Three-dimensional model generation method and device based on industrial part genes and medium

Abstract

The invention provides a three-dimensional model generation method and device based on industrial part genes and a medium, and relates to the technical field of intelligent industrial manufacturing. The method comprises the steps of obtaining sampling data of a to-be-generated industrial part, the sampling data comprising surface sampling data and edge sampling data; industrial part gene data of the to-be-generated industrial part are obtained, and the industrial part gene data correspond to different phenotypes of the industrial part; a trained three-dimensional generation model is obtained, the three-dimensional generation model is realized based on a diffusion converter, the three-dimensional generation model comprises a surface denoising network and an edge denoising network, and the trained three-dimensional generation model is obtained by training a surface data training sample, an edge data training sample and an industrial part gene data training sample; and inputting the surface sampling data, the edge sampling data and the industrial part gene data into the trained denoising model to generate a three-dimensional model corresponding to the to-be-generated industrial part through the trained denoising model.

Description

Technical Field

[0001] This application relates to the technical field of intelligent industrial manufacturing, and particularly to a three-dimensional model generation method, device, and medium based on industrial part genes. Background Art

[0002] A Computer-Aided Design (CAD) model is a three-dimensional model created using CAD software. It can contain information such as the geometry, dimensions, and materials of an object, and can be visually presented and edited. CAD models are widely used in the design and manufacturing processes of fields such as architecture, machinery, electronics, and aviation. Especially in the industrial field, CAD models are widely applied. The CAD generation model usually uses the "sketch and extrusion" representation method. The three-dimensional (3D) shape range of industrial parts generated by this method is limited, limited to generating sketches composed of three primitive elements: straight lines, arcs, and circles, and limited to performing extrusion operations, unable to achieve complex 3D shapes composed of complex curves and surfaces.

[0003] In related technologies, Boundary Representation (B-rep) is another CAD model description method, which precisely expresses complex geometries through parametric curves, surfaces, and the topological relationships of faces-edges-vertices, and is especially suitable for free-form surface modeling.

[0004] However, the existing CAD model generation methods are limited to specific shape primitives, unable to meet the requirements for generating complex CAD models, and the efficiency of CAD model generation is low. Summary of the Invention

[0005] This application provides a three-dimensional model generation method, device, and medium based on industrial part genes to solve the technical problems that the existing CAD model generation methods are limited to specific shape primitives, unable to meet the requirements for generating complex CAD models, and the efficiency of CAD model generation is low.

[0006] In a first aspect, this application provides a three-dimensional model generation method based on industrial part genes, including:

[0007] Obtaining sampling data of the industrial part to be generated, where the sampling data includes surface sampling data and edge sampling data;

[0008] Obtaining industrial part gene data of the industrial part to be generated, where the industrial part gene data corresponds to different phenotypes of the industrial part;

[0009] Obtain a trained 3D generation model, where the 3D generation model is implemented based on a diffusion transformer, and the 3D generation model includes a face denoising network and an edge denoising network. The trained 3D generation model is obtained by training with face data training samples, edge data training samples, and industrial part gene data training samples;

[0010] Input the face sampling data, the edge sampling data, and the industrial part gene data into the trained denoising model, so as to generate a 3D model corresponding to the industrial part to be generated through the trained denoising model.

[0011] The embodiment of the present application provides a CAD model generation method based on industrial part DNA. For the industrial part to be generated, sample its face sampling data and edge sampling data, and obtain the corresponding industrial part DNA data. Input the above face sampling data, edge sampling data, and face sampling data and edge sampling data into the trained 3D generation model, and then the CAD model corresponding to the industrial part to be generated can be generated through the trained 3D generation model. Here, the 3D generation model realizes face denoising and edge denoising based on the Diffusion Transformer (DiT) architecture. The DiT architecture solves the limitations of traditional diffusion models, can efficiently model long-range dependencies, capture richer global features, while improving the modeling ability of the denoising network, solves the problem that training is not easy to converge, and provides support for accurate denoising of the 3D generation model. The structure of the 3D generation model regards the vertex connection as an edge, and only designs two denoising networks for the surface and the edge, avoiding the design of the vertex denoising network, reducing the calculation amount, and improving the generation efficiency. On the basis of this model architecture, combined with industrial part DNA data, it can guide the generation of industrial parts, meet the requirements of complex CAD model generation, and improve the efficiency of CAD model generation.

[0012] Optionally, before obtaining the trained 3D generation model, it further includes:

[0013] Obtain an original 3D generation model;

[0014] Obtain face data training samples, edge data training samples, and industrial part gene data training samples;

[0015] Train the original 3D generation model according to the face data training samples, the edge data training samples, and the industrial part gene data training samples to obtain the trained 3D generation model.

[0016] Here, in the embodiments of the present application, a three-dimensional generation model is established in advance, and the model is trained based on surface data training samples, edge data training samples, and industrial part gene data training samples to obtain a trained three-dimensional generation model that can achieve accurate denoising and CAD model generation. During the training process, the introduction of industrial part DNA data can guide the generation of various industrial parts, solve the problem of part generation limitations, and improve generation accuracy. Through reliable training, the accuracy of the three-dimensional generation model can be improved, and the yield rate of the CAD model generated by the three-dimensional generation model can be increased.

[0017] Optionally, the obtaining of the original three-dimensional generation model includes:

[0018] Construct a surface denoising sub-model, where the surface denoising sub-model includes a surface encoder, a surface decoder, and a surface denoising network, and the surface denoising network is composed of multiple diffusion transformers;

[0019] Construct an edge denoising sub-model, where the edge denoising sub-model includes an edge encoder, an edge decoder, and an edge denoising network, and the edge denoising network is composed of multiple diffusion transformers;

[0020] Establish an original three-dimensional generation model according to the surface denoising sub-model and the edge denoising sub-model;

[0021] Wherein, the output of the surface denoising sub-model is connected to the input of the edge denoising sub-model.

