A multi-stage manufacturing feature recognition method based on graph and minimum disjoint feature volume suppression

Abstract

The present invention discloses a multi-stage manufacturing feature recognition method based on graph and minimum non-intersecting feature volume suppression, which is used to automatically convert the geometric and topological information of the three-dimensional model of the part into the feature information concerned in the manufacturing process. First, the geometric and topological information of the part is extracted and represented as an attribute adjacency graph AAG. Then the AAG is decomposed into feature subgraphs and the type of the corresponding feature of the subgraph is identified using a subgraph isomorphism algorithm. Then a method for constructing and suppressing the minimum non-intersecting volume of the identified and non-intersecting features is proposed to repair the boundaries of the features on the part that are changed or even destroyed during the intersection process. The above process is repeated with the result of feature suppression as a new input part until all features are identified. The method proposed in the present invention can not only identify all features on parts with complex and intersecting features, but also show the volume shape changes of each feature in the process from the blank to the final part.

Description

Technical Field

[0001] The invention relates to manufacturing feature recognition, and more specifically to a multi-stage manufacturing feature recognition method based on graph and minimum non-overlapping feature volume suppression, and belongs to the technical field of intelligent manufacturing. Background Art

[0002] The information barrier between design and manufacturing is a key factor restricting the intelligentization of manufacturing. The design stage focuses on low-dimensional geometric elements (faces, edges, points) and their topological information, and uses these geometric elements as information carriers to specify manufacturing requirements (precision, surface roughness, etc.). The manufacturing stage focuses on high-dimensional feature information (Hole, Step, Boss, etc.), and obtains the final part shape by selecting appropriate processing operations for the manufacturing features on the part. In traditional process planning activities, process engineers need to spend a lot of time to "translate" CAD models and obtain high-dimensional manufacturing feature information from low-dimensional geometric topological information. This process relies heavily on the experience and knowledge of engineers. Therefore, how to translate the part solid model in an automated way and convert low-dimensional geometric topological information into high-dimensional manufacturing feature information has become one of the hot issues in the manufacturing field.

[0003] The invention patent with publication number CN108009527A, "An intelligent feature recognition method for STEP-NC2.5D manufacturing features", converts the geometric topological information of the part into a graph representation, and then decomposes the minimum subgraph as the input of the neural network to identify the manufacturing features. This method can accurately and efficiently identify features with complete topological boundaries on the part, but it will ignore features on the part whose boundaries are changed or even destroyed due to feature intersections. The invention patent with publication number CN107463533B, "A method for manufacturing feature recognition of a three-dimensional CAD solid model based on PCA and CNN", has identified the manufacturing features on the part by segmenting the area corresponding to the feature from the CAD solid model point cloud data as the input of the trained CNN feature recognizer. However, how to correctly segment all the features contained in the part from the three-dimensional model remains a challenge. The invention patent "Method for Identifying Intersecting Manufacturing Features of Three-Dimensional CAD Models" with publication number CN103164582A uses the 3D CAD model represented by MBD as the information input source, and uses the processing surface clustering algorithm that integrates manufacturing semantics to construct the processing area sub-graph, and then uses the processing area sub-graph as the manufacturing feature trace, combined with the annotation dimension information, to optimize and merge the processing area sub-graph, thereby realizing the identification of manufacturing features. The invention patent "A Processing Feature Identification and Modeling Method Based on MBD Model" with publication number CN105354353A extracts product manufacturing information, and in the identification of processing feature geometric attributes, one or more geometric voxel information is synthesized into processing features according to the topological relationship of geometric voxels.

[0004] In summary, the above results are of great value in realizing design and manufacturing integration and intelligent process design. However, the above methods only identify the types of features contained in the parts, and rarely consider the volume changes of the features from the blank to the final part. Secondly, the intersection of features will lead to different degrees of loss of geometric and topological elements and destruction of feature boundaries, which makes the above methods unsatisfactory in dealing with intersecting features. Summary of the invention

[0005] In view of the above shortcomings of the prior art, the purpose of the present invention is to provide a multi-stage manufacturing feature recognition method based on graph and minimum non-intersecting feature volume suppression, which can realize feature recognition in complex intersections on parts and can repair damaged feature boundaries to achieve more accurate feature recognition. At the same time, the method also shows the shape change process of the feature in the process of processing the part from the blank to the part.

