Paper folding design method and system based on diffusion model and physical information neural network

Abstract

The application provides a paper folding design method and system based on a diffusion model and a physical information neural network, and belongs to the technical field of paper folding structures, and comprises the following steps: obtaining point cloud data of a target three-dimensional paper folding shape and corresponding text description information; generating a two-dimensional crease pattern by using a graph diffusion model according to the point cloud data and the text description information, and ensuring that the crease pattern conforms to a specified three-dimensional shape by using conditional coding; and optimizing the two-dimensional crease pattern by using a physical information neural network to obtain a final two-dimensional paper folding diagram. By combining the diffusion model and the physical constraint neural network, high-precision modeling of the crease pattern is realized, it is ensured that the generated pattern can be folded in geometry and physics, and therefore the design efficiency and accuracy are improved.

Description

Technical Field

[0001] This invention belongs to the field of origami structure technology, and particularly relates to an origami design method and system based on a diffusion model and a physical information neural network. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Origami reverse engineering is a computational technique used to derive the corresponding two-dimensional crease pattern from a known three-dimensional folded shape. This technique has wide applications in materials science, robotics, and biomedicine. By reverse-engineering the two-dimensional crease pattern, foldable structures with complex geometries can be designed, achieving goals of space saving and improved functionality. For example, origami structures can be used to manufacture retractable satellite components or miniature medical devices that can unfold from a compact state when needed. However, current reverse origami design methods present certain challenges.

[0004] Conventional origami reverse engineering relies heavily on geometric rules and manual experience. Designers manually derive the crease patterns by unfolding a three-dimensional model step by step into a flat surface. However, this method is extremely time-consuming when dealing with complex shapes and it's difficult to guarantee that the generated crease patterns conform to actual physical constraints. This experience-based design approach typically relies on existing origami theories, such as Kawasaki's Theorem and Maekawa's Theorem, which ensure that origami patterns can be folded physically flat, but they cannot automatically derive two-dimensional crease patterns for arbitrary three-dimensional structures.

[0005] In recent years, researchers have begun exploring reverse design methods that use computer algorithms to automatically generate origami patterns. Some studies have used geometric methods, such as continuous geometric models, to solve the design problem of origami structures. These methods can generate crease patterns that conform to geometric constraints to a certain extent, but they often show limitations when generating complex and diverse origami structures.

[0006] With the advancement of deep learning technology, some studies have attempted to automate origami reverse design using generative models. For example, some methods use deep generative models to generate two-dimensional crease patterns that conform to geometric and physical constraints. However, these models often require complex post-processing steps to ensure the foldability of the generated patterns, and the generated patterns still have shortcomings in terms of physical feasibility and accuracy. Summary of the Invention

[0007] To overcome the shortcomings of the existing technologies, this invention proposes an origami design method and system based on a diffusion model and a physical information neural network. This method is used to derive the corresponding two-dimensional crease patterns from known three-dimensional folding structures. By combining the diffusion model and the physical constraint neural network, it can generate diverse two-dimensional crease patterns that meet the requirements of physical folding, thus solving the shortcomings of traditional methods in generating crease patterns in terms of complexity and physical feasibility.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0009] Firstly, a paper-folding design method based on a diffusion model and a physical information neural network is disclosed, including:

[0010] Obtain point cloud data and corresponding text description information of the target 3D origami shape;

[0011] Based on the point cloud data and text description information, a two-dimensional crease pattern is generated using a graphical diffusion model, and conditional coding is used to ensure that the crease pattern conforms to the specified three-dimensional shape.

[0012] The two-dimensional crease pattern is optimized using a physical information neural network to obtain the final two-dimensional origami diagram.

[0013] Secondly, an origami design system based on a diffusion model and a physical information neural network is disclosed, including:

[0014] The data acquisition module is configured to acquire point cloud data of the target 3D origami shape and corresponding text description information.

[0015] The preliminary crease generation module is configured to: generate a two-dimensional crease pattern using a graphical diffusion model based on the point cloud data and text description information, and use conditional coding to ensure that the crease pattern conforms to the specified three-dimensional shape;

[0016] The origami generation module is configured to optimize the two-dimensional crease pattern using a physical information neural network to obtain the final two-dimensional origami diagram.