[0022] Here, in the embodiments of the present application, two encoders, two decoders, and two denoising networks are trained in advance. Specifically, the surface encoder and the edge encoder are used to encode the input data into the latent variable space to improve the processing efficiency. The surface denoising network and the edge denoising network are both composed of multiple diffusion transformers, which can improve the denoising performance of the surface denoising network and the edge denoising network. The embodiments of the present application also connect the output of the surface denoising sub-model to the input of the edge denoising sub-model to introduce the features of the surface denoising network into the edge denoising network, thereby guiding the model to better remove edge noise and improving the denoising effect and generation accuracy of the three-dimensional generation model.

[0023] Optionally, the training of the original three-dimensional generation model according to the surface data training samples, the edge data training samples, and the industrial part gene data training samples to obtain the trained three-dimensional generation model includes:

[0024] Train the surface encoder and the surface decoder according to the surface data training samples to obtain a trained surface encoder and a trained surface decoder;

[0025] Train the edge encoder and the edge decoder according to the edge data training samples to obtain a trained edge encoder and a trained edge decoder;

[0026] Freeze the parameters of the trained surface encoder, the trained surface decoder, the trained edge encoder, and the trained edge decoder, and train the surface denoising network and the edge denoising network according to the surface data training samples, the edge data training samples, and the industrial part gene data training samples to obtain a trained surface denoising network and a trained edge denoising network;

[0027] Obtain the trained 3D generation model according to the trained surface encoder, the trained surface decoder, the trained edge encoder, the trained edge decoder, the trained surface denoising network, and the trained edge denoising network.

[0028] Here, in the training process of the 3D generation model in the embodiments of the present application, the surface encoder and decoder, and the edge encoder and decoder are first trained, and then in the subsequent training process, the data processing efficiency is improved. After the surface encoder and decoder, and the edge encoder and decoder are trained, the parameters of the surface encoder and decoder, and the edge encoder and decoder are respectively frozen to train the two sequential denoising models of the surface denoising network and the edge denoising network to obtain a surface denoising network and an edge denoising network with good denoising effects, realizing the training of the 3D generation model, with high training efficiency and strong reliability.

[0029] Optionally, the surface data training samples include multiple surface noise samples, the surface latent feature samples corresponding to the surface noise samples, and the corresponding surface denoising samples;

[0030] The edge data training samples include multiple edge noise samples, the edge latent feature samples corresponding to the edge noise samples, and the corresponding edge denoising samples;

[0031] The industrial part gene data training samples include multiple industrial part gene data samples.

[0032] Optionally, the sampled data is boundary representation data;

[0033] Correspondingly, generating the 3D model corresponding to the industrial part to be generated through the trained denoising model includes:

[0034] Determine the target boundary representation data of the tree structure corresponding to the sampled data according to the output result of the trained denoising model;

[0035] Generate the 3D model corresponding to the industrial part to be generated according to the target boundary representation data.

[0036] Among them, in the model representation of the CAD model in the embodiments of the present application, the "sketch and extrusion" representation method that is difficult to represent complex curves and surfaces is abandoned, and B-rep is used for representation. The geometry and topology of the B-rep model are unified into a hierarchical tree with a fixed graph topology, where the node features encode geometric information, and the repeated nodes implicitly encode topological information, having geometric accuracy and editing flexibility, and being able to ensure the generation efficiency of the CAD model of industrial parts.

[0037] Optionally, the industrial part gene data includes at least one of shape, size, material, type, use, and assembly relationship.

[0038] Based on different division dimensions, the embodiments of the present application define a plurality of DNA data including shape, size, material, type, use, and assembly relationship, which are used to guide the accurate, reliable, and flexible generation of industrial parts.

[0039] In a second aspect, the embodiments of the present application provide a three-dimensional model generation device based on industrial part genes, including:

[0040] A sampling module, configured to obtain sampling data of the industrial part to be generated, where the sampling data includes surface sampling data and edge sampling data;

[0041] A first acquisition module, configured to obtain the industrial part gene data of the industrial part to be generated, where the industrial part gene data corresponds to different phenotypes of the industrial part;

[0042] A second acquisition module, configured to obtain a trained three-dimensional generation model, where the three-dimensional generation model is implemented based on a diffusion transformer, and the three-dimensional generation model includes a surface denoising network and an edge denoising network, and the trained three-dimensional generation model is trained by surface data training samples, edge data training samples, and industrial part gene data training samples;

[0043] A generation module, configured to input the surface sampling data, the edge sampling data, and the industrial part gene data into the trained denoising model, so as to generate a three-dimensional model corresponding to the industrial part to be generated through the trained denoising model.

[0044] Optionally, before the second acquisition module is configured to obtain a trained three-dimensional generation model, the above device further includes:

[0045] A model construction module, configured to obtain an original three-dimensional generation model;

[0046] A sample acquisition module, configured to obtain surface data training samples, edge data training samples, and industrial part gene data training samples;

[0047] A training module, configured to train the original 3D generation model based on the surface data training samples, the edge data training samples, and the industrial part gene data training samples, so as to obtain the trained 3D generation model.

[0048] Optionally, the model construction module is configured to:

[0049] Construct a surface denoising sub-model, where the surface denoising sub-model includes a surface encoder, a surface decoder, and a surface denoising network, and the surface denoising network is composed of multiple diffusion transformers;

[0050] Construct an edge denoising sub-model, where the edge denoising sub-model includes an edge encoder, an edge decoder, and an edge denoising network, and the edge denoising network is composed of multiple diffusion transformers;

[0051] Establish an original 3D generation model according to the surface denoising sub-model and the edge denoising sub-model;

[0052] Wherein, the output of the surface denoising sub-model is connected to the input of the edge denoising sub-model.

[0053] Optionally, the training module is configured to:

[0054] Train the surface encoder and the surface decoder according to the surface data training samples, so as to obtain a trained surface encoder and a trained surface decoder;

[0055] Train the edge encoder and the edge decoder according to the edge data training samples, so as to obtain a trained edge encoder and a trained edge decoder;

[0056] Freeze the parameters of the trained surface encoder, the trained surface decoder, the trained edge encoder, and the trained edge decoder, and train the surface denoising network and the edge denoising network according to the surface data training samples, the edge data training samples, and the industrial part gene data training samples, so as to obtain a trained surface denoising network and a trained edge denoising network;

[0057] Obtain the trained 3D generation model according to the trained surface encoder, the trained surface decoder, the trained edge encoder, the trained edge decoder, the trained surface denoising network, and the trained edge denoising network.

