two adjacent triangular facets, specifically:

[0063] All vertices of the triangular facets connected with any vertex in the B-Rep model are denoted as adjacent points;

[0064] Surface fitting is performed on the any vertex and all adjacent points thereof, and the curvature of the fitted surface at the any vertex is calculated;

[0065] In this embodiment, the least square method is adopted for surface fitting, wherein the least square method and the curvature calculation are known technologies, which will not be repeated here.

[0066] the difference between the curvatures of the two vertices connected by the common edge of the two adjacent triangular facets as the curvature difference of the two adjacent triangular facets;

[0067] In this embodiment, the absolute value of the difference between the curvatures of the two vertices connected by the common edge of the two adjacent triangular facets is taken as the curvature difference of the two adjacent triangular facets.

[0068] It should be noted that the greater the curvature, the greater the curvature of the fitting surface at the any vertex, the greater the curvature difference, the greater the change in curvature between the two vertices, reflecting that the two triangular facets are located in different curved surface regions with great difference in bending degree, and the common edge thereof is a strong geometric feature boundary, and the geometric discontinuity on both sides of the common edge is more obvious.

[0069] Further, based on the direction difference degree and the curvature difference, the region attribution degree is determined, specifically:

[0070] The curvature difference is normalized, and the product of the normalized result and the direction difference degree is negatively mapped as the region attribution degree of the two adjacent triangular facets;

[0071] In this embodiment, the maximum-minimum value normalization method is used for normalization, wherein the maximum-minimum value normalization method is a known technology and will not be described here. Secondly, the specific process of negative mapping is that the reciprocal of the product of the normalized result and the direction difference degree is taken as the region attribution degree of the two adjacent triangular facets, wherein, in order to avoid the denominator being 0, a preset value greater than 0 is added to the denominator when calculating the ratio, wherein the preset value greater than 0 is 0.01, and as other embodiments, the implementer can set it according to the actual situation; secondly, as other embodiments, the product of the normalized result and the direction difference degree can be denoted as the result of is taken as the region attribution degree, wherein is an exponential function with a natural constant as the base.

[0072] It should be noted that the more consistent the direction of the two adjacent triangular facets in the geometric space and the more similar the curvatures, that is, the smaller the curvature difference and the direction difference degree, the greater the region attribution degree, indicating that the two adjacent triangular facets have more smooth transition geometric features, reflecting that the two adjacent triangular facets are more likely to belong to the same smooth and continuous region.

[0073] Further, based on the region attribution degree, all triangular facets on each face in the B-Rep model are classified, specifically:

[0074] Each triangular facet on each face in the B-Rep model is abstracted as a graph node, an edge is connected between the graph nodes corresponding to two triangular facets directly adjacent to each other, and the region attribution degree of the two adjacent triangular facets is taken as the edge weight of the edge, so as to construct a corresponding undirected weighted graph of each face in the B-Rep model;

[0075] It should be noted that the construction of the weighted undirected graph is a known technology, which will not be repeated here.

[0076] The spectral clustering algorithm is used to cluster the triangular facets corresponding to all the graph nodes in the weighted undirected graph, and all the triangular facets in each clustering cluster are merged as each sub-region of each face in the B-Rep model;

[0077] It should be noted that the spectral clustering algorithm is a known technology, which will not be repeated here, and the number of clustering clusters of the spectral clustering algorithm can be determined by the elbow method.

[0078] It should be noted that the triangular facets in the sub-region are similar in geometric characteristics, and the triangular facets are smoothly transitioned, and different sub-regions can have different geometric structures and process semantics.

[0079] At this point, each sub-region of each face in the B-Rep model is obtained.

[0080] The surface evaluation module is configured to calculate a discrimination coefficient of each face by the discreteness of the bending characteristics of the vertices of the triangular facets in each face sub-region in a local range and the number of the divided sub-regions, and distinguish all the faces into process faces and auxiliary faces; evaluate the continuity of the concave-convex changes of the different sub-regions in each process face in the triangular facets on the boundary, calculate a continuity evaluation value of each process face, and obtain the structure complexity of each process face in combination with the number of the sub-regions in the process face and the discreteness of the bending characteristics.