[0017] Thirdly, an electronic device is disclosed, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the origami design method based on the diffusion model and physical information neural network described above.

[0018] Fourthly, a computer-readable storage medium is disclosed for storing computer instructions, which, when executed by a processor, complete the steps of the origami design method based on the diffusion model and physical information neural network described above.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] 1. Diversity and physical feasibility of generating crease patterns: This invention combines a diffusion model and a physical constraint neural network to generate diverse two-dimensional crease patterns that meet the requirements of physical folding, thus solving the shortcomings of traditional methods in terms of complexity and physical feasibility in generating crease patterns.

[0021] 2. Enhanced design controllability: By combining 3D point cloud and multimodal conditional embedding with natural language description, this invention achieves precise control over the generation of crease patterns, which can meet specific origami design requirements.

[0022] 3. Improved computational efficiency: Compared with traditional geometric derivation and optimization methods, this invention significantly improves the efficiency of generating crease patterns and reduces the design time for complex shapes by combining diffusion generation models and neural networks.

[0023] 4. Combination of physical and geometric optimization: The physical constraint neural network ensures that the generated crease pattern is not only geometrically accurate, but also physically foldable, which solves the shortcomings of existing origami patterns in terms of physical feasibility.

[0024] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0025] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0026] Figure 1 This is a flowchart of the origami design method based on diffusion model and physical information neural network as described in Embodiment 1 of the present invention.

[0027] Figure 2 The images shown are examples of 3D point cloud origami, 2D crease diagrams, and 3D crease shape origami cranes as described in Embodiment 1 of the present invention.

[0028] Figure 3 This is a schematic diagram of the data augmentation and preprocessing results described in Embodiment 1 of the present invention.

[0029] Figure 4 This is a model diagram of the origami design method based on diffusion model and physical information neural network described in Embodiment 1 of the present invention.

[0030] Figure 5 This is a schematic diagram of the graph-based conditional diffusion model described in Embodiment 1 of the present invention.

[0031] Figure 6 This is a schematic diagram of the physical information neural network described in Embodiment 1 of the present invention.

[0032] Figure 7 This is a schematic diagram of the two-dimensional origami diagram generated as described in Embodiment 1 of the present invention.

[0033] Figure 8 The image shows an example of the results of a set of origami reverse design methods based on a diffusion model and a physical information neural network in Embodiment 1 of the present invention. Detailed Implementation

[0034] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0035] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0036] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0037] Example 1

[0038] In one or more embodiments, an origami design method based on a diffusion model and a physical information neural network is disclosed, such as... Figure 1 As shown, it includes the following steps:

[0039] Step S1: Obtain the point cloud data of the target 3D origami shape and the corresponding text description information;

[0040] During the training phase, this embodiment uses publicly available 3D classic origami models (from OrigamiSimulator) and custom-generated crease pattern data for experiments. The 3D data comes from online origami simulators (including FreeformOrigami and OrigamiSimulator), and includes 3D point clouds of the origami models and corresponding 2D crease patterns. This data is used as the training dataset for the model. The dataset includes origami models of different shapes, such as cranes, boats, and frogs, covering various folding complexities. Figure 2 The image shown is an example of a 3D point cloud origami, a 2D crease diagram, and a 3D crease shape origami crane.

[0041] The original 3D origami data and 2D origami diagram structure are augmented to enrich the training data and enhance the model's generalization ability. Augmentation methods include standard data augmentation techniques such as rotation and flipping, noise perturbation, and scale transformation. The 3D point cloud and crease patterns are rotated and mirrored along different axes to increase the directional diversity of the data. A small amount of Gaussian noise is added to the point cloud data to simulate uncertainties and data errors in real-world applications. The 3D origami model is scaled to generate origami patterns of different sizes, thereby improving the model's robustness to different scales. Vertices are deleted, edges are added, and edges are removed from the 2D origami diagram structure. Figure 3 As shown, this demonstrates the effect of a 2D origami diagram after data augmentation.

[0042] It is important to note that all data augmentation operations maintain consistency across the 3D point cloud and the corresponding 2D crease pattern to ensure that the model captures the true geometric characteristics of the origami structure.