[0058] Optionally, the surface data training samples include multiple surface noise samples, the surface latent feature samples corresponding to the surface noise samples, and the corresponding surface denoising samples;

[0059] The edge data training samples include a plurality of edge noise samples, the corresponding edge latent feature samples of the edge noise samples, and the corresponding edge denoising samples;

[0060] The industrial part gene data training samples include a plurality of industrial part gene data samples.

[0061] The sampling data is boundary representation data;

[0062] Correspondingly, the generation module is specifically configured to:

[0063] Determine the target boundary representation data of the tree structure corresponding to the sampling data according to the output result of the trained denoising model;

[0064] Generate a three-dimensional model corresponding to the industrial part to be generated according to the target boundary representation data.

[0065] Optionally, the industrial part gene data includes at least one of shape, size, material, type, use, and assembly relationship.

[0066] In a third aspect, the present application provides a three-dimensional model generation device based on industrial part genes, including: a memory, a processor;

[0067] The memory stores computer execution instructions;

[0068] The processor executes the computer execution instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementation manners of the first aspect.

[0069] In a fourth aspect, the present application provides a computer-readable storage medium, in which computer execution instructions are stored, and when the computer execution instructions are executed by a processor, they are used to implement the above first aspect and / or various possible implementation manners of the first aspect.

[0070] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the above first aspect and / or various possible implementation manners of the first aspect.

[0071] The 3D model generation method, device, and medium based on industrial part genes provided by this application sample the surface sampling data and edge sampling data of the industrial part to be generated, and obtain the corresponding industrial part DNA data. Inputting the above surface sampling data, edge sampling data, and the surface sampling data and edge sampling data into the trained 3D generation model, the CAD model corresponding to the industrial part to be generated can be generated through the trained 3D generation model. Here, the 3D generation model realizes surface denoising and edge denoising based on the DiT architecture. The DiT architecture solves the limitations of traditional diffusion models, can efficiently model long-range dependencies, capture richer global features, while improving the modeling ability of the denoising network, solves the problem that training is not easy to converge, and provides support for the accurate denoising of the 3D generation model. The structure of the 3D generation model, by regarding the vertex connection as an edge, only designs two denoising networks for the surface and the edge, avoiding the design of the vertex denoising network, reducing the computational amount, and improving the generation efficiency. Based on this model architecture, combined with industrial part DNA data, it can guide the generation of industrial parts, meet the requirements for generating complex CAD models, and improve the efficiency of CAD model generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The drawings herein are incorporated into the specification and form a part of this specification, showing embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0073] Figure 1 A schematic diagram of the mapping representation from biological genes to industrial part genes provided by an embodiment of this application;

[0074] Figure 2 A schematic diagram of the architecture of a 3D model generation system based on industrial part genes provided by an embodiment of this application;

[0075] Figure 3 A flowchart of a 3D model generation method based on industrial part genes provided by an embodiment of this application;

[0076] Figure 4 A schematic diagram of the structure of a 3D generation model provided by an embodiment of this application;

[0077] Figure 5 A schematic diagram of the structure of a 3D model generation device based on industrial part genes provided by an embodiment of this application;

[0078] Figure 6 A schematic diagram of the structure of a 3D model generation device based on industrial part genes provided by an embodiment of this application.

[0079] Through the above-mentioned drawings, specific embodiments of the present application have been shown, and more detailed descriptions will be provided hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. Detailed Embodiments

[0080] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application.

[0081] First, the terms related to the present application are explained:

[0082] CAD model: A CAD model is a three-dimensional model created using computer-aided design (CAD) software. It can contain information such as the geometry, dimensions, and materials of an object, and can be used for visualization, presentation, editing, and modification. CAD models are widely used in the design and manufacturing processes of fields such as architecture, machinery, electronics, and aviation.

[0083] "Sketch and Extrusion": It is one of the main formats for describing CAD models. The workflow is to draw loops of two-dimensional (2D) curves as external and internal boundaries to create a 2D profile; extrude the 2D profile into a 3D shape; add or subtract 3D shapes to construct a complex CAD model.

[0084] B-rep: It is one of the main formats for describing CAD models and is widely used in free-form surface modeling to represent complex geometries. B-rep allows parametric curves and surfaces to replace the planes and linear edges used in mesh modeling. A B-rep consists of a set of interconnected faces, edges, and vertices. A face is the visible region of a parametric surface, bounded by a closed loop formed by its adjacent edges, while an edge is the visible region of a parametric curve, trimmed by the vertices that define its start and end points. Recording the adjacency relationships of adjacent edges and vertices enables the structure to provide a complete description of the final solid shape.

[0085] Currently, using the "sketch and extrusion" representation method, the range of 3D shapes of industrial parts generated is limited. It is only limited to generating sketches composed of three primitives: straight lines, arcs, and circles, and is only limited to performing extrusion operations, unable to achieve complex 3D shapes composed of complex curves and surfaces. Another representation method is B-rep, which is the main format for describing shapes and is widely used in free-form surface modeling to express complex geometries. Existing methods for generating CAD models based on boundary representation are limited to specific 3D shape primitives (such as prismatic), restricting the complexity of the results and the amount of data that can be trained. The inference speed of existing methods for generating CAD models based on boundary representation is slow, and it is difficult to generate well-structured geometric structures and correct topological relationships. When given constraints (such as categories), the yield rate of the generated models is low. Considering the existing technologies comprehensively, the generation of CAD models is limited to specific shape primitives, unable to meet the requirements for generating complex CAD models, and the efficiency of CAD model generation is low.

[0086] To solve the above problems, the embodiments of the present application provide a three-dimensional model generation method, device, and medium based on industrial part genes. For the industrial part to be generated, its surface sampling data and edge sampling data are sampled, and the corresponding industrial part DNA data is obtained. The above surface sampling data, edge sampling data, and surface sampling data and edge sampling data are input into a trained three-dimensional generation model, and the CAD model corresponding to the industrial part to be generated can be generated through the trained three-dimensional generation model.

