AI and vector bundle-based CAD large model method and system

Abstract

The application discloses a CAD large model method and system based on AI and vector bundle, and belongs to the field of computer-aided design. The method comprises the following steps: S1, constructing a mathematical model of manifold space, and mapping three-dimensional information of an object to a linear space through a vector bundle; S2, collecting drawing data of multiple industries, and constructing a drawing data set of multiple industries; S3, pre-training the drawing data set of multiple industries through the mathematical model of manifold space, and extracting feature template tensors corresponding to drawings of each industry; S4, based on the feature template tensors, using a large model with an embedded M-MoE architecture to encode, decode and cyclically optimize and update parameters; and S5, mapping the manifold tensor output in the previous step to a three-dimensional space, generating CAD drawings and reports, and cyclically optimizing and outputting results through a loss function. The application constructs a CAD large model based on artificial intelligence and vector bundle mathematical theory, and realizes intelligent drawing generation, automatic auditing, language interactive editing and full life cycle management.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design, and in particular to a method and system for constructing large CAD models based on artificial intelligence (AI) and vector bundle mathematical theory, which is used to realize intelligent drawing generation, automatic review, interactive language editing and full life cycle management functions. Background Technology

[0002] Currently, the core of CAD drawing software is parametric modeling, which includes various methods such as polygon mesh modeling, freeform surface modeling, and parametric feature modeling. Among these, parametric feature modeling is the primary method used in the production and manufacturing process.

[0003] The core of parametric feature modeling in CAD drawing software includes a geometric kernel and a parametric solver. The core of the geometric kernel consists of a mathematical expression function library and a geometric algorithm library, primarily including Siemens' Para-Solid and Dassault Systèmes' ACIS (both closed-source geometric kernels) and the open-source OCC geometric kernel from France. However, these CAD drawing software products still have the following shortcomings:

[0004] 1. Lack of intelligent reasoning ability: Existing CAD drawing software lacks the ability to think and reason, cannot make rule judgments or review two-dimensional and three-dimensional drawings, and is still in a low-level drawing tool stage.

[0005] 2. Insufficient interactivity: Existing CAD drawing software lacks voice-based editing capabilities, requiring users to manually modify and edit drawings.

[0006] 3. Inability to automatically generate drawings and reports: Existing CAD drawing software does not have the ability to automatically generate drawings and reports.

[0007] The aforementioned shortcomings result in significant manpower and resources being consumed in the review, editing, and generation of drawings, leading to high costs, low efficiency, and a high error rate. There is an urgent need to improve and innovate existing CAD drawing software to provide a more intelligent CAD drawing design solution. Summary of the Invention

[0008] The purpose of this invention is to overcome the problems existing in the design of existing CAD drawings and to provide a method and system for large CAD models based on AI and vector bundles.

[0009] The objective of this invention is achieved through the following technical solution:

[0010] Firstly, a method for creating large CAD models based on AI and vector bundles is provided, including the following steps:

[0011] S1. Construct a mathematical model of manifold space and map the three-dimensional information of the object to linear space through vector bundles;

[0012] S2. Collect drawing data from multiple industries and construct a multi-industry drawing dataset;

[0013] S3. Pre-train the multi-industry drawing dataset using the manifold space mathematical model, and extract the feature template tensor corresponding to each industry drawing;

[0014] S4. Based on the feature template tensor, a large model with an embedded M-MoE architecture is used for encoding, decoding, and iterative optimization to update parameters.

[0015] S5. Map the manifold tensor output from the previous step to three-dimensional space to generate CAD drawings and reports, and iteratively optimize the output results using a loss function.

[0016] In some embodiments, mapping the three-dimensional information of the object to a linear space via a vector bundle includes:

[0017] By representing the structure, annotation, production information, and full lifecycle data of objects through vector bundles, the local homeomorphism of the manifold space is mapped to the linear Euclidean space R. n .

[0018] In some embodiments, the collection of drawing data from multiple industries and the construction of a multi-industry drawing dataset includes:

[0019] Component features are extracted using a few-sample detection algorithm;

[0020] A feature dataset was established by comparing the differences between CAD drawings using the Hungarian algorithm.

[0021] In some embodiments, step S3 includes:

[0022] Substituting the feature dataset into the manifold space mathematical model for pre-training yields the homologous template tensor, the three-dimensional structure template tensor corresponding to the i-th level vector bundle, and the property template tensor corresponding to the i-th level vector bundle.