[0043] Step S2: Based on the point cloud data and text description information, a two-dimensional crease pattern is generated using a graphical diffusion model, and conditional coding is used to ensure that the crease pattern conforms to the specified three-dimensional shape.

[0044] Step S2-1: Use multimodal conditional embedding to combine 3D point cloud data and text description to generate conditional embedding data for a graphic diffusion model, in order to control the generation of crease patterns.

[0045] like Figure 4 As shown, the 3D point cloud features are extracted using the PointNet network, and the text description is encoded using a pre-trained CLIP model to generate semantic embeddings, thus obtaining a high-dimensional feature representation of the point cloud. ,in, For the number of points, The feature dimension is used. The text description is encoded by a pre-trained CLIP model to obtain semantic embedding vectors. To achieve effective fusion between different modalities (point clouds and text), a cross-attention mechanism is introduced. Specifically, this involves text embedding... Point cloud features as a query vector As a vector of keys and values, the computation process across attention is as follows:

[0046]

[0047] in, For query vector, For key vectors, For value vectors, , , Let be a learnable linear transformation parameter matrix, T be the transpose, and d be... k This is a scaling factor for the attention score. Through this cross-attention mechanism, textual conditions act as control signals to guide the weighted aggregation of point cloud features, thereby generating conditional embedding data with fused semantics. :

[0048]

[0049] This fusion feature These embeddings are mapped into a unified latent space and used as conditional inputs to guide subsequent diffusion models in generating crease patterns that satisfy semantic constraints. These embeddings are fused through a cross-attention mechanism, mapping conditional information to the same latent space to ensure that the crease patterns generated by the diffusion process meet specified conditions.

[0050] Step S2-2: Embed random noise and conditions into the input graphic diffusion model;

[0051] In this model, 3D point cloud data and corresponding text descriptions are encoded and mapped to the same latent space through a cross-attention mechanism architecture in CLIP. The diffusion process starts with random noise and gradually diffuses in reverse to generate a 2D crease pattern, while satisfying the geometric constraints of origami, including crease types such as boundaries, mountain creases, and valley creases. Through the graphical diffusion model, 2D crease patterns that conform to the expected geometry can be generated.

[0052] Multimodal conditional embedding is used to combine 3D origami point clouds and natural language descriptions, and to control the generation process using multimodal conditions to ensure that the generated crease pattern is consistent with the input 3D geometry and meets specific origami parameter requirements.

[0053] Step S2-3: Use a graph-based diffusion model to generate a two-dimensional crease pattern based on the initial random noise.

[0054] In this embodiment, as Figure 5 As shown, a graph-based conditional diffusion model is proposed, consisting of a GATv2 (GraphAttention Network v2) graph attention network, pooling layers, and convolutional layers with a stride of 2. The graph-based conditional diffusion model generates conditional crease patterns from random noise through a graph diffusion mechanism. The specific model structure is as follows:

[0055] Each graph convolutional layer uses GATv2 to capture the local geometric relationships and topological structure of nodes in the crease pattern. GATv2 enables the model to dynamically focus on important geometric relationships by adaptively adjusting the weights of features from neighboring nodes. The core mechanism of GATv2 is to assign different weights to nodes through an attention mechanism, the specific calculation formula of which is as follows:

[0056]

[0057] in, For nodes In the Features of the layer; For nodes The set of neighboring nodes; This is a trainable weight matrix; For activation functions; For nodes and nodes The attention weights between them are calculated using the GATv2 attention mechanism. For node j at the th l Layer features. By using multi-layer GATv2 graph convolution, the model can capture multi-scale features of crease patterns, improving the geometric accuracy of generated crease patterns.

[0058] Pooling layers are used to reduce the dimensionality of the output features of graph convolutional blocks, thereby reducing computational cost while preserving key geometric information. Adaptive pooling is employed, which shrinks the graph convolutional features to a fixed size to ensure consistent feature dimensions, facilitating subsequent convolutional operations. Adaptive pooling aggregates node features to extract global information from the graph, making the model more efficient when handling complex geometric structures.

[0059] The features after GATv2 graph convolution and pooling are input into a convolutional layer with a stride of 2 to further extract global features and improve the computational efficiency of the model. The convolutional kernel size is 3×3 to capture the detailed features of the graph. This convolutional operation further reduces the dimensionality and fuses the generated crease pattern features, providing support for the final crease pattern generation.