[0006] The technical solution of the present invention is achieved in this way:

[0007] A multi-stage manufacturing feature recognition method based on graph and minimum disjoint feature volume suppression comprises the following steps:

[0008] 1) Extract the geometric and topological information of the part and represent it as AAG: Taking the 3D model file of the part that meets the STEP standard as input, first parse the geometric elements and their topological relationships on the part, and represent them as an attribute adjacency graph AAG based on the concavity and convexity analysis of the edge; the attribute adjacency graph AAG represents the geometric and topological information of the part with a four-tuple graph G = {V, E, T, L} that is easy to process by computers, where: V is the node set of the graph, corresponds to a face on the part; E is the set of edges on the graph, Represents node v i , v j The common edge between the corresponding part surfaces; T(v i )∈{0,1,...} represents the node v i The corresponding part surface type; L(e(v i , v j ))∈{-1,0,1} represents the upper edge of the part e(v i , v j ) is a concave edge, a neutral edge, a convex edge;

[0009] 2) Decompose AAG into feature subgraphs according to the feature boundaries: On the part, the convex edge is considered as the feature boundary. By removing all the convex edges on the part AAG, AAG is decomposed into a series of feature subgraphs;

[0010] 3) Use the subgraph isomorphism algorithm to identify the feature type corresponding to each feature subgraph: Convert the feature identification problem into a subgraph isomorphism problem, and design a subgraph isomorphism algorithm to identify the type of features contained in each feature subgraph obtained by decomposing the AAG;

[0011] 4) Repair the boundaries of intersecting features: First, a method for constructing the minimum non-intersecting volume (MNV) of identified features is proposed; on this basis, the MNV of identified and non-intersecting features is suppressed on the part to repair the feature boundaries that have been changed or even destroyed during the intersection process; the unsuppressed features will be retained on the part solid model;

[0012] 5) Use the part solid model with unsuppressed features retained in step 4) as a new part input, and repeat steps 1) to 4) until all features on the part are identified.

[0013] Preferably, in step 1), the method for analyzing the concavity and convexity of the edge is as follows:

[0014] Given faces f1, f2 and the edge l between them, P mid (x0, y0, z0) is the midpoint of edge l, faces f1 and f2 are at point P mid The normal vectors n1 and n2 at are uniquely determined;

[0015] If n1 and n2 are collinear, it means that f1 and f2 are smoothly transitioned at edge l, and edge l is a neutral edge. l =0;

[0016] Otherwise, at n1 and n2 and point P mid The plane formed is constructed with r as radius and P as center. mid The auxiliary circle O of the surface f1 and the point p2 are the intersection points of the auxiliary circle O and the surface f1 and the surface f2 respectively. x It is the midpoint of the line segment formed by connecting points p1 and p2;

[0017] If you click p x Inside the part solid model, the edge l is a convex edge, label l =1;

[0018] If you click p x Outside the part solid model, the edge l is a concave edge, label l =-1.

[0019] Furthermore, the radius r of the auxiliary circle O is determined by the following formula, where tol represents the accuracy tolerance of the input model, argmin{|l1|,|l2|,...,|l n |} represents the length of the shortest side of the part;

[0020] tol<r<<argmin{|l1|, |l2|,...,|l n |}.

[0021] The subgraph isomorphism in step 3) of the present invention means that, given a data graph g=<V,E,T,L> and query graph q =<V′,E′,T′,L′> , if there exists an injective function f:V→V′ that satisfies (a) and (b), then the subgraph sub(g) of g is said to be isomorphic to q, denoted by

[0022] (a) Satisfies T(u)=T′(f(u))

[0023] (b) Satisfies L(e(u1,u2))=L′(e(f(u1),f(u2)))

[0024] The subgraph isomorphism algorithm is used to identify the feature type corresponding to the feature subgraph. The specific method is to use the feature subgraph obtained by decomposing the attribute adjacency graph AAG as the data graph set G = {g1, g2, ..., g m}, the predefined feature template is used as the query graph set Q = {q1, q2, ..., q n}; for each g i ∈G to find a q j ∈Q satisfies Then it is considered that the decomposed feature subgraph g i The corresponding feature type is q j .

[0025] The minimum non-intersecting volume MNV of the feature in step 4) is constructed according to the following steps:

[0026] Step 1: Construct an approximate feature volume: For features with a base surface, the base surface is stretched along the base surface normal vector to construct an approximate feature volume; for features without a base surface, a virtual base surface is constructed perpendicular to the axial direction of the feature component surface, and the approximate feature volume is constructed by bidirectional stretching along the axial direction of the component surface;

[0027] Step 2: Remove the intersection of the approximate feature volume and the part entity through Boolean operation;

[0028] Step 3: Extend the adjacent surface of the feature and divide the approximate feature volume obtained by Boolean operation in Step 2 into a series of feature sub-volumes;

[0029] Step 4: Generate MNV: For any feature subvolume obtained in Step 3, if its surface overlaps with the component surfaces of the feature, then the feature subvolume is part of the minimum non-intersecting volume of the feature; perform Boolean union operation on all feature subvolumes that can constitute MNV to obtain the final minimum non-intersecting volume MNV.

[0030] In step 4) of the present invention, when suppressing features, intersecting features cannot be suppressed at the same time; in each identified feature, the intersection relationship is first determined based on the component faces and adjacent faces of the feature, and then the most non-intersecting features are selected for suppression.