[0081] According to the geometric characteristics of the sheet metal part, all the faces on the B-Rep model of the sheet metal part can be divided into forming surfaces and connecting surfaces. The forming surfaces are generally formed by processes such as stamping and coining, can reflect the main contour of the sheet metal part, and include process features such as holes and grooves. The connecting surfaces are usually formed by shearing processes, and serve to connect the forming surfaces of the sheet metal part.

[0082] The connecting surfaces mainly serve to connect the forming surfaces of the sheet metal part, have few process features, the triangular facets have high similarity in geometric characteristics, the number of the sub-regions obtained after the merging of the triangular facets is small, and the connecting surfaces are generally planes. The forming surfaces can reflect the main shape of the sheet metal part, are composed of a plurality of process features, the number of the sub-regions obtained after the merging of the triangular facets is large, and the forming surfaces can include complex geometric structures such as planes and cylindrical surfaces.

[0083] Based on the above analysis, the discriminant coefficient is calculated by the number of different sub-regions on each face in the B-Rep model and the bending condition of the vertices of the triangular facets in the sub-regions, specifically:

[0084] The average of the corresponding curvatures at the vertices of all triangular facets in each sub-region is denoted as the region curvature;

[0085] The dispersion degree of the region curvatures of all sub-regions of each face in the B-Rep model is calculated;

[0086] In this embodiment, the dispersion degree is measured by calculating the coefficient of variation of the region curvatures of all sub-regions of each face in the B-Rep model, wherein the calculation of the coefficient of variation is a known technique and will not be described here.

[0087] The number of all sub-regions of each face in the B-Rep model is counted, and the product of the number and the dispersion degree is taken as the discriminant coefficient of each face in the B-Rep model.

[0088] It should be noted that the greater the dispersion degree, the lower the planeness of different sub-regions in each face, and the less likely it is to be a connecting surface of the sheet metal part. The more the number, the more likely it is that the B-Rep model has more geometric structures and process semantics in each face, reflecting the higher possibility that the face belongs to a forming surface, and the greater the discriminant coefficient, the higher the process complexity of the face on the B-Rep model, and the higher the possibility that the face belongs to a forming surface.

[0089] The segmentation threshold of the discriminant coefficients of all faces in the B-Rep model is obtained, and the faces with discriminant coefficients greater than or equal to the segmentation threshold are denoted as process faces, and otherwise, as auxiliary faces.

[0090] In this embodiment, the maximum inter-class variance method is used to obtain the segmentation threshold, wherein the maximum inter-class variance method is a known technique and will not be described here.

[0091] It should be noted that the process face contains more process features, i.e., it is likely to be a forming surface of the sheet metal part, and the auxiliary face mainly serves to support and connect the process face and contains fewer process features, i.e., it is likely to be a connecting surface of the sheet metal part.

[0092] Secondly, in the process of identifying processing features by using a graph neural network, in order to adapt to the input of the graph neural network, it is necessary to map the key information in the B-Rep model to an attribute adjacency graph, and to realize the identification of processing features by graph convolution operation.

[0093] The attribute adjacency graph is defined as a tuple , is a set composed of nodes in the attribute adjacency graph, and each node corresponds to a B-Rep face, is a set composed of edges between nodes in the attribute adjacency graph, and each edge corresponds to a B-Rep edge.

[0094] A large number of process features are contained on the process surface, and the intersection relationship of the process features of the complex sheet metal part is often complex and is prone to be coupled and interfered with each other, so as to destroy the geometric information and topological information of the process features. The expression of the geometric information in the B-Rep model depends on the parameterized surface and the edge line. In the process of converting the B-Rep model into the attribute adjacency graph for feature recognition, less attention is paid to the coupled interference phenomenon between the process features on the surface. The node attribute information in the attribute adjacency graph focuses on the overall spatial information of the surface, and the attention degree to the coupled interference phenomenon of the process features on the surface is low.

[0095] Further, the step flow chart of the method for obtaining the continuity evaluation value of each process surface provided by the application is as shown in Figure 2 .