[0060] Specifically, the diffusion model generates a two-dimensional crease pattern starting from random noise. At each diffusion step, conditional coding adjusts the generation process to ensure the pattern gradually conforms to the three-dimensional target shape and geometric constraints. The inverse formula for the diffusion process is as follows:

[0061]

[0062] in, Generate the result for step t-1. For the first Step-by-step generation of results For noise prediction networks, and These are noise parameters for the diffusion process. It is Gaussian noise. Conditional embedding of text and point cloud data.

[0063] Step S3: Optimize the two-dimensional crease pattern using a physical information neural network to obtain the final two-dimensional origami diagram.

[0064] In this embodiment, as Figure 6 As shown, a Physical Information Neural Network (PINN) is used to geometrically and physically optimize the initial two-dimensional crease pattern generated by the diffusion model. By incorporating physical and geometric loss terms, PINN ensures that the generated crease pattern conforms to the geometric requirements of actual origami and can also physically achieve folding. The specific design is as follows:

[0065] Step S3-1: The input to PINN is the two-dimensional crease pattern output by the diffusion model, including vertex positions, edge (crease) connection information, and initially generated folding angles.

[0066] Step S3-2: PINN uses a 3-layer perceptron (MLP) and a graph convolutional neural network (such as GATv2) to jointly encode nodes and edges, extracting structural features of the crease pattern. Specifically, the MLP is used to extract local attribute features, and GATv2 is used to model the topological relationships between nodes, forming high-level node embeddings and edge features.

[0067] Step S3-3: Input the above encoded features into PINN's physical optimization module. This module introduces geometric constraint loss (such as angle closure and side length consistency) and physical constraint loss (such as foldability) into the loss function. Through end-to-end optimization, the final output crease pattern is not only geometrically reasonable, but also conforms to the actual physical laws of origami.

[0068] Physical constraints ensure that the crease pattern geometrically meets the basic requirements of origami, such as edge connections at vertices and folding angles. The core of this module is the closure constraint, which guarantees that the sum of the crease angles formed by each internal vertex in the pattern is close to 2π (360 degrees), thus ensuring the crease pattern is geometrically closed. The specific formula for calculating the loss is as follows:

[0069]

[0070] in, Let be the set of all internal vertices. The angle formed by adjacent edges. This loss term ensures that the generated pattern meets geometric requirements by penalizing geometric closure deviations.

[0071] Geometric constraints ensure that the crease pattern is physically foldable. These mainly include two types of constraints: Maekawa constraints and Kawasaki constraints. Loss function. for:

[0072]

[0073] Among them, the Maekawa constraint ensures that the difference between the number of mountain folds and valley folds at each internal vertex is 2, which is one of the necessary conditions for the origami pattern to be physically foldable. The loss term is defined as follows:

[0074]

[0075] here, and These represent the number of mountain bends and valley bends at vertex i, respectively.

[0076] Kawasaki constraints are used to ensure that the pattern satisfies Kawasaki's theorem, which states that the sum of the relative folding angles at each internal vertex should be equal. The constraints are calculated as follows:

[0077]

[0078] Where n is the number of creases at the vertex. and This represents the relative crease angle. Furthermore, the loss function of the origami design method based on the diffusion model and physical information neural network includes diffusion loss, geometric loss, and physical constraint loss. The diffusion loss controls the reduction of noise during the generation process; the geometric loss is used to constrain the vertex positions and edge types of the crease pattern; and the physical constraint loss ensures that the generated crease pattern is physically foldable. The total loss function is expressed as:

[0079]

[0080] in, For the spread of loss, For geometric constraints, For physical constraints, and , , are constants, representing the weights of the loss function.

[0081] in, The specific expression is:

[0082]

[0083] Where E is the expected value. The crease pattern structure has not undergone diffusion. It is real Gaussian noise. For the noise predicted by the model, These are conditional features derived from the fusion of text and point clouds. For the first The results are generated through a step-by-step diffusion process.

[0084] In this embodiment, the loss function combines diffusion loss, geometric loss, and physical constraint loss to guide the model in generating crease patterns that conform to both origami geometry and physical folding requirements. The three networks form a unified architecture that can be trained in an end-to-end learning manner.