[0031] Specifically, in step 2), the attribute adjacency graph AAG is decomposed into feature subgraphs as follows to construct the vertex convex connection matrix P on the part: n×n , where n represents the number of vertices on the part; the element p in the matrix P is ij = 1 means there is a convex edge between vertices i and j, otherwise p ij =0; for each row i of the matrix P, if This means that the convex edge with point i as the endpoint cannot form a convex ring; remove the convex edge in the matrix P that satisfies The rows and columns where the vertices are located are simplified to P′, thereby filtering out non-boundary convex edges; the convex edges corresponding to the non-zero elements in the matrix P′ on the AAG are removed, and the AAG can be decomposed into characteristic subgraphs.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. The present invention utilizes a multi-stage feature recognition and suppression method to achieve feature recognition in complex intersection situations on parts.

[0034] 2. In the process of analyzing the geometry and topology information of parts, the present invention designs a method for judging the convexity of edges based on geometric reasoning, realizes the judgment of the concavity of adjacent edges between surfaces, and on this basis represents the geometry and topology information of parts in the form of AAG.

[0035] 3. The present invention proposes a method for constructing and suppressing the minimum non-intersecting feature volume (MNV) to repair the feature boundary that has been changed or destroyed during the intersection process. At the same time, the geometric information and topological information of the feature can be associated with the feature volume information that is more concerned in the manufacturing process, which can provide a useful reference for the processing of parts.

[0036] 4. The intersecting features on the parts of the present invention are identified and suppressed in multiple stages, demonstrating the reverse processing process of the parts from the blank to the final product. It can be used to model the shape changes of parts in each processing step in the digital twin of the machining industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the characteristic information of the example part;

[0038] Figure 2 It is a schematic diagram of the minimum disjoint feature volume and feature suppression of the present invention;

[0039] Figure 3 The overall flow chart of feature identification and suppression of the present invention is as follows;

[0040] Figure 4 The present invention parses the part STEP file to obtain a schematic diagram of geometric and topological information;

[0041] Figure 5 The figure is a schematic diagram of a new method for judging the concavity and convexity of a part model;

[0042] Figure 6 AAG schematic diagram of an example part of the present invention;

[0043] Figure 7 for Figure 6 Schematic diagram of AAG decomposition results of example parts;

[0044] Figure 8 It is a schematic diagram of the characteristic template of the present invention;

[0045] Fig. 9 This is a schematic diagram of an example of subgraph isomorphism of the present invention;

[0046] Fig.10 A schematic diagram for constructing an approximate characteristic volume of the present invention;

[0047] Fig.11 A schematic diagram showing the intersection of the approximate feature volume and the part entity;

[0048] Fig.12 This is a schematic diagram of the characteristic minimum non-intersecting volume of the present invention;

[0049] Fig.13 This is a schematic diagram of the four-stage feature recognition and suppression process for an example part;

[0050] Fig.14 The AAG of the first stage of the example part and the schematic diagram of the decomposed sub-graphs;

[0051] Fig.15 This is an example of a part feature F of the present invention. 4-1 Schematic diagram of the boundary repair process. DETAILED DESCRIPTION

[0052] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] Before introducing the present method, in order to better understand the present invention, several important concepts are first introduced and defined, and then the implementation process of the present method is described in detail.

[0054] Introduction to related concepts

[0055] Concept A. Faces of feature (FF) refers to the collection of faces that make up a feature. A manufacturing feature on a part is formed by several adjacent faces, such as Figure 1 The faces {f0, f1} in the figure form the hole feature F1, so the component faces FF1 of feature F1 = {f0, f1}. It should be noted that in STEP files, symmetrical faces such as cylindrical faces and conical faces are usually represented as two semi-cylindrical faces or two semi-conical faces. Figure 1 The cylindrical surface in the part is composed of two symmetrical semi-cylindrical surfaces f0 and f1.

[0056] Concept B: Neighbor faces of feature (NF) refers to the set of faces on a part that are adjacent to the component faces of a feature. Figure 1 The adjacent face set NF1 of the hole feature F1 of the part is {f2, f6, f7, f8, f9, f 10 , f 12 , f 13 , f 14}.

[0057] The component faces and adjacent faces of a feature not only express the geometric and topological information of the feature, but also serve as the basis for determining whether the features intersect. a and F b ,like Then F a and F b Do not intersect; otherwise F a Intersects at F b .For example Figure 1 The part in (containing three manufacturing features - hole F1, slot F2 and slot F3) has Therefore, hole F1 and slot F2 intersect; Therefore, groove F2 and groove F3 do not intersect.

[0058] Concept C, Base face of feature: The base face of feature is defined as the face of the feature that is perpendicular to the tool axis direction (TAD) and is connected to the other faces of the feature by concave edges.