[0087] To efficiently generate CAD models of complex and diverse industrial parts based on the three-dimensional generation model, the main problems that need to be considered in the embodiments of the present application include: Problem 1: Solve the expression problem of the CAD model, mainly to figure out what the input and output of the network should be: that is, use what kind of expression method to represent the CAD model, then encode this expression method as the input, send it into the generation model, and decode the output of the model to obtain the final CAD model. Problem 2: Research an end-to-end network. The network needs to solve the encoding and decoding problems of the CAD model, that is, how to encode the 3D model file into an input signal that the network can understand, and how to decode and reconstruct the 3D model from the output of the network. How to design the network framework and backbone modules so that they can generate specific 3D models according to the constraints. Problem 3: How to optimize the network structure so that it can efficiently generate CAD models and improve various indicators such as the yield rate of the generation model.

[0088] In view of the above problems, an optional implementation of the 3D generation model in the embodiments of the present application is that the 3D generation model realizes surface denoising and edge denoising based on the DiT architecture. The DiT architecture solves the limitations of traditional diffusion models, can efficiently model long-range dependencies, capture richer global features, while improving the modeling ability of the denoising network, solves the problem that training is not easily convergent, and provides support for accurate denoising of the 3D generation model. The structure of the 3D generation model, by regarding the vertex connection as an edge, only designs two denoising networks for the surface and the edge, avoiding the design of the vertex denoising network, reducing the computational amount, and improving the generation efficiency. Based on this model architecture, combined with industrial part DNA data, it can guide the generation of industrial parts, meet the requirements for generating complex CAD models, and improve the efficiency of CAD model generation.

[0089] Optionally, the embodiments of the present application are implemented based on the DNA of industrial parts. In the concept of the digital family tree of industrial parts, the basic attributes of parts can be mapped to similar elements in biological DNA. Exemplarily, Figure 1 A schematic diagram of the mapping representation from biological genes to industrial part genes provided by the embodiments of the present application. It can be understood that, Figure 1 It is only a schematic feasible way of industrial part DNA and does not affect the protection scope of the embodiments of the present application. It is possible to Figure 1 understand the evolution and phenotypes of industrial parts. Including Shape, Size, Material, Type, Purpose, Fitting Relationship, etc.

[0090] Shape: The shape configuration of the part is similar to the appearance characteristics of an organism. Different parts have different shapes, such as plate-shaped, T-shaped, block-shaped, etc.

[0091] Size: The size of the part, including length, width, and height, is similar to physical characteristics such as the height and weight of an organism. The sizes of industrial parts are mainly divided into large, medium, and small.

[0092] Material: The material composition of the part, such as metal, plastic, or rubber, corresponds to the concept of surface characteristics in biological DNA and reflects the inherent attributes of the part. For specific surface characteristics, the embodiments of the present application do not make specific limitations and can be determined according to actual situations, and no examples are shown in the figure.

[0093] Type: The classification of the part, that is, the specific category or lineage to which it belongs, such as gears, pipes, or bolts. Similar to species classification in biological classification, this DNA defines the category and structure of the part.

[0094] Usage: The expected function or role of a part in a system, similar to a human occupation. This attribute determines the contribution of the part to the overall system, such as functions like transmission, fastening, clamping, etc.

[0095] Assembly relationship: This DNA is a unique attribute of a part and has no direct correspondence with organisms. It describes the assembly relationship between a part and other parts in the system, especially how they are combined and interact. The assembly relationship determines the compatibility and performance of the part in the overall assembly, including press fitting, threaded connection, clamping fit, etc.

[0096] These elements together constitute the DNA of industrial parts, providing a comprehensive perspective on their characteristics, functions, and evolution in the manufacturing system. Just as biological DNA compresses the genetic information required for the development and adaptation of organisms, the DNA of industrial parts makes it possible to trace their historical evolution, modification, and future potential, supporting optimization and customization in modern manufacturing. At the same time, these DNAs can further promote the generation of industrial parts under different constraints.

[0097] The embodiments of this application draw on the concept of genes in the biological world and define the genes of industrial parts from different dimensions, corresponding to different phenotypes of industrial parts.

[0098] The embodiments of this application use the DNA of industrial parts to guide the generation of industrial parts in the three-dimensional generation model of part generation, making the generated industrial part effect more interpretable.

[0099] Optionally, Figure 2 is a schematic diagram of the architecture of a three-dimensional model generation system based on industrial part genes provided by the embodiments of this application. In Figure 1 above, the above architecture includes at least one of a data acquisition device 201, a processing device 202, and a display device 203.

[0100] It can be understood that the structure illustrated in the embodiments of this application does not constitute a specific limitation on the architecture of the three-dimensional model generation system based on industrial part genes. In other feasible embodiments of this application, the above architecture may include more or fewer components than shown, or combine certain components, or split certain components, or have different component arrangements, which can be specifically determined according to the actual application scenario and are not limited here. Figure 1 The components shown can be implemented in hardware, software, or a combination of software and hardware.

[0101] In the specific implementation process, the data acquisition device 201 may include an input / output interface or a communication interface. The data acquisition device 201 can be connected to the processing device through the input / output interface or the communication interface.

[0102] The processing device 202 can sample the surface sampling data and edge sampling data of the industrial part to be generated, obtain the corresponding industrial part DNA data, and input the above surface sampling data, edge sampling data, and the surface sampling data and edge sampling data into the trained 3D generation model, and then the CAD model corresponding to the industrial part to be generated can be generated through the trained 3D generation model.

[0103] Optionally, the processing device 202 is further configured to construct and train a 3D generation model based on industrial part DNA.

[0104] The display device 203 can also be a touch display screen or the screen of a terminal device, and is used to receive user instructions while displaying the above content to implement interaction with the user.

[0105] It should be understood that the above processing device can be implemented by a processor reading and executing instructions in a memory, or can also be implemented by a chip circuit.

[0106] In addition, the network architecture and service scenarios described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those of ordinary skill in the art know that with the evolution of the network architecture and the emergence of new service scenarios, the technical solutions provided in the embodiments of the present application are equally applicable to similar technical problems.

[0107] The technical solutions of the present application will be described in detail below with reference to specific embodiments:

[0108] Optionally, Figure 3 is a schematic flowchart of a 3D model generation method based on industrial part genes provided by an embodiment of the present application. The execution subject of the embodiment of the present application can be Figure 2 the processing device 202 therein, and the specific execution subject can be determined according to the actual application scenario. As Figure 3 shown, the method includes the following steps:

[0109] S301: Obtain the sampling data of the industrial part to be generated.