[0096] Firstly, the continuity is calculated by analyzing the concave-convex continuity of the edges of the triangular facets located at the boundaries of the sub-regions, and specifically, the continuity is calculated as follows:

[0097] For each sub-region in the B-Rep model, all the edges located on the boundaries thereof are extracted as the boundary edges;

[0098] It should be noted that the sub-region is composed of a plurality of triangular facets, and the triangular facet is composed of three edges and three vertices. Therefore, all the edges located on the boundaries of the sub-regions are the edges of the triangular facets.

[0099] For each boundary edge, the concave-convex property of the edge is judged. If the boundary edge is a convex edge, the concave-convex factor of the boundary edge is assigned as 1. If the boundary edge is a concave edge, the concave-convex factor is assigned as -1. If the boundary edge is a smooth edge, the concave-convex factor is assigned as 0.

[0100] It should be noted that the concave-convex property judgment of the edge is a known technology, and the specific method is as follows: for two planes , intersecting at the edge AB, the concave-convex judgment rule is as follows:

[0101] (1) a point C is taken on the plane , and an outward normal vector is taken at the point C. Similarly, an outward normal vector is taken at a point D on the plane .

[0102] (2) a point E is taken on the edge AB, and a vector at the point E is taken as . It is judged whether and satisfy the right-hand screw rule. If yes, they are in the same direction, otherwise, the direction of is reversed, so that and It satisfies the right-hand screw rule;

[0103] (3) and Cross product yields a vector The vector is calculated using the following formula. and The included angle ,like Values Then the common side AB is a concave side, if Values Then the common side AB is a convex side, if for Then the common edge AB is a smooth edge.

[0104] in, , It is an inverse cosine function. Represents a vector and Perform dot product. Represents the magnitude of a vector.

[0105] The concavity / convexity factors of all boundary edges on the boundaries of each sub-region are arranged in a clockwise order to form a concavity / convexity sequence;

[0106] If the concavity factor of each boundary edge in the concavity-convex sequence is the opposite of that of the next boundary edge, then the coherence weight of each boundary edge is assigned a preset first value.

[0107] If each boundary edge has the same concavity / convexity factor as the next boundary edge, then the continuity weight of each boundary edge is assigned a preset third value.

[0108] If there is exactly one 1 in the concavity / convexity factor of each boundary edge and the next boundary edge, then the continuity weight of each boundary edge is assigned a preset second value; wherein, the preset first value is less than the preset second value, and the preset second value is less than the preset third value.

[0109] In this embodiment, the first preset value is 0, the second preset value is 1, and the third preset value is 2. In other implementations, the clockwise direction can be set according to the actual situation.

[0110] It should be noted that, for ease of understanding, we assume that for the concave-convex sequence, the first... strip, If the boundary edge, the first Article and The concavity / convexity factor of the boundary edge takes the value of or , No. The continuity weight of the boundary edge is assigned to 0; if the first boundary edge... The concave-convex factor of the boundary edge of the i-th strip is The concave-convex factor of the boundary edge of the i-th strip is Or Or The concave-convex factor of the boundary edge of the i-th strip is The concave-convex factor of the boundary edge of the i-th strip is compared with the concave-convex factor of the first boundary edge in the concave-convex sequence, and the continuity weight value is obtained.

[0111] The sum of the continuity weight values of all boundary edges of each sub-region is taken as the continuity of each sub-region.

[0112] It should be noted that if the concave-convex factors of the adjacent two boundary edges are opposite numbers, that is, the continuity weight value is a preset first value, it indicates that the concave-convex of the local boundary changes suddenly, the boundary is discontinuous, if the continuity weight value is a preset second value, it indicates that the concave-convex of the boundary edge is partially continuous, which may be a smooth edge, if the continuity weight value is a preset third value, it indicates that the concave-convex of the boundary edge is completely continuous, and the boundary is smooth; therefore, the greater the obtained continuity, the more continuous and smooth the boundary of the sub-region, and the more complete the geometric features, on the contrary, it indicates that there is obvious concave-convex change in the boundary of the sub-region, and the geometric features may have discontinuity.

[0113] The sum of the continuity of all sub-regions in the process surface in the B-Rep model is taken as the continuity evaluation value of the process surface.