[0059] Concept D, Minimum non-intersecting feature volume (MNV): refers to the volume of the part of a feature that does not intersect with other features. a With features {F1, F2, ..., F l} intersect respectively, then the feature F a Minimum disjoint volume MNV a is defined as:

[0060]

[0061] Where S i Representation feature F i The characteristic body of , “-” represents the Boolean subtraction operation, “∑” represents the Boolean union operation, and “∩” represents the Boolean intersection operation.

[0062] Figure 1 The intersection of the three features on the part is as follows Figure 2 As shown in (a), two slot features (F2 and F3) share the feature volume (S2). This is because they are essentially the same slot, but are split into two slots during the feature intersection process. Hole F1 intersects with slots F2 and F3, and the common volume of their intersection is like Figure 2 (b) As shown. Therefore, the minimum non-intersecting characteristic volume of hole F1 like Figure 2 (c) as shown.

[0063] Concept E, feature suppression. Given feature F on part P i , feature suppression refers to the three-dimensional solid S of the part P and feature F i The minimum disjoint volume MNV i The Boolean operation between P ∪MNV i .

[0064] The purpose of feature suppression is to repair the feature boundaries destroyed during feature intersection by suppressing the MNV of features existing at the boundaries. Figure 2 (d) shows the feature suppression result of the hole feature F1. After feature suppression, the boundaries of the slots F2 and F3 that were destroyed in the feature intersection are repaired.

[0065] In order to solve the problem that the geometric and topological boundaries are lost in the feature intersection, resulting in the unrecognizable features, the present invention proposes an intersection feature recognition method based on graph and minimum non-intersecting feature body suppression. Figure 3 The overall process of feature recognition and suppression is given, including the following steps:

[0066] 1) Extract the geometric and topological information of the part to construct the attribute adjacency graph AAG: Taking the part 3D model file that meets the STEP standard as input, first parse the geometric elements and their topological relationships on the part, and represent them as the attribute adjacency graph AAG based on the convexity analysis of the edge.

[0067] 2) Decompose AAG into feature subgraphs: On the part, the convex edges are considered to be the boundaries of the features. By removing all the convex edges on the part AAG, the AAG is decomposed into a series of feature subgraphs.

[0068] 3) Use the subgraph isomorphism algorithm to identify the type of features corresponding to each feature subgraph: Convert the feature recognition problem into a subgraph isomorphism problem, and design a subgraph isomorphism algorithm to identify the type of features contained in each feature subgraph obtained by decomposing AAG.

[0069] 4) Repair the boundaries of intersecting features: First, a method to construct the minimum non-intersecting volume (MNV) of the identified features is proposed. On this basis, the MNV of the identified and mutually non-intersecting features is suppressed on the part to repair the feature boundaries that have been changed or even destroyed during the intersection process. The unsuppressed features will be retained on the solid model and recognized in the subsequent stage.

[0070] The solid model of the part after feature suppression at each stage will be input as a new part and the above four steps will be repeated until all features on the part are recognized.

[0071] The above four steps are described in detail below.

[0072] 1. Extract the geometric and topological information of the part and represent it as AAG

[0073] One of the keys to manufacturing feature recognition is to represent the geometric and topological information of parts in a form that is easy for computers to process. Thanks to the flexibility and uniqueness of graph representation, in the present invention, the geometric and topological information of parts is represented in the form of a quadruple graph G = {V, E, T, L}, where V is the node set of the graph, corresponds to a face on the part; E is the set of edges on the graph, Represents node v i , v j The common edge between the corresponding part surfaces; T(v i )∈{0,1,...} represents the node v i The corresponding part surface type; L(e(v i , v j ))∈{-1,0,1} represents the upper edge of the part e(v i , v j ) is a concave edge, a neutral edge, and a convex edge.

[0074] The STEP standard is a neutral file that does not rely on a specific system to represent a 3D model, and it concisely and completely defines geometric elements (faces, edges, points, etc.) and their topological relationships. In the present invention, the STEP neutral file of the part model is used as input, and it can be parsed in plain text through the API of common 3D software or according to the tree structure inside the file. The parsed geometric elements and their topological relationships are as follows: Figure 4 As shown in the figure, each face contains several edges, and each edge contains two points. If two faces share an edge, they are adjacent. The type attribute of the face records the type of the face (plane, cylinder, cone), which can be obtained directly from the STEP file. The label attribute of the edge records the convexity of the edge, which needs to be obtained through further geometric analysis.

[0075] The present invention proposes a method for judging the concavity and convexity of adjacent edges between curved surfaces, such as Figure 5 Given faces f1, f2 and the edge l between them, P mid (x0, y0, z0) is the midpoint of edge l, faces f1 and f2 are at point P mid The normal vectors n1 and n2 at are uniquely determined.

[0076] ·like Figure 5 As shown in (a), if n1 and n2 are collinear, it means that f1 and f2 are smoothly transitioned (tangent) at edge l, and edge l is a neutral edge label l =0.

