Manufacturing process strategy prediction method based on B-rep self-supervised learning

Abstract

The invention relates to the technical field of manufacturing process strategy prediction, in particular to a manufacturing process strategy prediction method based on B-rep self-supervised learning, which comprises the following steps of: preprocessing input B-rep data, extracting geometric and topological features of each surface, ring, edge and vertex, and encoding the geometric and topological features into a feature matrix; inputting the feature matrix into an encoder-decoder model, training the model through a self-supervised learning framework, and learning an embedded representation of each surface in a potential space from B-rep data; and training the model by using a self-supervised learning framework to obtain embedded representation of each surface in a potential space. According to the method, geometric and topological information of the model is effectively learned in unlabeled CAD data through a self-supervised neural network framework, and effective learning can be carried out on a small number of samples and unlabeled data, so that accurate and rapid process strategy prediction is provided for part processing.

Description

Technical Field

[0001] The present invention relates to the technical field of manufacturing process strategy prediction, and in particular to a manufacturing process strategy prediction method based on B-rep self-supervised learning. Background Art

[0002] In the field of manufacturing process strategy prediction, traditional manufacturing process decisions mainly rely on experienced engineers and their expertise. However, this method is not only time-consuming and labor-intensive, but also prone to errors, especially when it comes to complex or high-precision manufacturing tasks. In addition, as the manufacturing industry gradually shifts from large-scale production to a flexible manufacturing model with small batches and multiple varieties, the complexity of process planning has increased significantly. In this case, the traditional decision-making process is less efficient and difficult to quickly adapt to changing production needs.

[0003] To address this problem, some manufacturing process prediction methods based on machine learning have been proposed in recent years. These methods usually use the geometric data of 3D CAD models and analyze them by converting them into point clouds, voxels or meshes.

[0004] However, when the above technologies are used in practice, there are problems such as information loss, difficulty in data labeling and poor model generality, so they still face challenges when dealing with complex manufacturing process planning tasks. Summary of the invention

[0005] The purpose of the present invention is to provide a manufacturing process strategy prediction method based on B-rep self-supervised learning to solve the problems of information loss, data annotation difficulties and poor model universality, which still face challenges when dealing with complex manufacturing process planning tasks.

[0006] To achieve the above object, the present invention provides the following technical solution: a manufacturing process strategy prediction method based on B-rep self-supervised learning, comprising the following steps:

[0007] S1. CAD data preprocessing: preprocess the input B-rep data, extract the geometric and topological features of each face, ring, edge and vertex, and encode them into a feature matrix;

[0008] S2, self-supervised pre-training: Input the feature matrix pre-processed in step S1 into the encoder-decoder model, train the model through the self-supervised learning framework, and learn the embedded representation of each face in the latent space from the B-rep data;

[0009] S3, process strategy prediction: Use the embedded representation of each surface in the latent space obtained by training the model with the self-supervised learning framework in step S2, and calculate and output the process strategy label of each surface through the process strategy prediction network;

[0010] S4. Process planning result output: output the model segmentation results in the form of graphics or text.

[0011] Preferably, the B-rep data includes topological entities at different levels of faces, rings, edges and vertices.

[0012] Preferably, in step S1, different encoding processes are performed on different types of topological entities such as faces, rings, edges and vertices to form corresponding feature matrices.

[0013] Preferably, in the encoder-decoder model of step S2, the encoder realizes information aggregation through multiple layers of residual graph convolutional layers, and the decoder realizes self-supervised learning of potential representation through signed distance function reconstruction.

[0014] Preferably, the embedded representation of each face in the latent space obtained in step S3 is used to predict the process strategy on a small sample data set to achieve segmentation and recommendation of different manufacturing process strategies.

[0015] Preferably, the process strategy labels in step S3 include bottom wall milling, cavity milling, depth profile milling, adaptive milling and drilling milling processing strategies.

[0016] Preferably, in step S3, the face embedding representation and the connection matrix between faces are used as network input in the process strategy prediction network, and the process strategy label corresponding to each face is predicted through the process strategy prediction network including a residual graph convolution layer and a multi-layer perceptron module.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] 1. The present invention directly uses boundary representation as input, completely retains the geometric and topological information of the CAD model, greatly improves the accuracy and reliability of process strategy prediction, learns geometric features from large-scale unlabeled data, and can achieve high-precision strategy prediction under the conditions of a small number of samples, effectively reducing the demand and cost of data annotation. The potential representation obtained through self-supervised learning has stronger generalization ability, and can quickly adapt to new processing tasks in small-batch and multi-variety production scenarios to meet the needs of modern manufacturing. Compared with traditional methods that rely on manual experience, it significantly improves process planning efficiency, reduces manual intervention and decision-making time, and reduces the risk of human errors, thereby providing accurate and fast process strategy prediction for parts processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flowchart of the overall structure of a manufacturing process strategy prediction method based on B-rep self-supervised learning. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Embodiment 1