[0110] Among them, the sampling data includes surface sampling data and edge sampling data.

[0111] S302: Obtain the industrial part gene data of the industrial part to be generated.

[0112] Among them, the industrial part gene data corresponds to different phenotypes of the industrial part.

[0113] Optionally, the industrial part gene data includes at least one of shape, size, material, type, use, and assembly relationship.

[0114] Based on different partitioning dimensions, embodiments of the present application define multiple DNA data including shape, size, material, type, use, and assembly relationship to guide the accurate, reliable, and flexible generation of industrial parts.

[0115] S303: Obtain a trained three-dimensional generation model.

[0116] Among them, the three-dimensional generation model is implemented based on a diffusion transformer. The three-dimensional generation model includes a face denoising network and an edge denoising network. The trained three-dimensional generation model is obtained by training with face data training samples, edge data training samples, and industrial part gene data training samples.

[0117] S304: Input the face sampling data, edge sampling data, and industrial part gene data into the trained denoising model to generate a three-dimensional model corresponding to the industrial part to be generated through the trained denoising model.

[0118] Optionally, the sampling data is boundary representation data; correspondingly, generating a three-dimensional model corresponding to the industrial part to be generated through the trained denoising model includes: determining target boundary representation data of a tree structure corresponding to the sampling data according to the output result of the trained denoising model; generating a three-dimensional model corresponding to the industrial part to be generated according to the target boundary representation data.

[0119] In a possible implementation manner, in the model representation of the CAD model in embodiments of the present application, the "sketch and extrusion" representation method that is difficult to represent complex curves and surfaces is abandoned, and B-rep is used for representation. B-rep consists of geometric elements (faces, edges, edge, vertex adjacency matrix) with pairwise topological relationships (face-edge, edge-vertex adjacency matrix). To uniformly represent these two forms of data, topology and geometry, the geometry and topology of the B-rep model are unified into a hierarchical tree with a fixed graph topology, where node features encode geometric information, and repeated nodes (i.e., nodes with almost the same features during the generation process) implicitly encode topological information. The root node represents the entire CAD entity, so the tree has three levels: face, edge, and vertex (from top to bottom). The network input and output are the B-rep representations of such a tree structure.

[0120] Among them, in the model representation of the CAD model in embodiments of the present application, the "sketch and extrusion" representation method that is difficult to represent complex curves and surfaces is abandoned, and B-rep is used for representation. The geometry and topology of the B-rep model are unified into a hierarchical tree with a fixed graph topology, where node features encode geometric information, and repeated nodes implicitly encode topological information, having geometric accuracy and editing flexibility, and can ensure the generation efficiency of the CAD model of industrial parts.

[0121] An embodiment of the present application provides a method for generating a CAD model based on industrial part DNA. For the industrial part to be generated, surface sampling data and edge sampling data are sampled, and the corresponding industrial part DNA data is obtained. The above surface sampling data, edge sampling data, and the surface sampling data and edge sampling data are input into a trained three-dimensional generation model. Then, the CAD model corresponding to the industrial part to be generated can be generated through the trained three-dimensional generation model. Here, the three-dimensional generation model realizes surface denoising and edge denoising based on the DiT architecture. The DiT architecture solves the limitations of traditional diffusion models, can efficiently model long-range dependencies, capture richer global features, while improving the modeling ability of the denoising network, solves the problem that training is not easy to converge, and provides support for accurate denoising of the three-dimensional generation model. The structure of the three-dimensional generation model, by regarding the vertex connection as an edge, only designs two denoising networks for the surface and the edge, avoiding the design of the vertex denoising network, reducing the calculation amount, and improving the generation efficiency. Based on this model architecture, combined with industrial part DNA data, it can guide the generation of industrial parts, meet the requirements for generating complex CAD models, and improve the efficiency of CAD model generation.

[0122] Optionally, before obtaining the trained three-dimensional generation model, it further includes: obtaining the original three-dimensional generation model; obtaining surface data training samples, edge data training samples, and industrial part gene data training samples; training the original three-dimensional generation model according to the surface data training samples, edge data training samples, and industrial part gene data training samples to obtain the trained three-dimensional generation model.

[0123] Here, the embodiment of the present application pre-establishes a three-dimensional generation model and trains the model based on surface data training samples, edge data training samples, and industrial part gene data training samples to obtain a trained three-dimensional generation model that can realize accurate denoising and CAD model generation. During the training process, the introduction of industrial part DNA data can guide the generation of various industrial parts, solve the problem of part generation limitations, and improve the generation accuracy. Through reliable training, the accuracy of the three-dimensional generation model can be improved, and the yield rate of the CAD model generated by the three-dimensional generation model can be increased.

[0124] Optionally, obtaining the original three-dimensional generation model includes: constructing a surface denoising sub-model, where the surface denoising sub-model includes a surface encoder, a surface decoder, and a surface denoising network, and the surface denoising network is composed of multiple diffusion transformers; constructing an edge denoising sub-model, where the edge denoising sub-model includes an edge encoder, an edge decoder, and an edge denoising network, and the edge denoising network is composed of multiple diffusion transformers; establishing the original three-dimensional generation model according to the surface denoising sub-model and the edge denoising sub-model; where the output of the surface denoising sub-model is connected to the input of the edge denoising sub-model.

[0125] Here, the embodiments of the present application pre-train two encoders, two decoders, and two denoising networks. Specifically, the surface encoder and the edge encoder are used to encode the input data into the latent variable space to improve the processing efficiency. The surface denoising network and the edge denoising network are both composed of multiple diffusion transformers, which can improve the denoising performance of the surface denoising network and the edge denoising network. The embodiments of the present application also connect the output of the surface denoising sub-model to the input of the edge denoising sub-model to introduce the features of the surface denoising network into the edge denoising network, thereby guiding the model to better remove edge noise and improving the denoising effect and generation accuracy of the 3D generation model.

[0126] Exemplarily, Figure 4 is a schematic structural diagram of a 3D generation model provided by the embodiments of the present application. As Figure 4 shown, the 3D generation model provided by the embodiments of the present application is a sequential (Pipeline) model.