[0077] Otherwise, at n1 and n2 and point P mid The plane formed is constructed with r as radius and P as center. mid The auxiliary circle O. Points p1 and p2 are the intersection points of O with the surface f1 and the surface f2 respectively. x It is the midpoint of the line segment formed by connecting points p1 and p2.

[0078] If you click p x Inside the part solid model, such as Figure 5 (b) shows that edge l is a convex edge label l =1.

[0079] If you click p x Outside the part solid model, such as Figure 5 (c) shows that edge 1 is a concave edge label l =-1.

[0080] In order to ensure the existence of the intersection points p1 and p2 of the auxiliary circle O and the surface, and to ensure that the point p xThe distance to the part entity is greater than the minimum modeling software accuracy tol. Formula (1) gives the range of the radius r of the auxiliary circle O, where tol represents the accuracy of the input model, argmin{|l1|,|l2|,...,|l n |} represents the length of the shortest side of the part.

[0081] tol<r<<argmin{|l1l, |l2|,...,|l n |} (1)

[0082] Based on the judgment of the concavity and convexity of the edge, the faces and edges of the part are represented as nodes and arcs of the graph respectively, and the attribute adjacency graph AAG of the part is obtained. Figure 6 The AAG of an example part is shown, where different types of nodes represent the types of corresponding faces, and different types of arcs represent the concavity of corresponding edges.

[0083] 2. Decompose AAG into feature subgraphs according to feature boundaries

[0084] In general machining processes, the features on a part are obtained by removing a certain volume of material from the part blank. The convex edge left on the part during the material removal process is the boundary of the feature. Therefore, the part attribute adjacency graph AAG can be decomposed into feature subgraphs by removing the convex edges on the AAG. Specifically, the convex connection matrix P of the vertices on the part is constructed. n×n , where n represents the number of vertices on the part; the element p in the matrix P is ij = 1 means there is a convex edge between vertices i and j, otherwise p ij =0; for each row i of the matrix P, if This means that the convex edge with point i as the endpoint cannot form a convex ring; remove the convex edge in the matrix P that satisfies The rows and columns where the vertices are located are simplified to P′, thereby filtering out non-boundary convex edges; the convex edges corresponding to the non-zero elements in the matrix P′ on the AAG are removed, and the AAG can be decomposed into characteristic subgraphs. Figure 6 The decomposition results of AAG in Figure 7 shown.

[0085] 3. Use the subgraph isomorphism algorithm to identify the type of features corresponding to each feature subgraph

[0086] Refer to the manufacturing feature types defined in the STEP standard. Figure 8The graphic representations of 10 common milling manufacturing features are given, including slots, through holes, blind holes, circular slots, steps, bosses, tapered holes, and compound tapered holes. It should be noted that the feature recognition framework proposed in the present invention is independent of the manufacturing feature template library. As long as the uniqueness of the manufacturing feature graphic representation template is guaranteed, the features in the manufacturing feature template library can be freely added or modified without re-modifying the method. In addition to defining the feature template, the base surface of the feature is also specified in the template. For example Figure 8 The slots, blind holes, steps, bosses, closed cavities, and open slots in the figure are base features, and their base surface is f0. Through holes, circular slots, tapered holes, and compound tapered holes are baseless features.

[0087] In addition, in actual production, the type and number of the remaining faces of the open slot feature except the base face are not fixed. Therefore, the template of the closed cavity can be constructed as a closed-loop graph structure with a base face node (f0). The type of the base face node is a plane, which is connected to other nodes through concave edges. A label x that can match any type is set for the remaining nodes other than the base face. The template of the open slot feature is constructed as a chain graph structure with a base face node (f0). The type of the base face node is a plane, which is connected to other nodes through concave edges. Similarly, a label x that can match any type is set for the remaining nodes other than the base face.

[0088] Based on the constructed feature graph representation template, the subgraph isomorphism algorithm can be used to identify the type of features corresponding to each feature subgraph obtained by decomposing AAG. The concept F below gives the definition of subgraph isomorphism, and Table.1 summarizes the commonly used symbols.

[0089] Table 1 Definition of common symbols

[0090]

[0091] The feature matching process identifies the feature type corresponding to the feature subgraph obtained by AAG decomposition through subgraph isomorphism. Fig. 9 In the data graph g, the query graph q has an isomorphic mapping of {(u0, v0), (u1, v3), (u2, v4), (u3, v 10 )}. Subgraph isomorphism for feature recognition uses the feature subgraphs obtained by decomposing AAG as the data graph set G = {g1, g2, ..., g m}, the feature templates predefined above are used as the query graph set Q = {q1, q2, ..., q n}. The feature matching process is to i ∈G, find a q j ∈Q satisfies

[0092] Concept F: Subgraph isomorphism: Given a data graph g =<V,E,T,L> and query graph q =<V′,E′,T′,L′> , if there exists an injective function f:V→V′ that satisfies (a) and (b), then the subgraph sub(g) of g is said to be isomorphic to q, denoted by