[0022] See also Figure 1 The present invention provides a technical solution: a manufacturing process strategy prediction method based on B-rep self-supervised learning, comprising the following steps:

[0023] S1. CAD data preprocessing: preprocess the input B-rep data, extract the geometric and topological features of each face, ring, edge and vertex, and encode them into a feature matrix;

[0024] The B-rep data consists of topological entities at different levels, such as faces, rings, edges and vertices. There is a correlation between them. High-level topological entities can retrieve low-level topological entities. For example, face topology is a closed area surrounded by ring topology. Ring topology is composed of edge topologies of varying numbers. Edge topology has vertex topology to define the starting and ending points.

[0025] Surface topology entities are usually composed of planes, cylinders, cones, spheres, tori or B-spline surfaces, etc., which are used to construct the surface shape of the actual CAD model;

[0026] A ring is composed of multiple edges connected end to end, defining the boundaries of the component faces;

[0027] Edges can be composed of lines, arcs, circles, or B-spline curves;

[0028] The relationship between these entities forms a hierarchical graph. The surface topology entity can be retrieved from the ring topology entity, the ring topology entity can be retrieved from the edge topology entity, and the edge topology entity can be retrieved from the vertex topology entity. There is message transmission between different topological entities.

[0029] The unused parameters are encoded as zero. The geometric feature information of the faces, rings, edges and vertices of the B-rep data is extracted and expressed as a feature matrix X. f , X l , X e and X v The geometric information is extracted by the CAD kernel and compressed into a feature vector using MLP. In addition, in order to reduce the conversion, the coordinates involved are extracted using the same reference coordinate system.

[0030] The following is a detailed explanation of the input encoding for the four different topology types:

[0031] Face encode: For the encoding of CAD model faces, the first five bits of the feature matrix encode the type of surface (such as plane, cylinder, cone, sphere and torus) in a one-hot manner. B-spline surfaces are not considered in this invention. The normal direction of each face and the area parameters of the face are extracted. The radius parameters are extracted for cylindrical and conical surfaces, and the radius parameter is 0 for planes. At the same time, the coordinates of the centroid of the model face and the envelope of the face are encoded to facilitate subsequent matching.

[0032] Loop encode: For the encoding of CAD model loops, the first 10 bits of the feature matrix encode the loop type (inner loop, outer loop, etc.) in a one-hot manner. The length of the loop is set to the number of edges contained in the loop. The coordinates of the centroid and envelope of the surface composed of the loop are encoded to facilitate the matching of the surface and the loop.

[0033] Edge encode: For the encoding of CAD model edges, the first three bits of the feature matrix provide a one-hot encoding of a possible edge geometry type (straight line, circle, ellipse, spiral, intersecting curve), extract the edge's concavity and length as size parameters, and record the coordinate information of the two endpoints of the edge.

[0034] Vertices encode: For the encoding of CAD model vertices, its coordinate orientation is related to the relative position of the model faces and edges, but has nothing to do with the modeling coordinate system. Therefore, a relative coordinate system is established at the center of the model envelope to re-parameterize the vertex position information.

[0035] For CAD models, extract geometric information and construct the surface matrix and ring feature matrix X respectively. l ={l1,l2,......,l |l|}, edge matrix X e ={e1,e2,......,e |e|}, vertex matrix X v ={p1,p2,......,p |v|}. At the same time, build and The matrix, as an attribute connection graph, represents the correspondence between vertices and edges, edges and rings, and rings and faces. The input features are standardized on the training set, and the same scaling is applied to the validation set and the test set. So far, the geometric and topological information of the CAD model is processed into a structure suitable for neural network input by representing the CAD model with a lightweight matrix.

