A feature recognition method for steam turbine blades based on B-rep

Abstract

The present invention discloses a feature recognition method based on B-rep steam turbine blades. By extracting attributes from a B-rep (boundary representation) steam turbine blade model, different attributes such as faces, edges, and connection modes are assigned identifiers to construct an extended attribute adjacency graph of the blade. The extended attribute adjacency graph is then trimmed by suppressing invalid features. The VF2p algorithm is then used to calculate and sort the priority access nodes of the target feature subgraph of the knowledge base, and the target feature subgraph is reconstructed into a forest structure according to the attribute identifiers. Finally, a pre-explored extended attribute adjacency graph search is performed using a depth-first search method to identify the features and types of the blades. The present invention realizes the recognition of steam turbine blade features with efficient recognition speed, laying the foundation for the comprehensive digitalization of the steam turbine blade manufacturing field.

Description

Technical Field

[0001] The present invention relates to technologies in the field of auxiliary design and manufacturing, and in particular to a method for identifying features of steam turbine blades based on B-rep. Background Art

[0002] Blades, as key components of steam turbine generators and turbine engines, are widely used in aerospace, aviation, automotive, and energy applications. As a critical component of steam turbines, blade production processes, manufacturing precision, and processing efficiency directly impact the quality of finished turbines and the company's economic benefits. Currently, blade manufacturers typically utilize the basic functions of 3D CAD / CAM software for 3D modeling, process route determination, process model generation, and NC programming. However, blades are produced in small batches, with multiple specifications and in series. Even for products within the same series, the same NC machining process still requires the same preparatory steps, resulting in significant duplication of effort and significantly hindering efficiency. Therefore, fully digitizing the blade production process is crucial for improving blade production efficiency. Currently, the most significant challenge facing full-process digital manufacturing with NC processes is the inability to automatically extract model data. However, identifying and extracting features and types from 3D models is fundamental to fully digital manufacturing. Currently, most feature recognition research in universities is based on MBD, but its practical application in actual production and manufacturing is limited. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides a feature recognition method for turbine blades based on B-rep. The present invention classifies the turbine blades by extracting their attributes, constructs an extended attribute adjacency graph using the attributes, optimizes the target subgraph in the knowledge base using the VF2p algorithm, and performs full-graph retrieval through a depth-first search with pre-exploration to identify the characteristics and types of the blades.

[0004] The present invention achieves the above technical objectives through the following technical means.

[0005] A feature recognition method for steam turbine blades based on B-rep comprises the following steps:

[0006] Step 1: Extract attributes of the turbine blade model constructed using the B-rep (boundary representation) method, traverse all engineering surfaces of the entire model, extract the attributes of the engineering surfaces and the attributes of the edges connected to the surfaces;

[0007] Step 2: Mathematically represent the surface attributes, edge attributes, and connection methods of different engineering surfaces, construct surface nodes and edge nodes, and then combine them into the extended attribute adjacency graph G2 of the leaf;

[0008] Step 3: trim the extended attribute adjacency graph G2 by suppressing invalid features to reduce the amount of information in the target graph;

[0009] Step 4: Use the VF2p algorithm to calculate and sort the priority nodes of the target feature subgraph G1 of the knowledge base according to the primary and secondary features;

[0010] Step 5: Reconstruct the target feature subgraph G1 into a forest structure according to the attribute identification;

[0011] Step 6: Compare G2 with G1 in the database, and perform a pre-explored extended attribute adjacency graph search using a depth-first search to identify the characteristics and types of the leaves.

[0012] In the above scheme, in step one, the engineering surface is an independent surface with a closed boundary on a model; the attributes of the engineering surface are the unique identification ID of the surface, the type of surface, the normal vector of the surface and the number of all edges on the surface; the attributes of the edge are the unique identification ID of the edge, the concavity and convexity and the angle between the surfaces where the edge is located, the type of edge and the unique identification ID of the adjacent surface.

[0013] In the above solution, the extended attribute adjacency graph G2 is to add partition attributes on the basis of the traditional attribute adjacency graph, thereby eliminating the ambiguity that may exist in the traditional attribute adjacency graph representation;

[0014] The mathematical representation of surface attributes refers to constructing surface attribute enumeration and assigning different values ​​to each attribute;

[0015] The mathematical representation of edge attributes refers to constructing edge attribute enumeration and assigning different values ​​to each attribute;

[0016] The concavity and convexity and the angle between the surfaces refer to the construction of the concavity and convexity enumeration; when the angle α between the normal vectors of the two surfaces is defined as the clockwise direction of the inner radial direction of the blade is positive, if 0°<α<90° and 90°<α<180°, the edge between the two surfaces is defined as a concave edge; if α=90°, the edge between the two surfaces is defined as a right-angle concave edge; if α=180°, the two surfaces are defined as smoothly connected; if 180°<α<270° and 270°<α<360°, the edge between the two surfaces is defined as a convex edge; if α=270°, the edge between the two surfaces is defined as a right-angle convex edge.