[0114] It should be noted that the smaller the continuity evaluation value, the more dramatic the concave-convex change of the boundary between the process features in the process surface, the more discontinuous, the sub-region boundary may correspond to the boundary of the sharp turning process feature on the process surface, the stress is more concentrated in the manufacturing process, and the process feature coupling interference phenomenon is more likely to occur, the processing difficulty is greater, and higher automatic feature recognition accuracy is required.

[0115] Further, based on the number of sub-regions in the process surface, the dispersion degree and the continuity evaluation value, the structural complexity is calculated to evaluate the geometric structural complexity of the process features in the process surface, specifically:

[0116] The number of all sub-regions in each process surface, the dispersion degree and the continuity evaluation value are combined to form a feature vector.

[0117] All components in the feature vector are normalized, and a comprehensive evaluation algorithm is used to comprehensively evaluate the feature vectors of all process surfaces to obtain the comprehensive scores of each process surface as the structural complexity of each process surface; wherein the structural complexity of each auxiliary surface is assigned as 0.

[0118] In the embodiment, the Topsis algorithm is used for comprehensive evaluation, wherein the Topsis algorithm is a known technology and will not be described herein; wherein the quantity and the dispersion degree are maximum indicators, and the continuous evaluation value is a minimum indicator.

[0119] It should be noted that the quantity evaluation process evaluates the quantity of different process features in the process surface, the dispersion degree evaluates the degree of the process surface being flat, and the continuous evaluation value evaluates the continuity of the concave-convex of the process feature boundary. The greater the structure complexity, the more complex and irregular the process surface is in geometry and manufacturing process, which means higher processing difficulty. Since the planeness of the auxiliary surface is high, the geometric structure complexity is low, and the process features in the connecting curved surface are few, the recognition difficulty is low, and therefore, the structure complexity is small.

[0120] Thus, the structure complexity of each surface is obtained.

[0121] The feature recognition and process reasoning module is used to determine the sampling accuracy of each surface through the structure complexity, to discretize the surfaces and edges in the B-Rep model, to construct an attribute adjacency graph encoding the B-Rep model information, and to process the part based on the machining feature recognition result.

[0122] Further, the expression of geometric information in the B-Rep model depends on the parameterized surface and curve, which are defined by parameterized equations. Since such original surface information cannot be directly used as the input of the neural network, it is necessary to discretize the surface representation, but this will lose fidelity and inevitably lose the key mapping of the original B-Rep entity. Therefore, the B-Rep parameterized surface and curve are directly operated, the geometric features in the parameter domain are captured as regular UV grids, and a set of local features calculated from the original surface or curve is attached to each UV grid point. Since each UV point maintains a corresponding mapping relationship with a specific position on the original B-Rep entity, more accurate surface space geometric information can be captured.

[0123] Secondly, different surfaces have different complexities, a flat surface can be described with a few UV points, while a complex and twisted surface needs more UV points to accurately express. The sampling accuracy is determined according to the geometric structure complexity of the surface. The higher the complexity, the higher the sampling accuracy required, i.e. more UV points are needed to describe complex geometric features. Therefore, the sampling accuracy of each surface is determined through the structure complexity, specifically:

[0124] If the structure complexity is distributed in , the sampling accuracy of the surface is . If the structure complexity is distributed in , the sampling accuracy of the surface is . , if the structural complexity is distributed in , the sampling accuracy of the face is , if the structural complexity is distributed in , the sampling accuracy of the face is ; wherein , , , and are preset values, and ;

[0125] In the present embodiment, is 0, is 0.25, is 0.5, is 0.75, is 1, and as other implementation manners, the implementer can set them according to actual conditions.

[0126] All the faces of the B-Rep model are converted into UV coordinates, and each face is discretized into UV grid points under the UV coordinates, wherein the step formula between the sampling points in the U direction and the V direction is respectively , , wherein , represent the maximum value and the minimum value of each face in the U direction, , represent the maximum value and the minimum value of each face in the V direction, is the sampling accuracy corresponding to each face;

[0127] Therefore, the greater the step is, the smaller the step is, and the smaller the step is, means that the sampling points are more dense, that is, the number of UV grid points is more, and the accuracy is higher.

