Processing feature identification method and device based on hybrid neural network architecture, and product

Abstract

The invention discloses a processing feature recognition method and device based on a hybrid neural network architecture and a product, and relates to the field of part processing, and the method comprises the steps: carrying out the feature extraction of a graph structure, an initial node feature matrix and an initial edge feature matrix through a graph neural network, and obtaining a first local feature of each node, performing feature extraction on the initial node feature matrix through a KAN network to obtain a second local feature of each node, and fusing the first local feature and the second local feature of each node according to weights to obtain a target local feature of each node; pooling the target local feature of each node to obtain a first pooling result, pooling the initial node feature to obtain a second pooling result, and adding the first pooling result and the second pooling result to obtain a target node feature matrix; and performing feature extraction on the target node feature matrix through a graph neural network to obtain global features, and outputting processing features of the target product based on the global features.

Description

Technical Field

[0001] This application relates to the field of parts processing, and in particular to a method, equipment and product for processing feature recognition based on a hybrid neural network architecture. Background Technology

[0002] Machining features (such as holes, shafts, grooves, etc. in parts, as shown in the attached document) Figure 1 As shown, feature identification is commonly used in mechanical process planning. It involves identifying machining features using traditional rule-based or deep learning-based methods, and then extracting parameters (such as hole diameter and depth) for subsequent process planning. Traditional methods involve pre-defining rules, writing code, and calling a geometric kernel (open-source kernel OCC, Pythonocc) or using secondary development. Deep learning-based methods use neural networks to learn from datasets containing common machining features such as holes, shafts, grooves, and chamfers. After learning through the neural network, the machining features are classified, and the dataset may take the form of point clouds or graph structures.

[0003] Existing deep learning-based processing feature recognition methods classify processing features, but the recognition accuracy still needs improvement, for example, for relatively similar processing features (such as...). Figure 1 The hole and shaft shown Figure 2 In machining features such as grooves and bosses, these two types of machining features are often identified as the same, but in reality, they are different. For example, in CNC programming after the process route is determined, the tool entry method, programming method, and parameter settings will all be different. Therefore, it is important to distinguish between these two types of machining features. Thus, it is necessary to address the problem of low recognition accuracy in existing deep learning-based machining feature recognition methods. Summary of the Invention

[0004] The purpose of this application is to provide a processing feature recognition method, device, and product based on a hybrid neural network architecture, which can achieve high accuracy in processing feature recognition.

[0005] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a processing feature recognition method based on a hybrid neural network architecture, comprising: Construct a graph structure for the target product, and based on the graph structure, construct an initial node feature matrix and an initial edge feature matrix for the graph structure; wherein, the graph structure includes nodes and edges, the nodes are the faces of the target product, and the edges between two nodes are the connecting lines between two adjacent faces; The graph neural network extracts features from the graph structure, the initial node feature matrix, and the initial edge feature matrix to obtain the first local features of each node. The KAN network extracts features from the initial node feature matrix to obtain the second local features of each node. The first local features and the second local features of each node are then fused according to their weights to obtain the target local features of each node. Pooling is performed on the target local features of each node to obtain a first pooling result, pooling is performed on the initial node features to obtain a second pooling result, and the first pooling result and the second pooling result are added together to obtain the target node feature matrix; The graph neural network is used to extract features from the feature matrix of the target node to obtain global features, and the processing features of the target product are output based on the global features.

[0006] In one embodiment, the set of first local features of each of the nodes h conv Represented as: in, linear GNN This represents a linear layer in the graph neural network. g , h and he These represent the graph structure, the initial node feature matrix, and the initial edge feature matrix, respectively.

[0007] In one embodiment, the set of second local features of each of the nodes h kan Represented as: in, linear KAN This represents a linear layer of the KAN network. h This represents the initial node feature matrix.

[0008] In one embodiment, the set of target local features of each of the nodes LF Represented as: in, a This represents the adaptive fusion coefficient.

[0009] In one embodiment, the target node feature matrix h G Represented as: in, POOLThis indicates a pooling operation. h i express h The feature of the i-th node, N express h The total number of node features in the data. l k express LF The local feature of the k-th target in the middle, K express LF The total number of local features of the target in the image. W 1 and W 2 represents two coefficients respectively.

[0010] In one embodiment, the processing feature recognition method based on a hybrid neural network architecture further includes: Calculate the included angle between two adjacent faces in each of the aforementioned processing features. θ : in, N 1 and N 2 represents the normal vectors of two adjacent faces; For any of the aforementioned processing features, if the included angle between two adjacent faces... If the processing feature is determined to be a concave feature, and the included angle between two adjacent surfaces is... If so, the processing feature is determined to be a convex feature.