[0017] In the above scheme, the VF2p algorithm classifies nodes and edges according to the characteristics of turbine blades and pre-explores the target feature subgraph G1 by prioritizing nodes and matching features;

[0018] Priority node sorting refers to the matching of subgraph isomorphisms. To avoid the same mini_sub subgraph structure being accessed multiple times, it is necessary to determine the search order and try to traverse from the nodes with higher priority. Among them, a mini_sub subgraph refers to a subgraph G1 that reaches any state k, that is, when G1 and G2 have reached a matching set M(s), there are already k pairs of matching nodes. The subgraph composed of k G1 subgraph nodes that have completed the matching is called a mini_sub subgraph.

[0019] The priority node refers to the probability that the subgraph isomorphism can be completed within k rounds of traversal from any point n. The greater the probability, the higher the priority, and vice versa. The priority is determined by two constraints: Constraint 1: the priority is given to the surface nodes with the smaller probability of finding the corresponding node in the large graph, that is, the surface nodes with the same attributes; Constraint 2: the priority is given to the points with more connections to the matched points, that is, the number of edges of the corresponding attributes of the points in the large graph that match the points in the subgraph should be greater than the number of edges of the corresponding attributes of the points in the subgraph. The more edges of the large graph points corresponding to the subgraph nodes, the fewer matching candidate nodes in the large graph, and the higher the priority of the subgraph nodes; Constraint 3: in the subgraph, the more connections any node has with nodes other than itself, the higher its priority.

[0020] The method for calculating the priority nodes uses the maximum likelihood estimation method, and the calculation formula is:

[0021]

[0022] λ v1 (u) refers to the number of nodes with the same attributes as node u found in the large graph v1, Pl refers to the probability calculation based on the constraint-label, and λ v1 (u) is obtained by taking the quotient of all points in the large graph, P d Based on constraint 2, the probability of finding the number of edges in the large graph to be greater than or equal to the node, P dm It is based on constraint three and is the quotient of the sum of the nodes connected to node u in the subgraph except itself and the total number of points in the small graph G1. α is the weight factor.

[0023] In the above scheme, the value of the weight factor α estimated by the maximum likelihood method is set to α = 0.6.

[0024] In the above scheme, depth-first search means traversing the nodes of the graph along the depth of the large graph, searching the branches of the graph as deep as possible; when all the edges of node v have been explored or the node does not meet the conditions during the search, the search will backtrack to the starting node of the edge where node v was found; the whole process is repeated until all nodes have been visited;

[0025] Pre-exploration refers to modifying the fallback algorithm based on the depth-first search algorithm to determine whether the node can continue to match after two searches. Pre-exploration will prune routes with a matching probability of 0 and retain routes with a probability of successful matching.

[0026] The pre-exploration judgment follows the following rules:

[0027] F n (s c ,u n ,v n )∧F g (s c ,u n ,v n );

[0028] Among them, F n Refers to judging by node label, Fg refers to judging based on the topological structure of the graph, S c Indicates that the current state has completed the matching point set, u n Indicates the point to be matched in the subgraph for the next step of matching, v n Indicates the points to be matched in the next step of matching in the large image.

[0029] In the above scheme, the turbine blade model attribute extraction: based on the NX12.0 software expressed based on the B-rep model, secondary development is performed using UF / API to obtain the turbine blade model attributes.

[0030] Technical effects:

[0031] Compared to existing technologies, this invention can rapidly extract features from turbine blade B-rep models, identify blade types, and provide parameters for subsequent CNC machining automation. Independent of MBD models, this invention is applicable to mainstream 3D software such as UG and CATIA, which build models based on B-rep rules. Furthermore, compared to existing feature recognition methods based on the VF2 subgraph matching algorithm, this invention restructures the data structure and algorithm flow, specifically optimizing undirected graphs. This significantly increases the speed and efficiency of the feature recognition method. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 1 is an overall flow chart of the method of the present invention.

[0033] Figure 2 This is a schematic diagram of the surface nodes and edge nodes described in this method.

[0034] Figure 3 This is a schematic diagram of the data structure of the target graph G2 to be identified in this method.

[0035] Figure 4A three-dimensional representation of the feature suppression described in this method.

[0036] Figure 5 Schematic diagram of the implementation method of feature suppression according to the present invention.

[0037] Figure 6 This is a schematic diagram of a three-dimensional graph of a rhombus blade root without feature suppression and its attribute adjacency graph as described in the present invention.