[0128] Further, the topology and geometry information in the B-Rep model is the key information for identifying the machining feature, therefore, the accuracy of the machining feature identification is largely determined by how to utilize the B-Rep information, to extract the key attribute information and geometry information from the faces and edges of the B-Rep based on the attribute adjacency graph, form the feature descriptor, and respectively assign it to the nodes and edges in the attribute adjacency graph, in this way, the attribute adjacency graph not only captures the topology structure of the B-Rep model, but also integrates the effective B-Rep information, so as to more comprehensively describe the geometry and topology information in the B-Rep.

[0129] ​​In the UV coordinates, the 3D absolute coordinates, the 3D absolute surface normal vector and the clipping mask of each UV grid point in each face are obtained, so each face is represented by a vector containing the geometric information of ;

[0130] It should be noted that the 3D absolute coordinates refer to the position of the UV grid point in the actual three-dimensional space; the 3D absolute surface normal vector refers to a direction vector perpendicular to the face at the position of the UV grid point; and the clipping mask is used to identify whether the UV grid point is located in the visible or clipping area, for example, if there is a UV grid point inside a hole in an inner ring, it is invisible, so the clipping mask is 0, otherwise, the clipping mask is 1.

[0131] The attribute information of each face is obtained, including type, area, centroid coordinates, normal vector, ring number and bounding box ratio;

[0132] It should be noted that the type of the face can be a plane, a cylindrical surface, etc., and the normal vector is a direction vector perpendicular to the center of the face; the ring number is the sum of the inner ring and the outer ring, i.e. the outer boundary of the face, and the inner ring is a boundary ring inside the face, which usually appears in the form of a hole and internal features; the bounding box is determined by the UV coordinates of the surface to evaluate the shape characteristics of the surface, specifically: the length of the face in the U direction and the V direction is obtained, and the bounding box is the ratio of the shorter direction length to the longer direction length, so the attribute information is represented by a 10-dimensional vector.

[0133] The vectors corresponding to the geometric information and the attribute information of each face are spliced to give the attribute adjacency graph as node features;

[0134] Similarly, in the UV coordinates, each edge in the B-Rep model is discretized into a 1D grid structure with a preset value , the 3D absolute coordinates, the tangent vector and the normal vector of the point mapped to the adjacent two faces of each UV grid point on each edge are obtained, so each edge in the B-Rep model is represented by a vector containing the geometric information of .

[0135] In this embodiment, the preset value is 10, i.e. each edge in the B-Rep model is discretized into a 1D grid structure with a size of 10, and the step size is , , which represents the edge length. As other embodiments, the implementer can set it according to the actual situation.

[0136] It should be noted that the tangent vector is a unit vector in the tangent direction of the edge at the position of the UV grid point.

[0137] The attribute information of each edge in the B-Rep model is obtained, including type, length, concave-convex, closure and dihedral angle.

[0138] It should be noted that the type of edge can be a straight line, a circle, a curve, etc.; the length refers to the edge length; the concave-convex is a convex edge, a concave edge or a smooth edge; the closure is whether the starting point and the ending point coincide; the dihedral angle is the dihedral angle between the two faces connected by the edge; therefore, the attribute information is represented by a 5-dimensional vector.

[0139] The vectors corresponding to the geometric information and the attribute information of each edge in the B-Rep model are spliced to attribute adjacency graph as edge features;

[0140] Thus, the attribute adjacency graph encoding the B-Rep model information is obtained.

[0141] The attribute adjacency graph encoding the B-Rep model information is taken as the input of the graph neural network for processing feature recognition.

[0142] In this embodiment, the graph neural network model is trained using the public MFCAD++ dataset, which is generated based on the python OCC open source library and contains 24 typical machining features. The training strategy uses the Adam W optimizer, the learning rate uses the cosine annealing strategy, the loss function uses the cross entropy loss function, and the graph neural network is a known technology, which will not be described here.

[0143] It should be noted that in order to facilitate the learning of the graph neural network, the extracted geometric information and attribute information are uniformly encoded into standardized vector form using an input encoder, wherein the input encoder is composed of 4 parts, namely 2D CNN encoding UV grid information on the surface, 1D CNN encoding UV grid information on the edge, 2 multi-layer perceptron (MLP) encoding the attribute information of the surface and edge in the B-Rep model, and the input encoding information is shown in the following table:

[0144] Table 1 Input encoding information table

[0145]

[0146] Based on the recognized machining features, the pre-defined process knowledge base is matched to generate the optimal machining path and process parameters, so as to machine the part, realizing the full-process automation and intelligentization from design to manufacturing of the part.