[0036] S2, self-supervised pre-training: Input the feature matrix pre-processed in step S1 into the encoder-decoder model, train the model through the self-supervised learning framework, and learn the embedded representation of each face in the latent space from the B-rep data;

[0037] The encoder-decoder model specifically includes:

[0038] Encoder: In the encoder structure, the Structured BREP Graph ConvolutionalNetwork network is referenced to process the heterogeneous input data of the CAD model. Structured BREP Graph ConvolutionalNetwork is a hierarchical graph convolutional neural network, and the BREP graph nodes are divided into 4 layers according to their topological entity types. On this basis, the present invention focuses on the face topology of B-Rep, so only the upward transmission part of the SBGCN network is retained as the network structure of the encoder. By using three graph convolutional layers (MRGCN) in the encoder, the aggregation of topological information at different levels of B-Rep is completed. The message transmission process is specifically as follows:

[0039] Formally, the extracted input feature matrix is ​​defined as a set of node sets, and the value of the node set is N = X f ∪X l ∪X e ∪X v , where X f , X l , X e , X v They are the feature matrices of faces, rings, edges, and vertices respectively. In addition, the connection matrices of face topology and ring topology are obtained respectively Connection matrix of ring topology and edge topology Connectivity matrix of edge topology and vertex topology To obtain the connection relationship between different topological entities.

[0040] First, a shared MLP linear layer is used to transform all input feature vectors into the same 64-dimensional feature vector:

[0041]

[0042] Got N (l) =H f ∪H l ∪H e ∪H vThe feature vector dimension of the new node set is 64. Then three graph convolutional layers (MRGCN) are used to implement message passing, which are denoted as MRGCNLayer1, MRGCNLayer2, and MRGCNLayer3. MRGCNLayer1 first aggregates the vertex information to the edges they are bound to, MRGCNLayer2 aggregates the bounded edge information to the topological ring, and finally MRGCNLayer3 aggregates the ring information to the topological surface. The single graph convolutional layer MRGCN obtains the information set after message passing as

[0043]

[0044] in, After the message is passed, N (l) The eigenvector matrix in ;

[0045] τ(v) means return N (l) A certain eigenvector matrix in ;

[0046] u∈Ν τ(u):τ(v) (v) indicates that the two topological features are connected, that is, the connection matrix exists;

[0047] MLP (l) It is a standard fully connected structure, consisting of linear layers, batch normalization, and ReLU. The main reason for using this structure is that the relationship of convolution operations depends on the layers, and convolution operations are only performed between adjacent layers (different layers are represented by the subscripts of the neighborhood operators). The entire message passing order follows the hierarchy between BREP topological entities: vertex to edge, edge to ring, ring to face.

[0048] Through the message passing of the encoder, the aggregation of different topological structure information of BREP is realized, and the embedded representation of the surface is obtained as a part of the input of the decoder of the present invention.

[0049] Decoder: Generally speaking, the representation of geometric models can usually be divided into explicit representation and implicit representation:

[0050] Explicit Representation means that all points on the geometric model can be represented directly or through parameter mapping. Common representations include point clouds, grids, and voxels. Implicit Representation of geometric models usually describes the distance relationship between points and defines the shape through equations or inequalities instead of directly giving the coordinates of the points.

[0051] Implicit representation usually uses an equation F(x, y, z) = 0 to define a surface or volume, where points that satisfy the equation form a geometric shape. Common implicit representations include constructive solid geometry (CSG) and distance function representation.

[0052] A signed distance function (SDF) is a continuous function that outputs the shortest distance from a given point in space to a given surface. The positive or negative value indicates whether the point is inside (negative) or outside (positive) the closed surface. The decoder is used to complete the SDF reconstruction representation of the model surface as the self-supervised learning task of the present invention.

[0053] In SDF representation, shape S can be represented as:

[0054] S={f θ (x,z)=0|x∈R 3 ,z∈R m} (3)

[0055] Among them, f θ (x,z):R 3 ×R m →R is a signed surface distance function that can be implemented using a neural network. This neural network model determines the shape of the output through the potential vector z and the sampling coordinates. By the following formula, f θ The ability to simultaneously capture and process the overall and local shape features of CAD models.

[0056]

[0057] in, Indicates the conversion of x from the global coordinate system to R i The coordinate system corresponding to x after the local coordinate system;

[0058] A latent representation vector representing the local shape;

[0059] d(x,p) is the distance function, shape S represents the face of the model, and the potential vector It represents the output of the decoder of the present invention. The local coordinates are obtained by performing UV sampling on the surface. At the same time, the UV sampling is normalized so that the UV coordinate values ​​are all between [-0.1, 1.1] for easy encoding.

[0060] In general, the decoder is building a function that maps the (u, v) coordinates of the CAD model surface to the three-dimensional coordinate position of the point (x, y, z). At the same time, in the parameter domain, the same (u, v) coordinates are mapped to the value d of the signed distance function from the point to the boundary. Combining these two functions, the encoder of the present invention realizes the reconstruction representation of the (u, v) coordinates mapped to the (x, y, z, d) four-dimensional space represented by the SDF. After experiments, the decoder is set to three fully connected residual layers.