[0038] Figure 7 This is a schematic diagram of a three-dimensional graph of a rhombus blade root and its attribute adjacency graph after feature suppression as described in the present invention.

[0039] Figure 8 Schematic diagram of some characteristic subgraphs described in this method.

[0040] Figure 9 This is a module structure diagram of the VF2p algorithm described in the present invention.

[0041] Figure 10 This is a module structure diagram for calculating node access probability in the VF2p algorithm described in the present invention.

[0042] Figure 11 This is a module structure diagram for determining the final node access probability in the VF2p algorithm described in the present invention.

[0043] Figure 12 This is a module structure diagram for constructing the Children structure in the VF2p algorithm described in the present invention.

[0044] Figure 13 Schematic diagram of the data structure of Children in the VF2p algorithm described in the present invention.

[0045] Figure 14 This is a module structure diagram of the depth-first search with pre-exploration in the VF2p algorithm described in the present invention.

[0046] Figure 15 This is a module structure diagram for finding the next candidate node in the VF2p algorithm described in the present invention.

[0047] Figure 16 Schematic diagram of the VF2p algorithm pre-exploration to achieve constraint one according to the present invention.

[0048] Figure 17 Schematic diagram of achieving constraint 2 during the pre-exploration process of the VF2p algorithm of the present invention.

[0049] Figure 18 Schematic diagram of achieving constraint three during the pre-exploration process of the VF2p algorithm of the present invention. DETAILED DESCRIPTION

[0050] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0051] A feature recognition method for steam turbine blades based on B-rep comprises the following steps:

[0052] Step 1: Extract attributes of the turbine blade model constructed using the B-rep (boundary representation) method, traverse all engineering surfaces of the entire model, extract the attributes of the engineering surfaces and the attributes of the edges connected to the surfaces;

[0053] Step 2: Mathematically represent the surface attributes, edge attributes, and connection methods of different engineering surfaces, construct surface nodes and edge nodes, and then combine them into the extended attribute adjacency graph G2 of the leaf;

[0054] Step 3: trim the extended attribute adjacency graph G2 by suppressing invalid features to reduce the amount of information in the target graph;

[0055] Step 4: Use the VF2p algorithm to calculate and sort the priority nodes of the target feature subgraph G1 of the knowledge base according to the primary and secondary features;

[0056] Step 5: Reconstruct the target feature subgraph G1 into a forest structure according to the attribute identification;

[0057] Step 6: Compare G2 with G1 in the database, and perform a pre-explored extended attribute adjacency graph search using a depth-first search to identify the characteristics and types of the leaves.

[0058] In the above scheme, in step one, the engineering surface is an independent surface with a closed boundary on a model; the attributes of the engineering surface are the unique identification ID of the surface, the type of surface, the normal vector of the surface and the number of all edges on the surface; the attributes of the edge are the unique identification ID of the edge, the concavity and convexity and the angle between the surfaces where the edge is located, the type of edge and the unique identification ID of the adjacent surface.

[0059] In the above solution, the extended attribute adjacency graph G2 is to add partition attributes on the basis of the traditional attribute adjacency graph, thereby eliminating the ambiguity that may exist in the traditional attribute adjacency graph representation;

[0060] The mathematical representation of surface attributes refers to constructing surface attribute enumeration and assigning different values ​​to each attribute;

[0061] The mathematical representation of edge attributes refers to constructing edge attribute enumeration and assigning different values ​​to each attribute;

[0062] The concavity and convexity and the angle between the surfaces refer to the construction of the concavity and convexity enumeration; when the angle α between the normal vectors of the two surfaces is defined as the clockwise direction of the inner radial direction of the blade is positive, if 0°<α<90° and 90°<α<180°, the edge between the two surfaces is defined as a concave edge; if α=90°, the edge between the two surfaces is defined as a right-angle concave edge; if α=180°, the two surfaces are defined as smoothly connected; if 180°<α<270° and 270°<α<360°, the edge between the two surfaces is defined as a convex edge; if α=270°, the edge between the two surfaces is defined as a right-angle convex edge.

[0063] In the above scheme, the VF2p algorithm classifies nodes and edges according to the characteristics of turbine blades and pre-explores the target feature subgraph G1 by prioritizing nodes and matching features;

[0064] Priority node sorting refers to the matching of subgraph isomorphisms. To avoid the same mini_sub subgraph structure being accessed multiple times, it is necessary to determine the search order and try to traverse from the nodes with higher priority. Among them, a mini_sub subgraph refers to a subgraph G1 that reaches any state k, that is, when G1 and G2 have reached a matching set M(s), there are already k pairs of matching nodes. The subgraph composed of k G1 subgraph nodes that have completed the matching is called a mini_sub subgraph.