[0011] In one embodiment, the processing feature recognition method based on a hybrid neural network architecture further includes: For any of the processing features, determine whether the processing feature has a closed annular boundary line. If it does, the processing feature is determined to be a closed feature; otherwise, the processing feature is determined to be an open feature.

[0012] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the processing feature recognition method based on the hybrid neural network architecture described above.

[0013] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the processing feature recognition method based on a hybrid neural network architecture described above.

[0014] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the processing feature recognition method based on a hybrid neural network architecture described above.

[0015] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a processing feature recognition method, device, and product based on a hybrid neural network architecture. The key focus is on providing a novel hybrid neural network architecture. In this architecture, local feature extraction is first achieved through a graph neural network and a KAN network. The input to the graph neural network is the graph structure, the initial node feature matrix, and the initial edge feature matrix, while the input to the KAN network is the initial node feature matrix. Then, the graph neural network extracts global features from the local features. Finally, the processing features of the target product are output based on the global features. Compared to existing processing feature recognition architectures, this novel hybrid neural network architecture achieves higher processing feature recognition accuracy. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of some similar processing features in one embodiment of this application. Figure 1 ; Figure 2 This is a schematic diagram of some similar processing features in one embodiment of this application. Figure 2 ; Figure 3 This is a flowchart illustrating a processing feature recognition method based on a hybrid neural network architecture in one embodiment of this application. Figure 4 This is a geometric adjacency attribute diagram of the target product in one embodiment of this application; Figure 5 This is a schematic diagram of the adjacency matrix of the target product in one embodiment of this application; Figure 6 This is a schematic diagram of a recessed connection in one embodiment of this application; Figure 7 This is a schematic diagram of a convex connection in one embodiment of this application; Figure 8 This is a comparative diagram of closed and open features in one embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] In one embodiment of this application, a processing feature recognition method based on a hybrid neural network architecture is provided, specifically including steps S110, S120, S130 and S140.

[0021] Step S110: Construct the graph structure of the target product and the initial node feature matrix and initial edge feature matrix of the graph structure; wherein, the graph structure includes nodes and edges, nodes are faces of the target product, and the edge between two nodes is the connecting line between two adjacent faces.

[0022] Specifically, a corresponding graph structure can be constructed based on the CAD drawing of the target product. In the graph structure, each node corresponds to a face of the target product, and the node characteristics include the face type, such as a plane or a cylindrical face; each edge corresponds to a boundary line of the target product, and the existence of an edge between two nodes indicates that the corresponding two faces are adjacent faces in the target product, and there is a connecting line between them, which is a boundary line of the target product.

[0023] For example, the connection relationships between the faces are extracted from the CAD model of the target product to form a geometric adjacency attribute graph. This graph is then converted into an adjacency matrix, which represents the connection relationship between any two faces. The adjacency matrix is ​​an edge feature matrix.

[0024] Specifically, the geometric adjacency attribute graph is as follows: Figure 4 As shown, each F1 to F21 represents a different face, corresponding to a different node in the graph structure. A connecting line between two faces indicates that the two faces are connected in the target product. The adjacency matrix is ​​as follows: Figure 5 As shown, each matrix element corresponds to two faces. If the matrix element is 1, it means that the two faces are connected in the target product. If the matrix element is 0, it means that the two faces are not connected in the target product.

[0025] Through the above process, the type of each face and the connection relationship between any two faces can be obtained from the CAD model of the target product, thereby constructing the graph structure of the target product, as well as the initial node feature matrix and the initial edge feature matrix of the graph structure. The initial node feature matrix is ​​the set of features for each node in the graph structure, and the initial edge feature matrix is ​​the set of features for each edge in the graph structure. Node features include the node type, and edge features are the element values ​​in the adjacency matrix.

[0026] Step S120: The graph structure, initial node feature matrix and initial edge feature matrix are extracted by graph neural network to obtain the first local feature of each node. The initial node feature matrix is ​​extracted by KAN network to obtain the second local feature of each node. The first local feature and the second local feature of each node are fused according to weight to obtain the target local feature of each node.

[0027] Among them, the set of first local features of each node h conv Represented as: in, linear GNN This represents a linear layer in a graph neural network. g , h and he These represent the graph structure, the initial node feature matrix, and the initial edge feature matrix, respectively.

[0028] The set of second local features of each node h kan Represented as: in, linear KAN This represents a linear layer in a KAN network. h This represents the initial node feature matrix.