[0147] It should be noted that the pre-defined process knowledge base contains standard definitions of various machining features, machining methods, tool selection, cutting parameters, etc.

[0148] It should be understood that although Figure 1The steps in the flowcharts are shown in sequence according to the arrows, but the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the steps are not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, Figure 1 At least one part of the steps in the flowcharts can include a plurality of sub-steps or a plurality of stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of the sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least one part of other steps or sub-steps or stages of other steps.

[0149] The technical features of the above embodiments can be combined in any manner. For brevity, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not contradict each other, they should be considered within the scope of the present disclosure.

[0150] The above embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but should not be construed as a limitation on the present application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application, without departing from the technical solution of the present application, should be considered within the protection scope of the present application.

Claims

1. An automatic identification and reasoning system for spatial geometric features and process knowledge of parts, characterized in that, The system includes: The model acquisition module acquires the B-Rep model of the part, performs meshing on it, and obtains triangular facets located in each face of the B-Rep model; The surface segmentation module analyzes the angular deviation between two adjacent triangular facets within each face, as well as the curvature difference of the vertices connected by their common edges in a local range, and calculates the region affiliation degree of two adjacent triangular facets to construct an undirected weighted graph for each face, dividing the face into multiple sub-regions. The surface evaluation module is used to calculate the discrimination coefficient of each surface by considering the discreteness of the bending characteristics of the vertices of the triangular facets within each sub-region in a local range, as well as the number of sub-regions. This distinguishes all surfaces into process surfaces and auxiliary surfaces. The module also evaluates the continuity of the concavity and convexity changes of the edges of the triangular facets located on the boundary of different sub-regions within each process surface, calculates the coherence evaluation value of each process surface, and obtains the structural complexity of each process surface by combining the number of sub-regions within the process surface and the discreteness of its bending characteristics. The feature recognition and process reasoning module is used to determine the sampling accuracy of each face by means of structural complexity, discretize the faces and edges in the B-Rep model, construct an attribute adjacency graph that encodes the information of the B-Rep model, and process the part based on its processing feature recognition results; The calculation of the region affiliation degree of two adjacent triangular facets includes: Calculate the dihedral angle between two adjacent triangular facets, and record the difference between the angle and the straight angle as the relative deviation; Calculate the similarity between the normal vectors of two adjacent triangular facets and perform a positive mapping on them; use the ratio between the relative deviation and the result of the positive mapping as the directional difference between the two adjacent triangular facets. In the B-Rep model, all vertices of the triangular facets connected to any vertex are denoted as neighboring points; a surface is fitted to the vertex and all its neighboring points, and the curvature of the fitted surface at the vertex is calculated. The difference in curvature between the two vertices connected by the common edge of the two adjacent triangular facets is taken as the curvature difference, and the curvature difference is normalized. The region affiliation degree is the result of negative mapping of the product of the normalization result and the directional difference degree; The calculation of the discriminant coefficient for each face includes: The mean curvature at the vertices of all triangular faces within each sub-region is denoted as the region curvature; the dispersion of the region curvature of all sub-regions within each face in the B-Rep model is calculated; the number of all sub-regions of each face in the B-Rep model is counted. The discrimination coefficient is the product of the quantity and the degree of dispersion.

2. The automatic identification and reasoning system for spatial geometric features and process knowledge of a part as described in claim 1, characterized in that, The construction of an undirected weighted graph for each face divides the face into multiple sub-regions, including: In the B-Rep model, each triangular facet on each face is abstracted as a graph node. An edge is connected between the graph nodes corresponding to two directly adjacent triangular facests, and the region affiliation degree of the two adjacent triangular facests is used as the edge weight to construct an undirected weighted graph. The spectral clustering algorithm is used to cluster the triangular faces corresponding to all graph nodes in the weighted undirected graph, and all triangular faces in each cluster are merged to form the sub-regions of each face in the B-Rep model.