[0061] S3, process strategy prediction: Use the embedded representation of each surface in the latent space obtained by training the model with the self-supervised learning framework in step S2, and calculate and output the process strategy label of each surface through the process strategy prediction network, specifically:

[0062] After pre-training is completed, the process strategy prediction module takes the face embedding representation as input and completes the process strategy segmentation task of the part through the process strategy prediction network. The specific steps are as follows:

[0063] Input feature representation: The embedded representation of each topological surface in the CAD model and the connection matrix between surfaces are used as network input.

[0064] Predicting process strategy: The network structure includes a residual graph convolution layer and a multi-layer perceptron (MLP) module. Under the action of the residual graph convolution layer, the information on each face is further aggregated to the adjacent faces, and then the probability value of the process strategy prediction is calculated through the MLP.

[0065] Strategy recommendation: The model finally outputs the process strategy label corresponding to each face. For example, rough machining strategies include bottom wall milling (FM), cavity milling (CM), depth contour milling (ZM), adaptive milling (AM), drilling (HD), etc. Labels of different colors are used to indicate the segmentation results of the model, so as to intuitively display the area where each process strategy is applicable in the CAD model;

[0066] S4. Process planning result output: output the model segmentation results in the form of graphics or text.

[0067] This patent innovatively embeds process information into the topological surface of the three-dimensional model, reducing the granularity of process planning.

[0068] Milling processing strategy: Milling processing strategy refers to the method of selecting processing tool path generation according to different processing shape areas. In actual process planning, users repeatedly select a group of faces or edges as the input of the processing strategy according to the characteristics of the milling strategy, and then the CAM software will call the tool path generation module to generate tool paths according to the selected milling processing strategy and face group information. In the processing process, choosing the right processing strategy is a key planning task, and the appropriate milling strategy is crucial to ensure the efficiency and accuracy of part processing. Commonly used milling strategies and their characteristics are shown in Table 1.

[0069] Table 1: Description of milling strategies and their characteristics

[0070]

[0071]

[0072] Definition 1: Roughing: Roughing is to remove the blank quickly so that the blank's geometric shape is close to the finished product in preparation for finishing. The main purpose of roughing is to pursue the material removal rate per unit time and to prepare for subsequent semi-finishing. The characteristics of roughing are that it can fully improve the processing efficiency, select as high a cutting amount as possible without affecting the processing quality, and adopt a reasonable processing technology to minimize tool wear.

[0073] Definition 2: Finishing: The finishing stage is to complete the final processing of each major surface, and use high-precision tools to process semi-finished parts completed by rough milling. The most important consideration in finishing is the surface quality of the workpiece. Depending on the characteristics and purposes of rough and fine machining, the selected machining strategies are also different.

[0074] Through the analysis of the processing data in the dataset, the processing strategies are limited to 5 types, and are divided into rough milling processing dataset and fine milling processing dataset according to different processing stages. Among them, the rough milling processing dataset contains five processing strategies: floor wall milling (FM), cavity milling (CM), depth profile milling Z-level Profile Milling ZM, adaptive milling AM and drilling (HD). The fine milling dataset includes three processing strategies: floor wall milling (FM), depth profile milling (Z-level Profile Milling), and drilling (HD). Table 1 details the advantages and disadvantages of each strategy and its applicable scenarios. Among them, the floor wall (FW) milling strategy is suitable for processing flat bottom straight wall areas, while Z-level profile milling (ZP) is usually used to process workpiece areas with steep side walls. Cavity milling (CM) is used for cavity processing, while adaptive milling (AM) is usually used for rough processing of complex contour areas. Drilling (DR) is used to process holes and threaded holes.

[0075] Different machining strategies are combined with the faces of the model to predict the machining strategies for the faces of the model. A B-rep model segmentation dataset for milling is provided, which not only contains the geometric information of the B-rep model, but also the milling process strategy corresponding to each topological surface, providing valuable data resources for subsequent process strategy prediction and model optimization.

[0076] By utilizing the boundary representation (B-rep), the core data structure of the CAD model, as the direct input for training the model in the self-supervised learning framework, the problem that traditional manufacturing process decisions rely on manual experience, are time-consuming, labor-intensive, and error-prone is effectively solved. At the same time, it overcomes the shortcomings of existing machine learning methods in terms of information retention, data labeling requirements, and model generality.