[0029] The set of target local features of each node LF Represented as: in, a This represents the adaptive fusion coefficient. a ϵ[0,1], during initialization a =0.5, indicating that initially, this coefficient has the same effect on both channels; during training, a It will automatically adjust so that the hybrid neural network model learns the optimal two-way feature fusion strategy.

[0030] In this embodiment, local feature extraction is achieved through graph neural networks and KAN networks, respectively, with each local feature being a node feature.

[0031] On the one hand, the graph structure, the initial node feature matrix, and the initial edge feature matrix are all used as inputs to the graph neural network. Feature extraction is achieved through the linear layers of the graph neural network to obtain the feature set. h conv It includes the first local features of each node, which is essentially a node feature matrix.

[0032] On the other hand, the initial node feature matrix is ​​used as input to the KAN network, and feature extraction is achieved through the linear layers of the KAN network to obtain the feature set. h kan It includes the second local features of each node, which is essentially a node feature matrix.

[0033] Finally, the two node feature matrices are fused to obtain a new node feature matrix, i.e., the set. LF In a matrix or set, each element represents a local feature of a target node.

[0034] Step S130: Pool the target local features of each node to obtain the first pooling result, pool the initial node features to obtain the second pooling result, and add the first pooling result and the second pooling result to obtain the target node feature matrix.

[0035] Specifically, the target node feature matrix h G Represented as: in, POOL This indicates a pooling operation. h i express h The feature of the i-th node, N express h The total number of node features in the data. l k express LF The local feature of the k-th target in the middle, K express LF The total number of local features of the target in the image. W 1 and W 2 represents two coefficients respectively.

[0036] Step S140: Extract global features from the feature matrix of the target node using a graph neural network, and output the processing features of the target product based on the global features.

[0037] After obtaining the target node feature matrix h G Then, it is input into a graph neural network: in, GF Represents global features. linear GNN This represents a linear layer in a graph neural network. Global features include the final feature representation of each node.

[0038] Finally, the global features are input into the output head of the hybrid neural network to obtain the processing features of the target product. These processing features include holes, shafts, grooves, chamfers, etc.

[0039] This embodiment focuses on providing a novel hybrid neural network architecture. In this architecture, local feature extraction is first achieved through a graph neural network and a KAN network. The graph neural network receives the graph structure, initial node feature matrix, and initial edge feature matrix as input, while the KAN network receives the initial node feature matrix as input. Then, the graph neural network extracts global features from the local features. Finally, the processing features of the target product are output based on the global features. Compared to existing processing feature recognition architectures, this novel hybrid neural network architecture achieves higher processing feature recognition accuracy.

[0040] To further improve the accuracy of distinguishing similar features, in this embodiment, the processing feature recognition method based on a hybrid neural network architecture further includes steps S150 and S160.

[0041] Step S150: Calculate the included angle between two adjacent faces in each machining feature. θ : in, N 1 and N 2 represents the normal vectors of two adjacent faces.

[0042] Step S160: For any processing feature, if the included angle between two adjacent faces... If the processing feature is determined to be concave, then the included angle between two adjacent faces is considered a concave feature. If so, the processing feature is determined to be a convex feature.

[0043] There are three types of connection relationships between two surfaces: no connection, concave connection, and convex connection. (Refer to...) Figure 6 and Figure 7 , Figure 6 and Figure 7 Let OC represent the concave and convex connections between two faces, respectively. AO and BO are the boundary lines of face 1 and face 2, respectively. Let E be the midpoint of BO and D be the midpoint of AO. Connect DE and take F as the midpoint of DE. The coordinates of F are (…). Fx, Fy, Fz ); The coordinate of O is ( Ox , Oy, Oz Using the +Z direction as a reference direction, the connection relationship between surfaces is determined by judging the positions of F and O. Fz > Oz This indicates that F is above O, and 0° < θ If the angle is less than 180°, the connection is a concave connection; otherwise, it means that 180° ≤ θ <360°, the connection relationship is convex connection.

[0044] θ The specific calculation method for the value is as follows: N 1. N 2 are the normal vectors of face 1 and face 2, respectively. These can be calculated using the Python OCC kernel. N 1. N 2.

[0045] By following the steps above, it is possible to distinguish whether the connection between two adjacent faces in a machining feature is concave or convex. This determines whether the machining feature is concave or convex, which can further differentiate between similar machining features.