[0065] The priority node refers to the probability that the subgraph isomorphism can be completed within k rounds of traversal from any point n. The greater the probability, the higher the priority, and vice versa. The priority is determined by two constraints: Constraint 1: the priority is given to the surface nodes with the smaller probability of finding the corresponding node in the large graph, that is, the surface nodes with the same attributes; Constraint 2: the priority is given to the points with more connections to the matched points, that is, the number of edges of the corresponding attributes of the points in the large graph that match the points in the subgraph should be greater than the number of edges of the corresponding attributes of the points in the subgraph. The more edges of the large graph points corresponding to the subgraph nodes, the fewer matching candidate nodes in the large graph, and the higher the priority of the subgraph nodes; Constraint 3: in the subgraph, the more connections any node has with nodes other than itself, the higher its priority.

[0066] The method for calculating the priority nodes uses the maximum likelihood estimation method, and the calculation formula is:

[0067]

[0068] λ v1 (u) refers to the number of nodes with the same attributes as node u found in the large graph v1, Pl refers to the probability calculation based on the constraint-label, and λ v1 (u) is obtained by taking the quotient of all points in the large graph, P d Based on constraint 2, the probability of finding the number of edges in the large graph to be greater than or equal to the node, P dmIt is based on constraint three and is the quotient of the sum of the nodes connected to node u in the subgraph except itself and the total number of points in the small graph G1. α is the weight factor.

[0069] In the above scheme, the value of the weight factor α estimated by the maximum likelihood method is set to α = 0.6.

[0070] In the above scheme, depth-first search means traversing the nodes of the graph along the depth of the large graph, searching the branches of the graph as deep as possible; when all the edges of node v have been explored or the node does not meet the conditions during the search, the search will backtrack to the starting node of the edge where node v was found; the whole process is repeated until all nodes have been visited;

[0071] Pre-exploration refers to modifying the fallback algorithm based on the depth-first search algorithm to determine whether the node can continue to match after two searches. Pre-exploration will prune routes with a matching probability of 0 and retain routes with a probability of successful matching.

[0072] The pre-exploration judgment follows the following rules:

[0073] F n (s c ,u n ,v n )∧F g (s c ,u n ,v n );

[0074] Among them, F n Refers to judging by node label, Fg refers to judging based on the topological structure of the graph, S c Indicates that the current state has completed the matching point set, u n Indicates the point to be matched in the subgraph for the next step of matching, v n Indicates the points to be matched in the next step of matching in the large image.

[0075] In the above scheme, the turbine blade model attribute extraction: based on the NX12.0 software expressed based on the B-rep model, secondary development is performed using UF / API to obtain the turbine blade model attributes.

[0076] Example

[0077] Combined with attachment Figure 1 The overall flow chart shown:

[0078] Step 1: Extract attributes from the turbine blade model constructed using the B-rep (boundary representation) method;

[0079] Step 2: Mathematically represent the attributes of different faces, edges, and connection methods, construct face nodes and edge nodes, and then combine them into the extended attribute adjacency graph G2 of the leaf, using a specific data structure to represent the actual structure of the leaf;

[0080] Step 3: trim the extended attribute adjacency graph by suppressing invalid features to reduce the information of the target graph;

[0081] Step 4: Use the VF2p algorithm to calculate and sort the priority access nodes of the knowledge base target feature subgraph G1 according to the primary and secondary features;

[0082] Step 5: Reconstruct the target feature subgraph G1 into a tree structure according to the priority order and attribute identification and prune it;

[0083] Step 6. Finally, a depth-first search is performed to retrieve the pre-explored extended attribute adjacency graph, identify the characteristics of the leaves, analyze the specific type through regular expressions, and store the above information for subsequent use.

[0084] Among them, in step 1, the attributes of the B-rep (boundary representation) turbine blade model are extracted. It is done by traversing all the engineering surfaces of the entire model, that is, independent surfaces with closed boundaries, extracting the attributes of the surface and the attributes of the edges connected to the surface, so as to construct the most basic unit nodes of the target graph G2: surface nodes and edge nodes. Most of the 3D industrial software on the market use the B-rep method to represent the model. Preferably, NX software is used to complete the operation. NX software provides a powerful secondary development tool set and encapsulates multiple geometric information query functions: UF_MODL_ask_face_type() is used to query the type of the surface; UF_MODL_ask_face_props() is used to obtain the normal vector of the surface; UF_MODL_ask_face_edges(); is used to obtain a list of edges, etc. The required information is obtained through the above API functions.

[0085] The properties of the engineering surface, refer to Figure 2 The data structure diagram shown is mainly composed of the following parts: the unique identification ID of the surface, which can be uniquely identified and found through a number; the type of surface. In turbine blades, the types of surfaces include planes, cylinders, free-form surfaces, etc.; the normal vector of the surface; the number of all edges on the surface; and the number of types of edges contained in the surface, which are stored in the corresponding surface node.