[0085] In this embodiment, the data path, pre-trained model path, model storage path, and initialization parameters (such as learning rate, regularization term, optimization algorithm, etc.) are configured for model training based on the origami design method using a diffusion model and a physical information neural network (Diffusion-PINN). During training, a 3D origami model and its corresponding conditional description (including point cloud and text information) are randomly selected from the training dataset and input into the multimodal conditional encoding module. The diffusion model receives initial random noise and gradually generates a preliminary 2D crease pattern under the guidance of multimodal conditions. The graph convolutional diffusion model uses a GATv2 graph convolutional network to extract multi-scale graph features and generates a crease pattern through a multi-step backdiffusion process. The generated preliminary crease pattern is input into the physical information neural network (PINN) and optimized using geometric and physical constraints to ensure that the pattern conforms to the physical and geometric rules of origami. The entire model continuously optimizes parameters through backpropagation and finally outputs a physically feasible and geometrically accurate 2D origami diagram.

[0086] In this embodiment, the model testing process is similar to the training process, using 3D point cloud data and condition descriptions. During the testing phase, parameters such as the model path, test data path, and output path are set. A preliminary crease pattern is generated using a diffusion model and then fed into PINN for optimization, ultimately outputting a 2D origami diagram that meets the requirements. The physical feasibility and geometric accuracy of the generated 2D crease pattern are verified by simulating the folding process.

[0087] The diffusion model uses random noise as input and generates an initial two-dimensional crease pattern through a multi-step back-diffusion process. The Physical Information Neural Network (PINN) receives these initial patterns and optimizes them using geometric and physical constraints.

[0088] The model uses multimodal conditional encoding (combining 3D point cloud and text description) to guide the diffusion process. In each training iteration, a 3D origami model and its conditional description are randomly selected from the dataset. The model optimizes the generated pattern through backpropagation, gradually making it conform to the target geometric and physical requirements.

[0089] The physical information neural network optimizes the initial crease pattern through geometric and physical loss terms, ensuring that the pattern is physically foldable and geometrically accurate. During the optimization process, the loss function incorporates geometric closure, Maekawa and Kawasaki theorems, and other principles.

[0090] During training, in one embodiment, the diffusion model and PINN are jointly trained and optimized. The model updates its parameters in each training epoch to ensure that the generated crease patterns gradually meet the origami design requirements. The AdamW optimizer is used for training, and a learning rate decay strategy is set to improve the convergence speed.

[0091] In this embodiment, as Figure 7 As shown, the generated 2D origami diagram and its performance in actual folding are demonstrated. The test results include the geometric accuracy and physical feasibility of the crease pattern, as well as the degree of consistency between the actual folded 3D shape and the original input shape.

[0092] Step S4: File Storage and Visualization. The generated 2D origami diagram is stored in the computer device's memory, and the generation process of the crease pattern and its physical folding effect are displayed through visualization tools. Users can choose different visualization modes to more intuitively understand the folding process and result of the generated pattern.

[0093] This origami reverse design system based on diffusion model and physical information neural network provides an efficient and reliable method to generate complex two-dimensional crease patterns, which is widely applicable to materials science, robotics, and biomedical engineering.

[0094] like Figure 8 As shown, the technical solution of this embodiment ensures the accuracy of the origami reverse process design, where the blue vertices represent the original origami model diagram structure and the yellow diagram represents the origami pattern diagram structure derived from multiple sampling inferences.

[0095] Example 2

[0096] In one or more embodiments, an origami design system based on a diffusion model and a physical information neural network is disclosed, specifically including:

[0097] The data acquisition module is configured to acquire point cloud data of the target 3D origami shape and corresponding text description information.

[0098] The preliminary crease generation module is configured to: generate a two-dimensional crease pattern using a graphical diffusion model based on the point cloud data and text description information, and use conditional coding to ensure that the crease pattern conforms to the specified three-dimensional shape;

[0099] The origami generation module is configured to optimize the two-dimensional crease pattern using a physical information neural network to obtain the final two-dimensional origami diagram.

[0100] Example 3

[0101] This embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps of the origami design method based on the diffusion model and physical information neural network described above.