[0046] For example, cavities and bosses, while having similar shapes, might be identified as the same processing feature, they are actually different features. However, they can be further distinguished by their concavity / convexity. Specifically, for a cavity, if the cavity is concave overall, formed by a bottom surface and side surfaces, and the connection between the bottom and side surfaces is concave, then the cavity can be identified as a concave feature. For a boss, if the boss is convex overall, having side surfaces and a top surface, and the connection between the side surfaces and the top surface is convex, then the boss can be identified as a convex feature. Therefore, by determining concavity / convexity, some similar features can be further differentiated.

[0047] Furthermore, the processing feature recognition method based on a hybrid neural network architecture also includes step S170.

[0048] For any processing feature, determine whether the processing feature has a complete circular boundary line. If it does, the processing feature is determined to be a closed feature; otherwise, the processing feature is determined to be an open feature.

[0049] For example, refer to Figure 8 L1 is a closed groove with a complete annular boundary line, L2 is an open groove without a complete annular boundary line, L3 is a closed hole with a complete annular boundary line, and L4 is an open hole without a complete annular boundary line.

[0050] The above explains the determination of concavity / convexity and closure. Steps S150 and S160 are the concavity / convexity determination steps, which determine whether the processed feature is concave or convex, thus further distinguishing similar features, such as differentiating cavities and bosses of the same shape. Step S170 is the closure determination step, which determines whether the processed feature is closed or open, further subdividing the same type of processed feature. For example, it can determine whether a hole is closed or open, or whether a groove is closed or open, etc.

[0051] Therefore, through the above steps S150 to S170, more granular processing feature recognition can be achieved, thereby improving the accuracy of processing feature recognition.

[0052] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0053] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0054] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0055] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0056] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0057] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0058] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0059] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for machining feature recognition based on hybrid neural network architecture, characterized in that, include: Construct a graph structure for the target product, as well as an initial node feature matrix and an initial edge feature matrix for the graph structure; wherein, the graph structure includes nodes and edges, the nodes are faces of the target product, and the edges between two nodes are connecting lines between two adjacent faces; The graph neural network extracts features from the graph structure, the initial node feature matrix, and the initial edge feature matrix to obtain the first local features of each node. The KAN network extracts features from the initial node feature matrix to obtain the second local features of each node. The first local features and the second local features of each node are then fused according to their weights to obtain the target local features of each node. Pooling is performed on the target local features of each node to obtain a first pooling result, pooling is performed on the initial node features to obtain a second pooling result, and the first pooling result and the second pooling result are added together to obtain the target node feature matrix; The graph neural network is used to extract features from the feature matrix of the target node to obtain global features, and the processing features of the target product are output based on the global features.

2. The processing feature recognition method based on a hybrid neural network architecture according to claim 1, characterized in that, The set of first local features of each of the nodes h conv Represented as: in, linearGNN This represents a linear layer in the graph neural network. g , h and he These represent the graph structure, the initial node feature matrix, and the initial edge feature matrix, respectively.

3. The processing feature recognition method based on a hybrid neural network architecture according to claim 2, characterized in that, The set of second local features of each of the nodes h kan Represented as: in, linearKAN This represents a linear layer of the KAN network. h This represents the initial node feature matrix.

4. The processing feature recognition method based on a hybrid neural network architecture according to claim 3, characterized in that, The set of target local features of each of the nodes LF Represented as: in, a This represents the adaptive fusion coefficient.

5. The processing feature recognition method based on a hybrid neural network architecture according to claim 4, characterized in that, The target node feature matrix h G is represented as: in, POOL This indicates a pooling operation. h i express h The feature of the i-th node, N express h The total number of node features in the data. l k express LF The local feature of the k-th target in the middle, K express LF The total number of local features of the target in the image. W 1 and W 2 represents two coefficients respectively.

6. The processing feature recognition method based on a hybrid neural network architecture according to claim 1, characterized in that, Also includes: Calculate the included angle between two adjacent faces in each of the aforementioned processing features. θ : in, N 1 and N 2 represents the normal vectors of two adjacent faces; For any of the aforementioned processing features, if the included angle between two adjacent faces... If the processing feature is determined to be a concave feature, and the included angle between two adjacent surfaces is... If so, the processing feature is determined to be a convex feature.

7. The processing feature recognition method based on a hybrid neural network architecture according to claim 1, characterized in that, Also includes: For any of the processing features, determine whether the processing feature has a closed annular boundary line. If it does, the processing feature is determined to be a closed feature; otherwise, the processing feature is determined to be an open feature.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the processing feature recognition method based on a hybrid neural network architecture as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the processing feature recognition method based on a hybrid neural network architecture as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the processing feature recognition method based on a hybrid neural network architecture as described in any one of claims 1-7.

Images