3. The automatic identification and reasoning system for spatial geometric features and process knowledge of a part as described in claim 1, characterized in that, The step of dividing all surfaces into process surfaces and auxiliary surfaces includes: obtaining the segmentation threshold of the discrimination coefficient of all surfaces in the B-Rep model, and recording the surfaces with discrimination coefficients greater than or equal to the segmentation threshold as process surfaces, and the rest as auxiliary surfaces.

4. The automatic identification and reasoning system for spatial geometric features and process knowledge of a part as described in claim 1, characterized in that, The calculation of the coherence evaluation value for each process surface includes: For each sub-region within each process surface, extract all edges located on its boundary as boundary edges; For each boundary edge, determine the concavity / convexity of the edge. If each boundary edge is convex, the concavity / convexity factor of the boundary edge is assigned a value of 1. If each boundary edge is concave, the concavity / convexity factor is assigned a value of -1. If each boundary edge is smooth, the concavity / convexity factor is assigned a value of 0. Concavity / convexity factors of all boundary edges on the boundaries of each sub-region are arranged in clockwise order to form a concavity / convexity sequence. If the concavity / convexity factor of each boundary edge in the concavity / convexity sequence is the opposite of that of the next boundary edge, then the coherence weight of each boundary edge is assigned a preset first value. If the concavity / convexity factor of each boundary edge is the same as that of the next boundary edge, then the coherence weight of each boundary edge is assigned a preset third value. If there is exactly one 1 in the concavity / convexity factor of each boundary edge and the next boundary edge, then the coherence weight of each boundary edge is assigned a preset second value. The preset first value is less than the preset second value, and the preset second value is less than the preset third value. The coherence evaluation value is the sum of the coherence weights of all boundary edges of all sub-regions within each process surface.

5. The automatic identification and reasoning system for spatial geometric features and process knowledge of a part as described in claim 1, characterized in that, The process of obtaining the structural complexity of each process surface includes: forming a feature vector by taking the number of all sub-regions in each process surface, the degree of dispersion, and the coherence evaluation value; performing positive processing on all components in the feature vector; using a comprehensive evaluation algorithm to comprehensively evaluate the feature vectors of all process surfaces; and obtaining the comprehensive score of each process surface as the structural complexity of each process surface.

6. The automatic identification and reasoning system for spatial geometric features and process knowledge of a part as described in claim 1, characterized in that, Determining the sampling accuracy for each face includes: setting the structural complexity of the auxiliary face to a preset value; if the structural complexity of each face is distributed in... The sampling accuracy of that surface is then determined. for If the structural complexity is distributed in The sampling accuracy of that surface is then determined. for If the structural complexity is distributed in The sampling accuracy of that surface is then determined. for If the structural complexity is distributed in The sampling accuracy of that surface is then determined. for ;in, , , , as well as All are preset values, and .

7. The automatic identification and reasoning system for spatial geometric features and process knowledge of a part as described in claim 6, characterized in that, The construction of the attribute adjacency graph encoding B-Rep model information includes: extracting and encoding the geometric and attribute information of each face and each edge in the B-Rep model; concatenating the geometric and attribute information of each face and assigning it as a node feature to the attribute adjacency graph; concatenating the vectors corresponding to the geometric and attribute information of each edge and assigning them as edge features to the attribute adjacency graph; thereby obtaining the attribute adjacency graph encoding B-Rep model information.

8. The automatic identification and reasoning system for spatial geometric features and process knowledge of a part as described in claim 7, characterized in that, The extraction and encoding of geometric and attribute information for each face and edge in the B-Rep model includes: Obtain the attribute information of each face and each edge in the B-Rep model; Convert all faces of the B-Rep model to UV coordinates. In UV coordinates, based on the sampling accuracy... Discretize each surface into The 2D mesh structure is used to obtain the 3D absolute coordinates, 3D absolute surface normal vectors, and clipping masks of all UV mesh points in each face, which serve as the geometric information of each face. In UV coordinates, each edge in the B-Rep model is discretized into a 1D mesh structure with a preset size. The 3D absolute coordinates, tangent vector, and normal vector of each UV mesh point on each edge are obtained as the geometric information of each edge.

Images