[0127] As Figure 4 shown, the encoder Encoder part includes two downsampling layer (DownBlock) modules, a residual layer (ResNetBlock) module, an intermediate processing layer (MidBlock) module, and a convolutional layer (Global Self -Attention and Convolution, GSC) module. The decoder Decoder part includes two upsampling layer (UpBlock) modules, a ResNetBlock module, a MidBlock module, and a GSC module. Among them, the MidBlock module is a Self-Attention module.

[0128] Based on Figure 4 the 3D generation model in, an alternative of the embodiments of the present application is: First, use the edges and faces expressed by the boundary to train two encoders respectively, so as to train the encoder and the decoder well. The encoder is responsible for encoding the input data into the latent variable space, where the model processes the input data more efficiently, and the decoder is responsible for decoding the network output to reconstruct the B-rep expression of the tree structure, which can be further converted into a CAD entity.

[0129] Optionally, the encoder can be a Variational Autoencoder (VAE). The VAE can generate continuous and diverse data, is stable and reliable, and the latent space has semantic meaning, improving the interpretability of the 3D generation model. The VAE can also achieve end-to-end training, can be combined with other models to meet the encoding and decoding requirements, and improve the processing efficiency of the 3D generation model of industrial parts.

[0130] Optionally, the original 3D generation model is trained according to the face data training samples, edge data training samples, and industrial part gene data training samples to obtain a trained 3D generation model, including: training the surface encoder and surface decoder according to the face data training samples to obtain a trained surface encoder and a trained surface decoder; training the edge encoder and edge decoder according to the edge data training samples to obtain a trained edge encoder and a trained edge decoder; freezing the parameters of the trained surface encoder, trained surface decoder, trained edge encoder, and trained edge decoder, and training the face denoising network and edge denoising network according to the face data training samples, edge data training samples, and industrial part gene data training samples to obtain a trained face denoising network and a trained edge denoising network; obtaining a trained 3D generation model according to the trained surface encoder, trained surface decoder, trained edge encoder, trained edge decoder, trained face denoising network, and trained edge denoising network.

[0131] Here, in the training process of the 3D generation model in the embodiments of the present application, the surface encoder / decoder and edge encoder / decoder are first trained, and then in the subsequent training process, the data processing efficiency is improved. After the surface encoder / decoder and edge encoder / decoder are trained, the parameters of the surface encoder / decoder and edge encoder / decoder are frozen to train the two sequential denoising models, namely the face denoising network and the edge denoising network, to obtain the face denoising network and edge denoising network with good denoising effects, realizing the training of the 3D generation model, with high training efficiency and strong reliability.

[0132] Optionally, the face data training samples include multiple surface noise samples, surface latent feature samples corresponding to the surface noise samples, and corresponding surface denoising samples; the edge data training samples include multiple surface noise samples, edge latent feature samples corresponding to the edge noise samples, and corresponding edge denoising samples; the industrial part gene data training samples include multiple industrial part gene data samples.

[0133] Optionally, the surface noise samples can be obtained by adding noise to the surface denoising samples and edge denoising samples.

[0134] Based on Figure 4 the 3D generation model in, training is required for denoising generation. To avoid the training instability caused by directly generating a complete tree-shaped B-rep, a sequential denoising method is adopted, first denoising the face and then denoising the edge.

[0135] To improve the inference speed, the embodiments of the present application propose to regard two points connected together as an edge, avoiding denoising the vertices and reducing the calculation amount.

[0136] In a possible implementation, based on Figure 4 the training process of the three-dimensional generation model is as follows:

[0137] Train the VAE: The edges and faces of the boundary representation correspond to the training of two different VAEs. However, the network structures of the two VAEs are the same, except that the edge or face data is used as the input during training. Taking the training of the edge VAE as an example: First, the input data is the shape features in the B-rep of CAD , which is a one-dimensional (1D) array of 3D points sampled along a parameterized curve. After passing through an encoder Encoder, the latent features in the latent space are obtained , and then decoded through a decoder Decoder to obtain the output . The encoder is trained through the mean squared error (MSE) reconstruction loss and the Kullback-Leibler divergence (KL) regularization term.

[0138] The loss function of the encoder is:

[0139]

[0140] where represents the loss function of the encoder, represents the KL regularization term, represents the MSE reconstruction loss, represents the mean and variance sampled by the encoder.

[0141] Train the surface and edge denoising network: After training the edge and face VAE models, freeze the parameters of these two models to train the sequential denoising models (i.e., the face denoising network and the edge denoising network). To more efficiently model long-range dependencies and capture richer global features, the DiT architecture is used instead of the UNet network as the core module of the denoising network. Compared with ordinary diffusion models based on the transformer architecture, the DiT architecture, through designs such as cross-attention blocks (connecting the embeddings of t and c into a sequence and adding an additional multi-head cross-attention layer after the self-attention block) and adaptive layer normalization blocks (replacing the standard layer normalization with adaptive layer normalization in the transformer block), while improving the modeling ability of the denoising network, solves the problem that the training is not easily convergent. Optionally, the denoising network has 8 DiT blocks, with 24 attention layers and 12 heads in each DiT block. The output of the denoising network will be reconstructed by the corresponding decoder for the denoised edge or face.

[0142] During training, the sampled face features are input , obtain the latent features through the surface encoder , and then , time step t, and class condition c (input if there is a class, otherwise not, corresponding to conditional generation and unconditional generation respectively) are embedded and then input into the surface denoising network . The structures of the edge denoising network and the surface denoising network are basically the same. The difference is that the edge denoising network additionally introduces the denoised surface feature F as an input, so as to guide the network to better remove the noise in the edge. Among them, the class condition c here is the industrial part DNA in the embodiments of the present application.

[0143] The loss functions of the surface and edge denoising networks are respectively:

[0144]

[0145]

[0146] Among them, is the loss function of the surface denoising network, is the loss function of the edge denoising network, and respectively represent the latent features of clean nodes and . The parameter t represents the time step, c represents the class label, is a constant hyperparameter, , which refers to the real noise added in the forward process at time step t and is the supervision signal for model training.