[0077] The present invention directly adopts B-rep as input, completely retains the geometric and topological information of the CAD model, provides a richer and more accurate data basis for the prediction of manufacturing process strategies, thereby significantly improving the accuracy and reliability of the prediction. By constructing a self-supervised pre-training model and using large-scale unlabeled data to learn the geometric features of the CAD model, the dependence on data annotation is effectively reduced, and the time and cost of data preparation are reduced. The potential representation learned by the present invention has stronger generalization ability, can adapt to parts of different types and shapes, and improves the adaptability and flexibility of process strategies. In addition, through automated process strategy prediction, manual intervention and decision-making time are reduced, the efficiency of process planning is significantly improved, and strong technical support is provided for the intelligence and automation of the manufacturing industry. It provides a new solution and technical path for process planning, production optimization and intelligent upgrading of the manufacturing industry, and is expected to play an important role in the future development of the manufacturing industry and promote the manufacturing industry to develop in a more efficient, smarter and more sustainable direction.

[0078] Embodiment 2

[0079] Different from the first embodiment, this embodiment uses point cloud or voxel representation as the input of the CAD model in step S1. This type of method converts the three-dimensional model into regularly sampled point cloud or voxel to adapt to the existing deep learning framework. By performing supervised learning on these discrete representation data, the process strategy prediction task can be completed. However, since the point cloud and voxel conversion will cause partial loss of geometric and topological information, and the accuracy depends on the sampling density, this method is not as good as the first embodiment in terms of information fidelity and prediction accuracy.

[0080] Embodiment 3

[0081] Different from the first embodiment, this embodiment uses a graph neural network (GNN) to learn the graph structure of the CAD model in step S2. The CAD model is represented as a graph structure composed of nodes and edges, where the nodes represent the faces or vertices of the model, and the edges represent geometric or topological relationships. By learning node features through a graph neural network, the process strategies for different areas can be predicted. However, the implementation of this method is usually more complicated, and the capture of model topology information is not as direct and efficient as in the first embodiment, so it may be limited when dealing with complex geometries and diversified process strategies.

[0082] Embodiment 4

[0083] Unlike the first embodiment, the present embodiment uses a traditional supervised learning method in step S2 to train the manufacturing process strategy through a large-scale labeled data set. This method requires a large amount of CAD model data to be manually labeled with the processing strategy corresponding to each surface, and then trains a deep learning model to predict the strategy. The accuracy of this solution depends on the size and labeling quality of the data set, but it is difficult to compare with the self-supervised learning solution of the first embodiment in terms of data labeling cost, time requirements and adaptability.

[0084] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0085] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A manufacturing process strategy prediction method based on B-rep self-supervised learning, characterized by: The following steps are involved: S1. CAD data preprocessing: preprocess the input B-rep data, extract the geometric and topological features of each face, ring, edge and vertex, and encode them into a feature matrix; S2, self-supervised pre-training: Input the feature matrix pre-processed in step S1 into the encoder-decoder model, train the model through the self-supervised learning framework, and learn the embedded representation of each face in the latent space from the B-rep data; S3, process strategy prediction: Use the embedded representation of each surface in the latent space obtained by training the model with the self-supervised learning framework in step S2, and calculate and output the process strategy label of each surface through the process strategy prediction network; S4. Process planning result output: output the model segmentation results in the form of graphics or text.

2. The manufacturing process strategy prediction method based on B-rep self-supervised learning according to claim 1, characterized in that: The B-rep data includes topological entities at different levels of face, ring, edge and vertex.

3. The manufacturing process strategy prediction method based on B-rep self-supervised learning according to claim 1, characterized in that: In the step S1, different types of topological entities such as faces, rings, edges and vertices are coded differently to form corresponding feature matrices.

4. The manufacturing process strategy prediction method based on B-rep self-supervised learning according to claim 1, characterized in that: In the encoder-decoder model of step S2, the encoder realizes information aggregation through multi-layer residual graph convolution layers, and the decoder realizes self-supervised learning of potential representation through signed distance function reconstruction.

5. The manufacturing process strategy prediction method based on B-rep self-supervised learning according to claim 1, characterized in that: The embedded representation of each face in the latent space obtained in step S3 is used to predict the process strategy on a small sample data set to achieve segmentation and recommendation of different manufacturing process strategies.

6. The manufacturing process strategy prediction method based on B-rep self-supervised learning according to claim 1, characterized in that: The process strategy labels in step S3 include bottom wall milling, cavity milling, depth profile milling, adaptive milling and drilling milling processing strategies.

7. The manufacturing process strategy prediction method based on B-rep self-supervised learning according to claim 1, characterized in that: In step S3, the face embedding representation and the connection matrix between faces are used as network input in the process strategy prediction network, and the process strategy label corresponding to each face is predicted through the process strategy prediction network including a residual graph convolution layer and a multi-layer perceptron module.

Images