[0086] Edge attributes, see Figure 2The data structure diagram shown is mainly composed of the following parts: the unique identification ID of the edge, which can be uniquely identified and found through a number; the convexity and angle between the faces where the edge is located; the type of edge, in turbine blades, the types of edges include straight lines, arcs, NURBS curves, etc.; the unique identification ID of the face adjacent to the edge, which is stored in the corresponding edge node.

[0087] In step 2, mathematical representations are performed on various attributes, such as faces, edges, and connection methods, to construct an extended attribute adjacency graph for the blade. Traditional attribute adjacency graphs represent models solely based on the concavity and convexity between two faces, which can introduce errors. Therefore, this invention builds on these traditional attribute adjacency graphs by more finely dividing attributes based on the characteristics of steam turbine blades. This eliminates potential ambiguity in attribute adjacency graph representations and improves feature recognition accuracy.

[0088] The mathematical representation of surface attributes is to construct a surface type enumeration, assigning plane (PLANAR_FACE) to 1, cylinder to 2, cone to 3, sphere to 4, and free-form surface to 5; traverse the surface ring to obtain the number and type of edges; obtain the surface normal vector by the coordinates of two points, fix one of them as the origin through coordinate transformation, and stipulate that the clockwise direction along the inner radial direction of the blade is the positive direction. The vector is expressed as:

[0089]

[0090] The mathematical representation of edge properties is to construct an edge type enumeration, assigning a value of 1 to a straight line (LINEAR_EDGE), 2 to an arc (ARC_EDGE), 3 to an elliptical edge (ELLIPTICAL_EDGE), and 4 to a NURBS curve edge. The convexity between faces is determined by the angle between their normal vectors. A concave-convex enumeration is constructed, defining that when the angle α between the normal vectors of two faces is 0°<α<90° and 90°<α<180°, the edge between the two faces is defined as a concave edge, with a value of 1; if α=90°, the edge between the two faces is defined as a right-angled concave edge, with a value of 2; if α=180°, the two faces are defined as smoothly connected, with a value of 3; if 180°<α<270° and 270°<α<360°, the edge between the two faces is defined as a convex edge, with a value of 4; if α=270°, the edge between the two faces is defined as a right-angled convex edge, with a value of 5. The angle calculation formula is:

[0091] Reference Figure 3As shown, the extended attribute adjacency graph is constructed using a special data structure. Unlike the general attribute adjacency graph that uses an N*N matrix to store data, the present invention uses a combination of a hash table and a red-black tree to construct the data structure. Using the face nodes stored in the hash table, the unique ID of the engineering face can be quickly addressed to the corresponding face node with a time complexity of O(1). The face node stores a pointer to the red-black tree header, which points to the red-black tree composed of the edge nodes of all edges of the face. The rotation of the red-black tree is driven by the unique ID of the edge. The edge node stores the unique ID of another adjacent face, which can be quickly locked in the hash table.

[0092] Preferably, the hash table uses the divisor remainder method for addressing. Let the length of the hash table, that is, the number of all engineering surfaces, be m, the unique identifier id of the surface be the key key, take any number p, let p≤m, and the addressing hash function be:

[0093] H(key)=key MOD p(p≤m);

[0094] In step three, the extended attribute adjacency graph is trimmed by suppressing invalid features. In order to reduce the volume of the graph and increase the speed of retrieval, the present invention adopts a pruning strategy of feature suppression. For the invalid features of the turbine blades (specifically the edge fillet features in the present invention), the fillet surface is suppressed to an edge. Feature suppression only requires edge fusion and face node elimination operations in the attribute adjacency graph, and does not require changes to the solid model.

[0095] Edge fusion operation, see Figure 4 As shown, this means that the faces S(n) (n≥1) connected to the invalid feature are connected in pairs, and an edge is constructed between the two faces accordingly. For the edge rounding feature, there is a cylindrical surface C whose edge set E has and only has four elements: a pair of straight line edges E1 and E2, and a pair of arc edges Arc1 and Arc2. For the straight line edges E1 and E2, let the angle λce1 between the vector λc of the cylindrical surface C and the vector of the adjacent surface CE1 of E1 be α1, and the angle between λc and λce2 be α2, and there must be α1 = α2. The entire G2 target graph is traversed to search for this feature. After each feature is captured and marked, the search continues along its Arc edge. After the search is completed, the imaginary edge set node is constructed based on the information of CE1 and CE2.