[0093] (a) Satisfies T(u)=T′(f(u))

[0094] (b) Satisfies L(e(u1,u2))=L′(e(f(u1),f(u2)))

[0095] Algorithm 1 provides a general framework for subgraph matching algorithms based on backtracking search. i ∈G,q j ∈Q, if it satisfies Then the matching mapping M from q to g is output. Line 1 indicates that for each node u∈V(q) in the query graph, its candidate matching node set C(u) is generated, and C(u) is defined as the set of all nodes in the data graph g that match the node u. Line 2 indicates setting a matching order for all nodes in the query graph q according to a certain matching order Δ. Line 3 indicates recursively searching the solution space starting from i=1 given q, g, C, Δ, and recording the matching results in M. If all nodes in the query graph q are matched (i=Δ+1), M is returned in Line 4. Line 5 indicates selecting the next matching node u given Δ. Line 6 calculates the local candidate vertex set LC (local candidate vertex set) of node u through formula (4), where In the given matching order Δ, N(u) represents the neighbor node after node u, and M[u′] represents the node in g that matches node u′. Line 7-11 recursively calls the Enumerate function to enumerate the entire search space.

[0096]

[0097]

[0098] 4. Repair the boundaries of intersecting features

[0099] The previous steps can identify features with complete boundaries on the part, but cannot identify features with broken boundaries in feature intersections. Therefore, the present invention proposes a method for constructing and suppressing the minimum non-intersecting volume (MNV) of identified and non-intersecting features to repair the boundaries of intersecting features on the part. The MNV of the identified features can be generated as follows:

[0100] Step 1: Generate an approximate feature volume by stretching the feature base surface (virtual base surface). Fig.10 As shown in (a), stretching the feature base surface along the normal vector direction of the feature base surface can generate an approximate feature volume. In order to ensure that the generated approximate volume can completely contain the MNV of the feature, the stretching distance L is specified as max{x max -x min ,y max -y min , z max -z min}, where (x max ,y max , z max ) and (x min ,y min , z min ) are the coordinate values ​​of the maximum and minimum vertices of the bounding box of the part. For features without base surfaces (circular holes, circular grooves, tapered holes, compound tapered holes, etc.), it is necessary to first construct a virtual base surface perpendicular to the axis of the cylindrical surface (conical surface). The diameter of the base surface is the maximum diameter of the cylindrical surface (conical surface) as shown in the figure. Fig.10 Then, by biaxially stretching the virtual base surface along the axial direction of the cylindrical surface (conical surface) of the feature, the approximate feature volume of the feature can be obtained.

[0101] Step 2: Use Boolean operations to remove the intersection of the approximate feature volume and the part entity. The approximate feature volume constructed in the previous step may have intersections with the part entity model. For example, Fig.10 The approximate volume of the tapered hole constructed in (b) is a cylinder rather than a cone. Therefore, it is necessary to use Boolean operations to remove the intersection of the approximate feature volume and the part solid model. Let V part and V f are the part volume and the generated approximate feature volume respectively. Then the volume after removing the intersection is V f -(V part ∩V f ), where “-” and “∩” represent Boolean difference and Boolean intersection operations respectively. Fig.11 A cross-section view of a tapered hole feature showing the above process.

[0102] Step 3: Extend the adjacent faces of the feature and divide the approximate feature volume into a series of feature sub-volumes. Extend all adjacent faces of the feature and divide the approximate feature volume into a series of sub-feature volumes (denoted as Sub). Fig.12 As shown, the approximate volume of a through hole feature is divided into 12 sub-features by adjacent faces.

[0103] Step 4: Generate MNV. For the sub-feature volume obtained in the previous step, if its surface overlaps with the component surface of the feature, then the sub-feature volume is part of the minimum non-intersecting volume of the feature. Perform a Boolean operation on all sub-feature volumes that can constitute MNV to obtain the final MNV. Fig.12 The cylindrical surfaces of sv4, sv6, sv7, sv8 and sv9 in the 12 sub-volumes divided from the feature volume of the through-hole feature F1 overlap with the component surfaces f0 and f1 of the feature F1. Therefore, the minimum non-intersecting volume of the feature F1 is MNV1=sv4∪sv6∪sv7∪sv8∪sv9.

[0104] Since MNV that suppresses two intersecting features at the same time will ignore the common volume of the intersecting parts, it is stipulated that intersecting features cannot be suppressed at the same time. In each identified feature, the intersection relationship is first determined based on the component faces and adjacent faces of the feature, and then the most non-intersecting features are selected for suppression. The parts after feature suppression at each stage will be used as new input for the next stage until all features on the part are identified.

[0105] The effectiveness of the method proposed in the present invention is verified by using an industrial mold base part including 18 features such as grooves, bosses and through holes.