[0023] In some embodiments, the large model includes an encoder, an M-MoE layer, and a decoder. The encoder uses four layers of multi-head attention to compute the feature template tensor and applies 48 layers of blocks. The M-MoE layer includes 128 M-MoE architectures, each of which includes one shared expert and 128 routing experts.

[0024] Secondly, a large CAD model system based on AI and vector bundles is provided, including:

[0025] The manifold space mathematical model construction module is used to construct a manifold space mathematical model, mapping the three-dimensional information of an object to a linear space through vector bundles;

[0026] The drawing dataset construction module is used to collect drawing data from multiple industries and build multi-industry drawing datasets.

[0027] The feature template tensor extraction module is used to pre-train the multi-industry drawing dataset using the manifold space mathematical model to extract the feature template tensors corresponding to the drawings of each industry.

[0028] The large model optimization module is used to encode, decode, and iteratively optimize and update parameters of a large model based on the feature template tensor using an embedded M-MoE architecture.

[0029] The CAD drawing generation module maps the manifold tensors output by the large model optimization module to three-dimensional space, generates CAD drawings and reports, and iteratively optimizes the output results through a loss function.

[0030] It should be further noted that the technical features corresponding to the above embodiments can be combined or substituted with each other to form new technical solutions without conflict.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] 1. This invention constructs a mathematical model using manifold space and vector bundle theory, combined with a large model based on multi-industry datasets and the M-MoE architecture, to achieve automatic drawing generation, intelligent review, interactive language editing, and full lifecycle management. This improves the intelligence and interactivity of CAD drawing software, significantly increasing efficiency and reducing costs.

[0033] 2. This invention pre-trains the multi-industry drawing dataset using the manifold space mathematical model, extracts the feature template tensors corresponding to the drawings of each industry, uses a few-shot detection algorithm to extract component features, and performs differential comparison of CAD drawings using the Hungarian algorithm, reducing computing power and saving costs, while taking into account both accuracy and computing speed. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating a large CAD model method based on AI and vector bundles, as shown in an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram illustrating the principle of constructing a mathematical model of a manifold space according to an embodiment of the present invention;

[0036] Figure 3 This is a schematic diagram of the overall architecture of the large model shown in an embodiment of the present invention;

[0037] Figure 4 This is a schematic diagram illustrating the generation of a large model report, as shown in an embodiment of the present invention.

[0038] Figure 5This is a schematic diagram illustrating the generation of large model drawings according to an embodiment of the present invention;

[0039] Figure 6 This is a schematic diagram illustrating the design functions of a large model as shown in an embodiment of the present invention.

[0040] Figure 7 This is a schematic diagram of the tensor at the manifold point as shown in an embodiment of the present invention. Detailed Implementation

[0041] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] It should be noted that the defects in the solutions in the prior art are all the results of the inventors' practice and careful research. Therefore, the discovery process of the above problems and the solutions proposed by the embodiments of this application in the following text should be the inventors' contributions to this application in the process of invention and creation, and should not be understood as technical content known to those skilled in the art.

[0043] In view of the technical problems pointed out in the background art, the present invention provides the following embodiments:

[0044] In one exemplary embodiment, reference is made to Figure 1 This invention provides a method for large-scale CAD models based on AI and vector bundles. It introduces new mathematical tools, placing all information, including 2D or 3D CAD text, drawings, and data, within a manifold space. Since real-world designs cannot be fully represented by linear spaces, and considering that computers can only process linear numbers (0s and 1s), it is necessary to map the manifold space locally homeomorphically to linear Euclidean space. Therefore, a theoretical tool for homeomorphizing the manifold space to Euclidean space is needed. This invention uses vector bundles and manifold tensor theory in manifold space to handle the complex representation problem of drawing.