[0096] Reference Figure 5 As shown in the figure, the node elimination operation is to delete all edge fillet face nodes and their associated edge set nodes, and replace the face nodes with the above-mentioned imaginary edge nodes. Figure 6As shown in the figure, a rhombus blade root structure is shown. The schematic diagram of its structural attribute adjacency graph when invalid features are suppressed is as follows. The blue part is marked as the face node of the cylindrical face; refer to Figure 7 As shown in the figure, this is a schematic diagram of the model with invalid features suppressed (in fact, the model will not be operated). After the invalid features are suppressed, the schematic diagram of its structural attribute adjacency graph is as follows Figure 7 As shown in Figure 3, invalid feature suppression greatly simplifies the volume of the target graph G2, indirectly speeding up the retrieval speed.

[0097] In step 4, the VF2p algorithm is used to calculate and sort the priority access nodes of the knowledge base target feature subgraph G1 according to the primary and secondary features. Figure 8 As shown in the figure, the VF2p algorithm is an improved VF2plus algorithm based on the VF2 algorithm. Compared with the traditional VF2 algorithm, the VF2p algorithm classifies nodes and edges according to the characteristics of turbine blades, and improves the retrieval efficiency of subgraph isomorphism to a considerable extent by sorting the priority of subgraphs and pre-exploring feature matching. Figure 9 As shown in Figure 1, the knowledge base stores a set of leaf feature subgraphs G1, which are used to judge and identify leaves. The subgraphs in the knowledge base are sorted according to the primary and secondary features, with the decisive features being placed at the first priority for matching, and the remaining features being ranked in the order of matching.

[0098] Reference Figure 10 As shown, priority node sorting uses a specific algorithm to systematically visit nodes in the G1 subgraph one by one. Subgraph matching uses a depth-first approach, a blind, fallback matching method. Without a plan for subgraph node access, there are a total of k! paths to any mini_sub subgraph structure. Planning node access can fix a single path to the mini_sub subgraph and improve the speed of successful recognition.

[0099] A mini_sub graph refers to a subgraph G1 that reaches any state s, that is, when G1 and G2 have reached a matching set M(s), there are already k pairs of matching nodes. The subgraph M(s) composed of the k G1 subgraph nodes that have completed the matching is called a mini_sub graph.

[0100] Priority refers to the probability that the subgraph isomorphism can be completed within k rounds by traversing from any point u. The greater the probability, the higher the priority, and vice versa. The priority is judged by three constraints: Constraint one, priority is given to nodes with a smaller probability of finding corresponding nodes in the large graph, that is, face nodes with the same attributes; Constraint two, priority is given to points that are more connected to matched points, that is, the number of edges of corresponding attributes of points in the large graph that match points in the subgraph should be greater than the number of edges of corresponding attributes of points in the subgraph. The more edges of large graph points corresponding to subgraph nodes, the fewer candidate nodes for matching in the large graph, and the higher the priority of subgraph nodes; Constraint three, in the subgraph, the more connections any node has with nodes other than itself, the higher its priority. From the above, it can be obtained that the precise calculation method of node priority Pfes is:

[0101] P fes (u)=P r (λ(v)=λ(u)∩d(v)≥d(u)∩P dm (u));

[0102] The priority of G1 subgraph nodes only affects the order of node access, but does not affect the correctness of the subgraph matching solution. The above exact calculation method has a very high time complexity of O(Nx), where x is the sum of the attribute types of the face nodes. Therefore, the maximum likelihood estimation method is used for approximate calculation, assuming that the three conditions are independent of each other. Figure 11 As shown, the calculation formula is optimized as follows:

[0103]

[0104] P l Refers to the probability calculation based on the constraint-label, P d Based on constraint 2, the probability of finding the number of edges in the large graph to be greater than or equal to the node, P dm Based on constraint 3, the sum of the nodes connected to node u in the subgraph other than itself and the total number of points in the small graph G1 is quotient; preferably, the weight factor α is set to 0.6. The time complexity of the above approximate solution is O(N);

[0105] In step 5, the target feature subgraph is reconstructed into a tree structure according to the attribute identifier. Figure 12 As shown, according to the priority described in step 4, the subgraph G1 is reorganized so that the face nodes are arranged in order of priority. Each node has a branch consisting of edge nodes, and the edge nodes are classified into different branches according to the type of face. This data structure is recorded as Childrenu0, that is, the tree structure with the head node being the highest priority u0 in G1, as shown in Figure 13 shown.

[0106] Step 6, refer to Figure 14As shown, a depth-first subgraph matching with pre-exploration is performed. Depth-first search involves traversing the nodes of the larger graph along its depth, searching the branches of the graph as deeply as possible. If all edges to node v have been explored or the search does not meet the conditions, the search will backtrack to the starting node of the edge that discovered node v. This process repeats until all nodes have been explored.