[0102] Example 4

[0103] This embodiment provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, complete the steps of the origami design method based on the diffusion model and physical information neural network described above.

[0104] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0105] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0106] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0107] The descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0108] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A paper-folding design method based on a diffusion model and a physical information neural network, characterized in that, include: Obtain point cloud data and corresponding text description information of the target 3D origami shape; Based on the point cloud data and text description information, a two-dimensional crease pattern is generated using a graphical diffusion model, and conditional encoding is used to ensure that the crease pattern conforms to a specified three-dimensional shape. The conditional encoding employs multimodal conditional embedding, combining the three-dimensional point cloud data and text description to generate conditional embedding data for the graphical diffusion model. Specifically: 3D point cloud features are extracted using the PointNet network, and text descriptions are encoded using a pre-trained CLIP model to generate semantic embeddings. The extracted 3D point cloud features and semantic embeddings are fused through a cross-attention mechanism to map conditional information to the same latent space to obtain conditional embedding data. The two-dimensional crease pattern is optimized using a physical information neural network to obtain the final two-dimensional origami diagram.

2. The origami design method based on diffusion model and physical information neural network as described in claim 1, characterized in that, The input to the graphical diffusion model is random noise and conditional embedding data.

3. The origami design method based on diffusion model and physical information neural network as described in claim 1, characterized in that, The graph diffusion model includes a graph attention network, pooling layers, and convolutional layers with a stride of 2; The graph attention network assigns different weights to nodes through an attention mechanism, expressed as follows: in, For nodes In the Features of the layer; For nodes The set of neighboring nodes; is a trainable weight matrix; For activation functions; For nodes and nodes The attention weights between them are calculated using the GATv2 attention mechanism. For node j at the th l Characteristics of the layer.

4. The origami design method based on diffusion model and physical information neural network as described in claim 3, characterized in that, The graphic diffusion model generates a two-dimensional crease pattern starting from random noise. The generation process is adjusted through conditional coding. The inverse formula for the diffusion process is as follows: in, Generate the result for step t-1. For the first Step-by-step generation of results For noise prediction networks, and These are noise parameters for the diffusion process. It is Gaussian noise. Conditional embedding of text and point cloud data.

5. The origami design method based on diffusion model and physical information neural network as described in claim 1, characterized in that, The physical information neural network includes a 3-layer perceptron and a graph convolutional neural network; the physical information neural network uses a loss function to control the reduction of noise during the generation process, constrains the vertex position and edge type of the crease pattern, and ensures that the generated crease pattern is physically foldable.

6. The origami design method based on diffusion model and physical information neural network as described in claim 5, characterized in that, The loss function includes diffusion loss, geometric loss, and physical constraint loss; The loss function is expressed as: in, For the spread of loss, For geometric loss, For physical constraint loss, and These are constants, representing the weights of the loss function; The specific expression for diffusion loss is: Where E is the expected value. The crease pattern structure has not undergone diffusion. It is real Gaussian noise. For the noise predicted by the model, F cond Conditional features derived from the fusion of text and point clouds. For the first The results are generated through a step-by-step diffusion process.

7. An origami design system based on a diffusion model and a physical information neural network, characterized in that, include: The data acquisition module is configured to acquire point cloud data of the target 3D origami shape and corresponding text description information. The preliminary crease generation module is configured to: generate a two-dimensional crease pattern using a graphic diffusion model based on the point cloud data and text description information, and use conditional encoding to ensure that the crease pattern conforms to a specified three-dimensional shape; the conditional encoding uses multimodal conditional embedding to combine the three-dimensional point cloud data and text description to generate conditional embedding data for the graphic diffusion model, specifically: 3D point cloud features are extracted using the PointNet network, and text descriptions are encoded using a pre-trained CLIP model to generate semantic embeddings. The extracted 3D point cloud features and semantic embeddings are fused through a cross-attention mechanism to map conditional information to the same latent space to obtain conditional embedding data. The origami generation module is configured to optimize the two-dimensional crease pattern using a physical information neural network to obtain the final two-dimensional origami diagram.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, complete the origami design method based on a diffusion model and a physical information neural network as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the origami design method based on a diffusion model and a physical information neural network as described in any one of claims 1-6.

Images