[0045] Step S1 specifically includes:

[0046] By representing the structure, annotation, production information, and full lifecycle data of objects through vector bundles, the local homeomorphism of the manifold space is mapped to the linear Euclidean space R. nBased on the conclusions regarding object representation in the three-dimensional world, a manifold basis space M is constructed from the spatial information of the object's three-dimensional space. This space is then inversely mapped to a vector bundle E1 (representing the object's three-dimensional structure), smoothly mapped from E1 to E2 (representing all annotation information such as object rules), smoothly mapped from E2 to E3 (representing the object's manufacturing and construction information), and smoothly mapped from E1 and E3 to E4 (representing the object's entire lifecycle information). i (The value of i can be adjusted as needed) Homeomorphic to linear Euclidean space R n The constructed mathematical model of the manifold space is as follows: Figure 2 As shown. In practical applications, the vector bundle tool of the manifold space is selected according to the requirements of the object's drawing and report, etc., with the M manifold base space as the vector bundle space E. i Given a basis vector bundle, the following operations can be performed:

[0047] Direct summation, scalar multiplication, existence dual space, tensor product, algebraic geometry operations, and topological operations.

[0048] If the above operational rules are satisfied, inner product dot product can be performed, which can be directly used as input for various models. Define the local coordinate system ₯(p,Ψ) of the manifold space, mapping it to R. n The space is transformed by the general linear group GL(n), where P is any manifold point in the manifold space, and Φ3(π) -1 ), Φ2 (π -1 ), Φ1 (π -1 The bundle space mapping shown is finally transformed to Euclidean space by a general linear group transformation.

[0049] A tensor on the manifold space is a partially linear function θ, representing a (r,s) type tensor at a point P on the manifold. This tensor represents, for example,... Figure 7 As shown.

[0050] Among them, T P *M represents the dual tensor at point P on the manifold, and T P M represents the tensor at point P on the manifold. This (r,s) type tensor satisfies the operational rules of vector bundles and can be used for subsequent data analysis and processing.

[0051] Furthermore, the E1 space is used as the function and geometric algorithm representation space of the geometric kernel to satisfy the two-dimensional or three-dimensional representation of the object world.

[0052] Step S2 specifically includes:

[0053] We collect basic text data and drawing data from dozens of industrial sectors, including oil and gas chemical industry, biopharmaceutical industry, tobacco industry, automotive industry, semiconductor electronics industry, robotics and drone industry, industrial design industry, e-cigarette industry, auto parts industry, and construction machinery industry, and store them in a local database. As needed, we use vector database tools and feature extraction tools to construct feature datasets for each industry based on the multimodal object drawing information and full lifecycle data from the database.

[0054] Due to the format differences among the databases in various industries, detailed information from the databases is interpreted to obtain a feature dataset. A few-shot detection algorithm is then used to extract component features, including component identification, line clustering, and legend matching, to achieve automatic identification of key components in the drawings. The few-shot detection algorithm includes the following steps:

[0055] (1) Based on a deep feature extraction network, region candidates are generated for CAD drawings, and a small sample support set is used to construct the feature embedding space between component classes;

[0056] (2) Using a metric learning framework, each candidate region is mapped to a high-dimensional feature vector, and a few-sample classification is achieved by matching the distance function to obtain the category label and location information of each component in the drawing;

[0057] (3) Further construct a connected structure graph for the identified component set, and apply density clustering algorithm to perform spatial clustering of the lines in the graph to identify the connection relationship;

[0058] (4) Combine the legend templates in the graphic element library and perform accurate matching of legend symbols in the drawing based on image matching and geometric correction strategies.

[0059] Furthermore, a feature dataset is established by performing differential comparison of CAD drawings using an improved Hungarian algorithm. The improved Hungarian algorithm includes the following steps:

[0060] (1) Convert the two CAD drawings to be compared into graphic structures respectively. and ,in Represents a set of component nodes. Indicates the topological connections between components;

[0061] (2) Calculate the matching cost between each pair of nodes. The cost function comprehensively considers visual feature distance, spatial location deviation, and local topological similarity. Specifically, the cost matrix can be defined as follows:

[0062]

[0063] in For adjustable weighting parameters, Indicates the Euclidean distance of the visual embedding of the element. This represents the difference in node position coordinates. The Jaccard distance represents the local topology of a node;

[0064] (3) Introduce a fault-tolerant mechanism to improve the traditional Hungarian algorithm, including adding empty nodes to match missing or newly added components to avoid unreasonable forced matching; dynamically adjusting the cost threshold to adapt to changes in component layout and labeling differences; and combining graph convolutional neural networks to calculate node embedding features to improve topology matching accuracy.

[0065] (4) Finally, the optimal matching pair is output, and the set of unmatched nodes is extracted as candidate element pairs for differences between drawings, and a differential feature map is automatically generated.