[0106] Based on the method proposed in this invention, the open source geometric modeling kernel OpenCasCade is selected to implement Boolean operations in the feature suppression process and the position relationship judgment between geometric elements (points, surfaces, and bodies). The subgraph isomorphism algorithm and the visualization of graph data are implemented by selecting the networkx tool. In the edge concavity judgment, the radius of the auxiliary circle is set to r = argmin{|l1|,|l2|,...,|l n |} / 100, where argmin{|l1|,|l2|,...,|l n |} is the length of the shortest side of the part.

[0107] Initial parts such as Fig.13 As shown above in the first stage, an AAG with 78 nodes and 214 edges is constructed. Fig.14 As shown in the figure. In the first stage, AAG is decomposed into 17 subgraphs. Table 2 records the types of features corresponding to the 17 subgraphs identified by the subgraph isomorphism algorithm. Afterwards, feature F is constructed. 1-1 、F1-2 、F 1-3 、F 1-4 、F 1-5 、F 1-7 、F 1-8 、F 1-11 、F 1-12 、F 1-14 、F 1-15 、F 1-16 and F 1-17 The minimum non-intersecting volume (MNV) of the feature F in Table 2 1-6 , F 1-9 , F 1-10 , F 1-13 is retained on the solid model. The result of the first stage feature suppression is as follows Fig.13 shown.

[0108] Table.2 Features identified in the first stage

[0109]

[0110] Except for the first stage, each subsequent stage uses the result of feature suppression in the previous stage as a new input. The feature recognition results of each stage are listed in Table 3, Table 4, and Table 5. After four stages of recognition and suppression, all 18 features on the part are recognized and suppressed. Finally, a part blank without any features is obtained, as shown in Fig.13 It is worth noting that only 17 feature subgraphs were obtained after the AAG decomposition constructed in the first stage, and 18 features were identified after 4 stages. This is because the feature F identified in the last stage 4-1 In with F 1-4 、F 1-16 、F 2-2 、F 2-3 and F 3-1 The boundary is destroyed during the intersection process, but the three-stage feature suppression repairs the F 4-1 The boundaries, such as Fig.15 Finally, in the final stage, F 4-1 .

[0111] Table.3 Features identified in the second stage

[0112]

[0113] Table 4 Features identified in the third stage

[0114]

[0115] Table.5 Features identified in the fourth stage

[0116]

[0117] The manufacturing feature recognition method proposed in the present invention realizes feature recognition in complex intersection situations by using a multi-stage feature recognition and suppression method. In the process of parsing the geometry and topology information of the part, a method for judging the convexity of the edge based on geometric reasoning is designed to realize the judgment of the convexity of the adjacent edges between the surfaces, and on this basis, the geometry and topology information of the part is represented in the form of AAG. Considering that the geometric topology information is lost due to excessive decomposition in the traditional AAG decomposition process, the convex edges of the non-feature boundaries are first filtered before the AAG decomposition to ensure the integrity of each feature subgraph information in the AAG decomposition process. For the situation where the geometric topology information of the features on the part is lost and the boundaries are destroyed during the intersection process, the destroyed feature boundaries are gradually repaired by constructing and suppressing the MNV of the features with complete boundaries on the part. In addition, the multi-stage feature recognition and suppression of the part shows the reverse processing process from the blank to the final part. The minimum non-intersecting volume of the suppressed features in each stage is actually the volume of the material removed during the processing of the feature, which provides an important reference for process planning activities and can be used to model the shape change process of each processing process part in the machining digital twin process.

Claims

1. A multi-stage manufacturing feature recognition method based on graph and minimum disjoint feature volume suppression, characterized in that: The steps include: 1) Extract the geometric and topological information of the part and represent it as AAG: Taking the 3D model file of the part that meets the STEP standard as input, first parse the geometric elements and their topological relationships on the part, and represent them as an attributed adjacency graph AAG based on the concavity and convexity analysis of the edge; the attributed adjacency graph AAG represents the geometric and topological information of the part with a four-tuple graph G = {V, E, T, L} that is easy to process by computers, where: V is the node set of the graph, corresponds to a face on the part; E is the set of edges on the graph, Represents node v i ,v j The common edge between the corresponding part surfaces; T(v i )∈{0,1,…} represents the node v i The corresponding part surface type; L(e(v i ,v j ))∈{-1,0,1} represents the upper edge of the part e(v i ,v j ) is a concave edge, a neutral edge, a convex edge; 2) Decompose AAG into feature subgraphs according to the feature boundaries: On the part, the convex edge is considered as the feature boundary. By removing all the convex edges on the part AAG, AAG is decomposed into a series of feature subgraphs; 3) Use the subgraph isomorphism algorithm to identify the feature type corresponding to each feature subgraph: Convert the feature identification problem into a subgraph isomorphism problem, and design a subgraph isomorphism algorithm to identify the type of features contained in each feature subgraph obtained by decomposing the AAG; 4) Repair the boundaries of intersecting features: First, a method for constructing the minimum non-intersecting volume (MNV) of identified features is proposed; on this basis, the MNV of identified and non-intersecting features is suppressed on the part to repair the feature boundaries that have been changed or even destroyed during the intersection process; the unsuppressed features will be retained on the part solid model; 5) The solid model of the part with unsuppressed features retained in step 4) is used as a new part input, and steps 1) to 4) are repeated until all features on the part are identified; The minimum non-intersecting volume MNV of the feature in step 4) is constructed as follows: Step 1: Construct an approximate feature volume: For features with a base surface, the base surface is stretched along the base surface normal vector to construct an approximate feature volume; for features without a base surface, a virtual base surface is constructed perpendicular to the axial direction of the feature component surface, and the approximate feature volume is constructed by bidirectional stretching along the axial direction of the component surface; Step 2: Remove the intersection of the approximate feature volume and the part entity through Boolean operation; Step 3: Extend the adjacent surface of the feature and divide the approximate feature volume obtained by Boolean operation in Step 2 into a series of feature sub-volumes; Step 4: Generate MNV: For any feature subvolume obtained in Step 3, if its surface overlaps with the component surfaces of the feature, then the feature subvolume is part of the minimum non-intersecting volume of the feature; perform Boolean union operation on all feature subvolumes that can constitute MNV to obtain the final minimum non-intersecting volume MNV.