[0147] Based on the above embodiments, the embodiments of the present application can achieve the following technical effects:

[0148] Express the CAD model using boundary representation B-rep, unify geometry and topology into a hierarchical tree with a fixed graph topology, and master the ability to generate CAD models by training a denoising model to recover a clean tree structure from a noisy tree structure. This data representation method solves the problem that the "sketch and extrusion" representation method is difficult to generate 3D shapes with complex curves or surfaces.

[0149] In the design of the denoising network, the problem of difficulty in generating a complete model at one time is avoided through sequential denoising, and by considering connecting two vertices as an edge, there is no need to separately design a vertex denoising network, reducing the computational amount and improving the inference speed.

[0150] By introducing the DiT architecture as the core module of the denoising network, directly modeling global dependencies through self-attention and cross-attention mechanisms, the expression ability of the network is improved, the problem of difficult training of diffusion models based on the transformer structure is solved, making the geometric and topological relationships of the generated CAD models more in line with industrial reality and having a higher yield rate. And by introducing industrial part DNA, the model is more interpretable.

[0151] Figure 5 The following is a schematic structural diagram of a three-dimensional model generation device based on industrial part genes provided by an embodiment of the present application, as Figure 5 shown, the device of the embodiment of the present application includes: a sampling module 501, a first acquisition module 502, a second acquisition module 503, and a generation module 504. The three-dimensional model generation device based on industrial part genes here can be the above-mentioned processing device itself, or a chip or integrated circuit that realizes the functions of the processing device. It should be noted here that the division of the sampling module 501, the first acquisition module 502, the second acquisition module 503, and the generation module 504 is only a logical function division, and physically the two can be integrated or independent.

[0152] Among them, the sampling module is used to obtain sampling data of the industrial part to be generated, where the sampling data includes surface sampling data and edge sampling data;

[0153] The first acquisition module is used to obtain industrial part gene data of the industrial part to be generated, where the industrial part gene data corresponds to different phenotypes of the industrial part;

[0154] The second acquisition module is used to obtain a trained three-dimensional generation model, where the three-dimensional generation model is implemented based on a diffusion transformer, and the three-dimensional generation model includes a surface denoising network and an edge denoising network. The trained three-dimensional generation model is trained by surface data training samples, edge data training samples, and industrial part gene data training samples;

[0155] The generation module is used to input the surface sampling data, edge sampling data, and industrial part gene data into the trained denoising model, so as to generate a three-dimensional model corresponding to the industrial part to be generated through the trained denoising model.

[0156] Optionally, before the second acquisition module is used to obtain a trained three-dimensional generation model, the above device further includes:

[0157] A model construction module, which is used to obtain an original three-dimensional generation model;

[0158] A sample acquisition module, which is used to obtain surface data training samples, edge data training samples, and industrial part gene data training samples;

[0159] A training module for training an original three-dimensional generation model based on surface data training samples, edge data training samples, and industrial part gene data training samples to obtain a trained three-dimensional generation model.

[0160] Optionally, the model construction module is used for:

[0161] Construct a surface denoising sub-model, where the surface denoising sub-model includes a surface encoder, a surface decoder, and a surface denoising network, and the surface denoising network consists of multiple diffusion transformers;

[0162] Construct an edge denoising sub-model, where the edge denoising sub-model includes an edge encoder, an edge decoder, and an edge denoising network, and the edge denoising network consists of multiple diffusion transformers;

[0163] Establish an original three-dimensional generation model based on the surface denoising sub-model and the edge denoising sub-model;

[0164] Among them, the output of the surface denoising sub-model is connected to the input of the edge denoising sub-model.

[0165] Optionally, the training module is used for:

[0166] Train the surface encoder and the surface decoder according to the surface data training samples to obtain a trained surface encoder and a trained surface decoder;

[0167] Train the edge encoder and the edge decoder according to the edge data training samples to obtain a trained edge encoder and a trained edge decoder;

[0168] Freeze the parameters of the trained surface encoder, the trained surface decoder, the trained edge encoder, and the trained edge decoder, and train the surface denoising network and the edge denoising network according to the surface data training samples, the edge data training samples, and the industrial part gene data training samples to obtain a trained surface denoising network and a trained edge denoising network;

[0169] Obtain a trained three-dimensional generation model according to the trained surface encoder, the trained surface decoder, the trained edge encoder, the trained edge decoder, the trained surface denoising network, and the trained edge denoising network.

[0170] Optionally, the surface data training samples include multiple surface noise samples, surface latent feature samples corresponding to the surface noise samples, and corresponding surface denoising samples;

[0171] The edge data training samples include multiple edge noise samples, edge latent feature samples corresponding to the edge noise samples, and corresponding edge denoising samples;

[0172] The industrial part gene data training samples include multiple industrial part gene data samples.

[0173] The sampled data is boundary representation data;

[0174] Correspondingly, the generation module is specifically configured to:

[0175] Determine the target boundary representation data of the tree structure corresponding to the sampled data according to the output result of the trained denoising model;

[0176] Generate a three-dimensional model corresponding to the industrial part to be generated according to the target boundary representation data.

[0177] Optionally, the industrial part gene data includes at least one of shape, size, material, type, use, and assembly relationship.

[0178] Reference Figure 6 , which shows a schematic structural diagram of a three-dimensional model generation device 600 based on industrial part genes suitable for implementing the embodiments of the present disclosure. The three-dimensional model generation device 600 based on industrial part genes can be a terminal device or a server. Among them, the terminal device can include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, personal digital assistants (Personal Digital Assistant, abbreviated as PDA), tablet computers (Portable Android Device, abbreviated as PAD), portable multimedia players (Portable Media Player, abbreviated as PMP), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs and desktop computers. Figure 6 The shown three-dimensional model generation device based on industrial part genes is only an example and should not impose any limitations on the functions and usage scope of the embodiments of the present disclosure.

[0179] As Figure 6 shown, the three-dimensional model generation device 600 based on industrial part genes can include a processing device (such as a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to the program stored in the read-only memory (Read Only Memory, abbreviated as ROM) 602 or the program loaded from the storage device 608 into the random access memory (Random Access Memory, abbreviated as RAM) 603. In the RAM 603, various programs and data required for the operation of the three-dimensional model generation device 600 based on industrial part genes are also stored. The processing device 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. The input / output (I / O) interface 605 is also connected to the bus 604.