[0066] Step S3 specifically includes:

[0067] Based on vector bundle and tensor theory of manifold spaces, the vector bundle E is represented by four sets of vector bundle bases of the manifold basis space M. i The tensor of the CAD information of the homologous objects, the i-th vector bundle, and the representation corresponding to the i-th level.

[0068] After the mathematical model of the manifold space is established, the various feature datasets established in step S2 are substituted into the pre-training of the mathematical model of the manifold space to obtain the homologous template tensor, the corresponding i-th vector bundle, and the corresponding i-th level representation.

[0069] The overall architecture of the large model of this invention is as follows: Figure 3 As shown, it includes an encoder, an M-MoE layer, and a decoder. In step S3, multiple template datasets are pre-trained. The large model is input with multimodal CAD drawing information (drawings, text, IoT data, etc.) and the feature dataset established in step S2. The output model becomes four corresponding vector bundles.

[0070] The core of the large model encoder is the improved Transformer. The encoder uses four layers of multi-head attention to compute the feature template tensor and applies 48 layers of blocks. It outputs three fourth-level tensors as inputs to the next layer of decoding. Specifically, the multiple tensors obtained in step S3 are used as inputs to the encoder for homology comparison, structural comparison, and physicochemical property comparison learning. The encoder iteratively adjusts the parameters of the above multiple tensors, and after learning and adjustment, it obtains new multiple tensors as inputs to the M-MoE layer.

[0071] The large model adds 128 M-MoE layers. Each MoE architecture includes one shared expert and 128 routing experts. Each drawing token activates multiple routing experts, constructing a multi-intelligence platform that jointly provides rule knowledge and industry-specific drawings for various industries. In practical applications, the number of expert layers can be adjusted according to industry needs, allowing different industries to call upon expert agents from different industries. Furthermore, each agent represents the corresponding knowledge and expert for each industry; calling an expert agent means calling the corresponding judgment rules and specifications.

[0072] Step S6 specifically includes:

[0073] The multiple tensors output from the M-MoE layer are used as input. The decoder performs homology adjustment, structural adjustment, and physicochemical property adjustment learning, iteratively adjusting the parameters of the multiple tensors to obtain new manifold space tensors, which serve as inputs for the next step of mapping to the three-dimensional linear space. Specifically, the tensors output from step S5 are used as input to update the structure of the four-level structure, and the physicochemical properties are used as constraints (such as minimizing the potential energy and maximizing the stability of each level) to update the geometric kernel and constraints. The parameters are updated iteratively through eight blocks of optimization.

[0074] Further, in step S7, the manifold space tensor output from the previous step is projected onto the three-dimensional spatial structure data and related functional data, converted into two-dimensional or three-dimensional drawings and reports, and then iteratively optimized. Specifically, the three-dimensional structure output is optimized using a minimum loss function, and this process is continuously iterated. In this embodiment, five rounds of iterations are designed, repeatedly executing the encoding and decoding process. After multiple adjustments, end-to-end two-dimensional or three-dimensional drawings and reports are output. Furthermore, a language interaction mechanism is introduced during the iterative optimization process, allowing users to interactively edit and fine-tune the generated drawings and reports using natural language, such as adjusting structural details, modifying dimension annotations, or adding explanatory text, thereby achieving human-computer collaborative optimization design output. CAD industry reports generated based on large CAD models, such as... Figure 4 As shown, the drawings generated based on the large CAD model are as follows: Figure 5 As shown.

[0075] In another exemplary embodiment, a large CAD model system based on AI and vector bundles is provided, comprising:

[0076] The manifold space mathematical model construction module is used to construct a manifold space mathematical model, mapping the three-dimensional information of an object to a linear space through vector bundles;

[0077] The drawing dataset construction module is used to collect drawing data from multiple industries and build multi-industry drawing datasets.

[0078] The feature template tensor extraction module is used to pre-train the multi-industry drawing dataset using the manifold space mathematical model to extract the feature template tensors corresponding to the drawings of each industry.

[0079] The large model optimization module is used to encode, decode, and iteratively optimize and update parameters of a large model based on the feature template tensor using an embedded M-MoE architecture.