[0107] The node exploration rules of depth-first search are as follows Figure 15 As shown, for the nodes in the subgraph G1, the search order is searched in the order of Childrenu0. c Empty, that is, the current matching set M(sc) has no nodes participating in the matching, read the next u of Childrenu0 n Participate in matching; if u n Still empty. Then Childrenu0 matching is complete. n When it is not empty, u n Try matching as a preselected node: Find u n edge set nodes, if u n Edge Set Node Proof n If it is an isolated node, it is necessary to globally traverse the target graph G2 to find the node u n Nodes with the same attributes. Then find out whether u exists in d(uc) n , if un∈d(uc), then the preselected node v of the target graph G2 n To v c The edge set nodes with u n Search among nodes with the same attributes; if Then v n Search among the nodes in G2 except M(sc) and d(vc).

[0108] After obtaining the preselected node (un, vn) as described above, a preliminary exploration is required to determine whether it can be formally added to M(sc). To mitigate the high time complexity caused by blind searches based on the depth-first search algorithm, the fallback algorithm based on the depth-first search algorithm is modified to determine whether the node has a chance of successfully matching in two subsequent searches. The preliminary exploration prunes routes with a zero probability of matching and continues to explore routes with a high probability of matching.

[0109] The pre-exploration judgment follows the following rules:

[0110] Convenient=F n (s c ,u n ,v n)∧F g (s c ,u n ,v n );

[0111] Reference Figure 16 As shown, F n Refers to judging by node attributes, used to check the new matching set state S after (un, vn) is added to the match n Is it still in a consistent state? n The expression is:

[0112] Fn needs to meet two conditions: first, for any nodes un and vn that have entered the pre-matching, the attributes within the two face nodes should be all the same; second, all the attributes of the edge nodes between the two and uc and vc should be all the same. Figure 16 The status indicates that the Fn judgment has passed.

[0113] F g It is based on the topological structure of the graph to make judgments, and it uses certain rules to judge whether the point has the possibility of successful matching in the future matching process. g The expression is:

[0114]

[0115] F g It consists of two parts, F ep1 It is to judge whether the next step explore1 of the current match can form a consistent state, F ep1 The detailed description is:

[0116]

[0117] Among them, 1≤q≤Label.size, for any face node attribute c, we have:

[0118]

[0119] F ep1 It is used to judge whether the number of edge set nodes of the node un currently entering the pre-matching has the same attributes as all u nodes in M(sc). It should be less than or equal to the number of edge set nodes of vn with the same attributes as all v nodes in M(sc). If it is greater, it proves that F ep1 Unqualified, that is, adding (un,vn) under M(sc) will cause M(sn) to fail to match in a future step and trigger a return. For example, Figure 17 As shown, that is F ep1 qualified.

[0120] Fep2 It is a further judgment based on the result of explore1, that is, whether explore2 can form a consistent state. ep2 The detailed description is:

[0121]

[0122] Among them, 1≤q≤Label.size, for any face node attribute c, we have:

[0123]

[0124] F exp2 To judge u n The number of edge nodes in the target graph G2 should be less than or equal to the number of nodes in the target graph G2 in s c The sum of all nodes in the state except the matched nodes. For example Figure 18 As shown, that is F ep2 qualified.

[0125] By matching the feature subgraphs in the knowledge base with the larger image to be identified, the matching features are extracted in text form, and the specific blade type is analyzed using regular expressions to complete the identification. The relevant data is then saved for subsequent CNC code writing.

[0126] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0127] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and purpose of the present invention.

Claims

1. A feature recognition method for steam turbine blades based on B-rep, characterized in that: The steps include: Step 1: Extract attributes of the turbine blade model constructed using the B-rep (boundary representation) method, traverse all engineering surfaces of the entire model, extract the attributes of the engineering surfaces and the attributes of the edges connected to the surfaces; Step 2: Mathematically represent the surface attributes, edge attributes, and connection methods of different engineering surfaces, construct surface nodes and edge nodes, and then combine them into the extended attribute adjacency graph G2 of the leaf; Step 3: trim the extended attribute adjacency graph G2 by suppressing invalid features to reduce the amount of information in the target graph; Step 4: Use the VF2p algorithm to calculate and sort the priority nodes of the target feature subgraph G1 of the knowledge base according to the primary and secondary features. The VF2p algorithm classifies nodes and edges based on the characteristics of steam turbine blades and pre-explores the target feature subgraph G1 by sorting the priority nodes and matching features. Step 5: Reconstruct the target feature subgraph G1 into a forest structure according to the attribute identification; Step 6: Compare G2 with G1 in the database, and perform a pre-explored extended attribute adjacency graph search using a depth-first search to identify the characteristics and types of the leaves.