[0180] Generally, the following devices can be connected to the I / O interface 605: input devices 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. The communication devices 609 can allow the three-dimensional model generation device 600 based on industrial part genes to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 6 the three-dimensional model generation device 600 based on industrial part genes with various devices is shown, it should be understood that it is not required to implement or have all the shown devices. More or fewer devices can be alternatively implemented or had.

[0181] In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, an embodiment of the present disclosure includes a computer program product that includes a computer program carried on a computer-readable medium, and the computer program includes program codes for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network through the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above functions defined in the methods of the embodiments of the present disclosure are performed.

[0182] It should be noted that the computer-readable medium described above can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0183] The above computer-readable medium can be included in the three-dimensional model generation device based on industrial part genes; or it can exist separately and not be assembled into the three-dimensional model generation device based on industrial part genes.

[0184] The above computer-readable medium carries one or more programs, and when the one or more programs are executed by the three-dimensional model generation device based on industrial part genes, the three-dimensional model generation device based on industrial part genes is caused to execute the method shown in the above embodiments.

[0185] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0186] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks may occur in a different order than noted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system for performing the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0187] The units involved in the embodiments described in the present disclosure may be implemented in software or in hardware. Among them, the name of the unit does not constitute a limitation to the unit itself in some cases. For example, the first acquisition unit may also be described as "the unit for acquiring at least two Internet protocol addresses".

[0188] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGA), application specific integrated circuits (ASIC), application specific standard products (ASSP), system on a chip (SOC), complex programmable logic devices (CPLD), and so on.

[0189] In the context of the present disclosure, a machine-readable medium may be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0190] The three-dimensional model generation device based on industrial part genes according to the embodiments of the present application can be used to execute the technical solutions in the foregoing method embodiments of the present application. The implementation principles and technical effects are similar and will not be elaborated herein.

[0191] The embodiments of the present application further provide a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the three-dimensional model generation method based on industrial part genes in any of the foregoing items.

[0192] The embodiments of the present application further provide a computer program product, including a computer program. When the computer program is executed by a processor, it is used to implement the three-dimensional model generation method based on industrial part genes in any of the foregoing items.

[0193] In several embodiments provided by the present application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0194] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0195] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include known common general knowledge or conventional technical means in the technical field not disclosed in the present application. The specification and examples are only to be considered as exemplary.

[0196] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A three-dimensional model generation method based on industrial parts genes, characterized in that: include: Acquire sampling data of the industrial part to be generated, wherein the sampling data includes surface sampling data and edge sampling data; Acquire industrial part gene data of the industrial part to be generated, wherein the industrial part gene data corresponds to different phenotypes of the industrial part; Acquire a trained three-dimensional generative model, wherein the three-dimensional generative model is implemented based on a diffusion transformer, the three-dimensional generative model includes a surface denoising network and an edge denoising network, and the trained three-dimensional generative model is obtained by training surface data training samples, edge data training samples, and industrial parts gene data training samples; The surface sampling data, the edge sampling data and the industrial part gene data are input into a trained denoising model, so as to generate a three-dimensional model corresponding to the industrial part to be generated through the trained denoising model.

2. The method according to claim 1, characterized in that Before obtaining the trained three-dimensional generation model, the method further includes: Obtaining the original three-dimensional generative model; Obtaining surface data training samples, edge data training samples and industrial parts gene data training samples; The original three-dimensional generation model is trained according to the surface data training samples, the edge data training samples and the industrial parts gene data training samples to obtain the trained three-dimensional generation model.

3. The method according to claim 2, characterized in that The obtaining of the original three-dimensional generation model comprises: Constructing a surface denoising sub-model, wherein the surface denoising sub-model includes a surface encoder, a surface decoder and a surface denoising network, and the surface denoising network is composed of a plurality of diffusion transformers; Constructing an edge denoising sub-model, wherein the edge denoising sub-model includes an edge encoder, an edge decoder and an edge denoising network, and the edge denoising network is composed of a plurality of diffusion transformers; Establishing an original three-dimensional generation model according to the surface denoising sub-model and the edge denoising sub-model; The output of the surface denoising sub-model is connected to the input of the edge denoising sub-model.

4. The method according to claim 3, characterized in that The training of the original three-dimensional generation model according to the surface data training samples, the edge data training samples and the industrial parts gene data training samples to obtain the trained three-dimensional generation model includes: Training the surface encoder and the surface decoder according to the surface data training samples to obtain a trained surface encoder and a trained surface decoder; Training the edge encoder and the edge decoder according to the edge data training samples to obtain a trained edge encoder and a trained edge decoder; Freezing the parameters of the trained surface encoder, the trained surface decoder, the trained edge encoder and the trained edge decoder, and training the surface denoising network and the edge denoising network according to the surface data training samples, the edge data training samples and the industrial parts gene data training samples to obtain a trained surface denoising network and a trained edge denoising network; The trained three-dimensional generation model is obtained according to the trained surface encoder, the trained surface decoder, the trained edge encoder, the trained edge decoder, the trained face denoising network and the trained edge denoising network.

5. The method according to claim 4, characterized in that The surface data training samples include a plurality of surface noise samples, surface potential feature samples corresponding to the surface noise samples, and corresponding surface denoising samples; The edge data training samples include a plurality of edge noise samples, edge potential feature samples corresponding to the edge noise samples, and corresponding edge denoising samples; The industrial parts gene data training samples include a plurality of industrial parts gene data samples.

6. The method according to any one of claims 1 to 5, characterized in that: The sampling data is boundary representation data; Accordingly, generating a three-dimensional model corresponding to the to-be-generated industrial part by using the trained denoising model includes: Determining target boundary representation data of a tree structure corresponding to the number of samples according to an output result of the trained denoising model; A three-dimensional model corresponding to the to-be-generated industrial part is generated according to the target boundary representation data.

7. The method according to any one of claims 1 to 5, characterized in that: The industrial parts gene data includes at least one of shape, size, material, type, purpose, and assembly relationship.

8. A three-dimensional model generation device based on industrial parts genes, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

10. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 7 when being executed by a processor.

Images