[0080] The CAD drawing generation module maps the manifold tensor output by the large model optimization module to three-dimensional space to generate CAD drawings and reports, and iteratively optimizes the output results through a loss function. In addition, the CAD drawing generation module also includes a language interaction submodule, which is used to interactively edit the generated CAD drawings and reports based on natural language commands, so as to modify and adjust the drawing structure, dimensions, symbol annotations and other contents, thereby improving human-computer interaction efficiency and design flexibility.

[0081] The large model platform and design functions built upon this system include, for example... Figure 6 As shown, this system can automatically generate, review, and edit drawings and reports using interactive language. This saves costs and improves efficiency. Furthermore, the system's large-scale model platform enables production management.

[0082] The above detailed embodiments are a description of the present invention. It should not be considered that the specific embodiments of the present invention are limited to these descriptions. For those skilled in the art, several simple deductions and substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the protection scope of the present invention.

Claims

1. A method for generating a CAD drawing using an AI and vector bundle based CAD large model, characterized in that, The method comprises the following steps: S1, constructing a manifold space mathematical model, mapping the three-dimensional information of an object to a linear space through a vector bundle; S2, collecting drawing data of multiple industries to construct a multi-industry drawing data set; S3, pre-training the multi-industry drawing data set through the manifold space mathematical model, and extracting feature template tensors corresponding to drawings of each industry; S4, based on the feature template tensors, using a large model embedded with an M-MoE architecture for encoding, decoding and cyclic optimization of updating parameters; S5, mapping the manifold tensor output in the previous step to a three-dimensional space to generate CAD drawings and reports, and cyclically optimizing the output results through a loss function. 2.The method for generating a CAD drawing using an AI and vector bundle-based CAD large model according to claim 1, wherein, The mapping of the three-dimensional information of the object to the linear space comprises: The structure of an object is represented by a vector bundle, and the structure, label, production information and life cycle data of the object are mapped from the manifold space to the linear Euclidean space R n . 3.The method for generating a CAD drawing using an AI and vector bundle-based CAD large model according to claim 1, wherein, The collection of drawing data of multiple industries to construct a multi-industry drawing data set comprises: Extracting element features using a few-shot detection algorithm; Performing differential comparison of CAD drawings through the Hungarian algorithm to establish a feature data set. 4.The method for generating a CAD drawing using an AI and vector bundle-based CAD large model according to claim 3, characterized in that, The step S3 comprises: Substituting the feature data set into the manifold space mathematical model to pre-train a homologous template tensor, a three-dimensional structure template tensor corresponding to the i-th level vector bundle, and a property template tensor corresponding to the i-th level vector bundle; i is 1, 2, 3, or 4, wherein the first level vector bundle represents the three-dimensional structure information of the object, the second level vector bundle represents the labeling information of the object, the third level vector bundle represents the production and manufacturing information of the object, and the fourth level vector bundle represents the whole life cycle information of the object; the mapping relationship between the vector bundles of different levels is as follows: The first level vector bundle is smoothly mapped to the second level vector bundle, and then the second level vector bundle is smoothly mapped to the third level vector bundle, and then the first level vector bundle and the third level vector bundle are smoothly mapped to the fourth level vector bundle. 5.The method for generating a CAD drawing using an AI and vector bundle-based CAD large model according to claim 1, wherein, The large model comprises an encoder, an M-MoE layer, and a decoder, wherein the encoder adopts four-layer multi-head attention to calculate the feature template tensor, and applies 48 layers of blocks; the M-MoE layer comprises 128 M-MoE architectures, and each M-MoE architecture comprises one shared expert and 128 routing experts.

6. A system for generating a CAD drawing using an AI and vector bundle based CAD large model, characterized by, It comprises: A manifold space mathematical model construction module for constructing a manifold space mathematical model and mapping the three-dimensional information of an object to a linear space through a vector bundle; A drawing data set construction module for collecting drawing data of multiple industries to construct a multi-industry drawing data set; A feature template tensor extraction module for pre-training the multi-industry drawing data set through the manifold space mathematical model and extracting feature template tensors corresponding to drawings of each industry; A large model optimization module for encoding, decoding and cyclic optimization of updating parameters based on the feature template tensors using a large model embedded with an M-MoE architecture; A CAD drawing generation module for mapping the manifold tensor output by the large model optimization module to a three-dimensional space to generate CAD drawings and reports, and cyclically optimizing the output results through a loss function.

Images