2. The feature recognition method based on B-rep steam turbine blade according to claim 1, characterized in that: In step 1, the engineering surface is an independent surface with a closed boundary on a model; the attributes of the engineering surface are the unique identification ID of the surface, the type of the surface, the normal vector of the surface and the number of all edges on the surface; the attributes of the edge are the unique identification ID of the edge, the concavity and convexity and the angle between the surfaces where the edge is located, the type of the edge and the unique identification ID of the adjacent surface.

3. The feature recognition method based on B-rep steam turbine blade according to claim 1, characterized in that: In step 2, the extended attribute adjacency graph G2 is to add partition attributes on the basis of the traditional attribute adjacency graph, eliminating the ambiguity that may exist in the traditional attribute adjacency graph representation; The mathematical representation of surface attributes refers to constructing surface attribute enumeration and assigning different values ​​to each attribute; The mathematical representation of edge attributes refers to constructing edge attribute enumeration and assigning different values ​​to each attribute; The concavity and convexity and the angle between the surfaces refer to the construction of the concavity and convexity enumeration; when the angle α between the normal vectors of the two surfaces is defined as the clockwise direction of the inner radial direction of the blade is positive, if 0°<α<90° and 90°<α<180°, the edge between the two surfaces is defined as a concave edge; if α=90°, the edge between the two surfaces is defined as a right-angle concave edge; if α=180°, the two surfaces are defined as smoothly connected; if 180°<α<270° and 270°<α<360°, the edge between the two surfaces is defined as a convex edge; if α=270°, the edge between the two surfaces is defined as a right-angle convex edge.

4. The feature recognition method based on B-rep steam turbine blade according to claim 1, characterized in that: In step four, Priority node sorting refers to the matching of subgraph isomorphisms. To avoid the same mini_subgraph structure being accessed multiple times, it is necessary to determine the search order and try to traverse from the nodes with higher priority. Among them, a mini subgraph refers to a subgraph G1 that reaches any state k, that is, when G1 and G2 have reached a matching set M(s), there are already k pairs of matching nodes. The subgraph composed of k G1 subgraph nodes that have completed the matching is called a mini_subgraph. The priority node refers to the probability that the subgraph isomorphism can be completed within k rounds of traversal from any point n. The greater the probability, the higher the priority, and vice versa. The priority is determined by two constraints: Constraint 1: the priority is given to the surface nodes with the smaller probability of finding the corresponding node in the large graph, that is, the surface nodes with the same attributes; Constraint 2: the priority is given to the points with more connections to the matched points, that is, the number of edges of the corresponding attributes of the points in the large graph that match the points in the subgraph should be greater than the number of edges of the corresponding attributes of the points in the subgraph. The more edges of the large graph points corresponding to the subgraph nodes, the fewer matching candidate nodes in the large graph, and the higher the priority of the subgraph nodes; Constraint 3: in the subgraph, the more connections any node has with nodes other than itself, the higher its priority. The method for calculating the priority nodes uses the maximum likelihood estimation method, and the calculation formula is: λ v1 (u) refers to the number of nodes with the same attributes as node u found in the large graph v1, P l Refers to the probability calculation based on the constraint-label, v1 (u) is obtained by taking the quotient of all points in the large graph, P d Based on constraint 2, the probability of finding the number of edges in the large graph to be greater than or equal to the node, P dm It is based on constraint three and is the quotient of the sum of the nodes connected to node u in the subgraph except itself and the total number of points in the small graph G1. α is the weight factor.

5. The feature recognition method based on B-rep steam turbine blade according to claim 4, characterized in that: In step 4, the value of the weight factor α estimated by the maximum likelihood method is set to α = 0.

6.

6. The feature recognition method based on B-rep steam turbine blade according to claim 1, characterized in that: In step 6, depth-first search refers to traversing the nodes of the graph along the depth of the large graph, searching the branches of the graph as deep as possible; when all the edges where node v is located have been explored or the node does not meet the conditions during the search, the search will backtrack to the starting node of the edge where node v was found; the whole process is repeated until all nodes have been visited; Pre-exploration refers to modifying the fallback algorithm based on the depth-first search algorithm to determine whether the node can continue to match after two searches. Pre-exploration will prune routes with a matching probability of 0 and retain routes with a probability of successful matching. The pre-exploration judgment follows the following rules: F n (s c ,u n ,v n )∧F g (s c ,u n ,v n ); Among them, F n Refers to judging by node label, Fg refers to judging based on the topological structure of the graph, S c Indicates that the current state has completed the matching point set, u n Indicates the point to be matched in the subgraph for the next step of matching, v n Indicates the points to be matched in the next step of matching in the large image.

7. The feature recognition method based on B-rep steam turbine blades according to claim 1, characterized in that: In step 1, the turbine blade model attributes are extracted: secondary development is performed using UF / API based on NX12.0 software expressed based on the B-rep model to obtain the turbine blade model attributes.

Images