2. The multi-stage manufacturing feature recognition method based on graph and minimum disjoint feature volume suppression according to claim 1, characterized in that: In step 1), the edge concavity analysis method is as follows: Given faces f1, f2 and the edge l between them, P mid (x0, y0, z0) is the midpoint of edge l, faces f1 and f2 are at point P mid The normal vectors n1 and n2 at are uniquely determined; If n1 and n2 are collinear, it means that f1 and f2 are smoothly transitioned at edge l, and edge l is a neutral edge. l =0; Otherwise, at n1 and n2 and point P mid The plane formed is constructed with r as radius and P as center. mid The auxiliary circle O of the surface f1 and the point p2 are the intersection points of the auxiliary circle O and the surface f1 and the surface f2 respectively. x It is the midpoint of the line segment formed by connecting points p1 and p2; If you click p x Inside the part solid model, the edge l is a convex edge, label l =1; If you click p x Outside the part solid model, the edge l is a concave edge, label l =-1.

3. The multi-stage manufacturing feature recognition method based on graph and minimum disjoint feature volume suppression according to claim 2, characterized in that: The radius r of the auxiliary circle O is determined by the following formula: tol represents the accuracy tolerance of the input model, argmin{|l1|,|l2|,…,|l n |} represents the length of the shortest side of the part; toll <r<<argmin{|l1|,|l2|,…,|l n |}。 4. The multi-stage manufacturing feature recognition method based on graph and minimum disjoint feature volume suppression according to claim 1, characterized in that: The subgraph isomorphism in step 3) means that, given a data graph g =<V,E,T,L> and query graph q =<V′,E′,T',L′> , if there exists an injective function f:V→V′ that satisfies (a) and (b), then the subgraph sub(g) of g is said to be isomorphic to q, denoted by (a) Satisfies T(u)=T′(f(u)) (b) Satisfies L(e(u1,u2))=L′(e(f(u1),f(u2))) The subgraph isomorphism algorithm is used to identify the feature type corresponding to the feature subgraph. The specific method is to use the feature subgraph obtained by decomposing the attribute adjacency graph AAG as the data graph set G = {g1, g2, …, g m }, the predefined feature template is used as the query graph set Q = {q1,q2,…,q n }; for each g i ∈G to find a q j ∈Q satisfies It is considered that the decomposed feature subgraph g i The corresponding feature type is q j .

5. The multi-stage manufacturing feature recognition method based on graph and minimum disjoint feature volume suppression according to claim 1, characterized in that: Step 4) When suppressing features, intersecting features cannot be suppressed at the same time; in each identified feature, the intersection relationship is first determined based on the component faces and adjacent faces of the feature, and then the most non-intersecting features are selected for suppression.

6. The multi-stage manufacturing feature recognition method based on graph and minimum disjoint feature volume suppression according to claim 1, characterized in that: In step 2), the attribute adjacency graph AAG is decomposed into feature subgraphs as follows to construct the vertex convex connection matrix P on the part: n×n , where n represents the number of vertices on the part; the element p in the matrix P is ij = 1 means there is a convex edge between vertices i and j, otherwise p ij =0; for each row i of the matrix P, if This means that the convex edge with point i as the endpoint cannot form a convex ring; remove the convex edge in the matrix P that satisfies The rows and columns where the vertices are located are used to simplify the matrix P to P′, thereby filtering out non-boundary convex edges; By removing the convex edges corresponding to the non-zero elements in the matrix P′ on the AAG, the AAG can be decomposed into characteristic subgraphs.

Images