Ship profile modeling method and system based on AI
By generating 3D models of ship profiles using AI-based convolutional neural networks and generative adversarial networks, the problems of low efficiency and error-proneness in traditional modeling are solved, achieving efficient and accurate profile modeling and reducing manual intervention and time costs.
Patent Information
- Application Number
- CN202511207404.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-27
- Publication Date
- 2025-12-16
AI Technical Summary
Traditional ship profile modeling relies on manual operation, which is inefficient, error-prone, and has poor applicability. It lacks intelligent processing capabilities, making it difficult to quickly respond to new ship types or special structures, and historical data is not fully utilized.
An AI-based approach is adopted, which extracts profile features through convolutional neural networks, generates 3D profile models using generative adversarial networks, and combines profile databases and generator discriminators for training and optimization to achieve fully automated modeling.
It enables efficient and accurate 3D modeling of profiles, reduces manual intervention and errors, improves modeling speed and quality, and provides a standardized modeling process, saving time and labor costs for ship design.
Smart Images
Figure CN121145338A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of shipbuilding technology, and in particular relates to an AI-based method and system for modeling ship profiles. Background Technology
[0002] In ship structural design, structural profiles, as a core component of the hull skeleton, are a crucial and extremely labor-intensive part of the entire design process, and a core element of ship structural design. The traditional design process, based on manually drawn CAD 2D drawings followed by 3D modeling using software, is outdated, highly dependent on manual operation, and suffers from the following significant drawbacks:
[0003] 1. Low efficiency: Manual modeling is time-consuming and prone to errors. Complex surfaces require repeated adjustments. Especially for a large ship containing tens of thousands of profiles, the modeling cycle can last for months, which seriously restricts the design and construction progress of the ship.
[0004] 2. High dependence on experience: The accuracy of modeling depends heavily on the engineer's personal technical level and experience. Novice engineers are prone to making modeling errors due to misreading drawings or operational mistakes, resulting in a low degree of standardization.
[0005] 3. Poor applicability: It lacks sufficient flexibility and adaptability, and lacks the ability to quickly respond to modeling new ship types or special structures.
[0006] 4. Low data utilization: A large number of historical two-dimensional drawings, which are valuable data assets, have not been fully explored for optimizing new ship models. They are lost as engineers change, making them difficult to pass on effectively.
[0007] In summary, while existing design models utilize semi-automated modeling tools to some extent, they primarily rely on engineers' human resources, lacking intelligent processing capabilities and struggling to respond promptly to new and complex scenarios. With the rapid development of AI technology in computer vision and generative modeling, there is an urgent need for a novel modeling technology solution to achieve a fully automated, high-precision, and intelligent modeling process for ship profiles, transforming two-dimensional drawings into three-dimensional models. Summary of the Invention
[0008] In view of the shortcomings of the prior art described above, this application provides an AI-based method and system for ship profile modeling. This method can reduce manual intervention and achieve an efficient, high-precision, and adaptive ship profile modeling process.
[0009] Firstly, this application provides an AI-based method for modeling ship profiles, comprising at least the following steps:
[0010] S1: Based on the historical CAD drawings of profiles, summarize the modeling elements of profiles, extract the structured data of profiles, and build a profile database;
[0011] S2: Use convolutional neural networks to extract profile features from the profile database, including the geometric features and topological relationships of the profiles;
[0012] S3: Based on the extracted profile features, a three-dimensional profile model is obtained by training a generative adversarial network.
[0013] In some implementations, the profiles include straight profiles and twisted profiles;
[0014] The features of the straight profile include: boundary, specifications and material, end cut, ball head direction and trajectory line;
[0015] The characteristics of the twisted profile include: boundary, specifications and material, end cut, ball head direction, trajectory line, profile support surface and sheet support surface angle;
[0016] Step S1 summarizes the modeling elements of the profile during the execution process, determines the mapping relationship between the profile features and the CAD drawings, identifies the profile's boundary, specifications, material, end cut, ball head direction and trajectory line, and classifies the trajectory line as a straight line, arc or ellipse; for twisted profiles, it further identifies the angle between the profile support surface and the sheet support surface.
[0017] In some implementations, step S2 includes:
[0018] S201: Prepare profile training data based on profile features. Each training data structure includes: profile cross-sectional shape image - profile end-cut shape image - profile external dimensions;
[0019] S202: Based on the data structure in S201, a convolutional neural network is used to extract the geometric features and topological relationships of the profiles.
[0020] In some implementations, preparing the profile training data in step S201 includes: vectorizing the image of the profile to generate a grayscale image; and normalizing the external dimensions of the profile to eliminate the influence of dimensions, so as to obtain a unified data structure for the input of the convolutional neural network.
[0021] In some implementations, the method further includes a step of training a convolutional neural network: the convolutional neural network employs a multi-scale structure and introduces an attention mechanism to improve the accuracy of profile feature extraction.
[0022] In some implementations, the profile feature output is a feature vector containing "curvature change sequence - cross-sectional shape classification - end-cut classification".
[0023] In some implementations, step S3 includes:
[0024] S301: Design a generative adversarial network: The generator takes the profile features extracted by the convolutional neural network as input and outputs a 3D voxel model of the profile; the discriminator takes the real 3D voxel model as input and outputs the realism probability and normality score.
[0025] S302: Rule Embedding and Adversarial Training: Embed physical feasibility loss, ship specification loss and feature consistency loss in adversarial training, and train the model.
[0026] In some implementations, step S3 further includes S303: the step of converting the voxel model into a solid model, wherein the system iteratively corrects the converted solid model to make the model conform to the latest design specifications.
[0027] In some implementations, step S3 further includes S304: a topology optimization step of converting the voxel model into a solid model.
[0028] Secondly, this application provides an AI-based modeling system for ship profiles, comprising:
[0029] The profile feature data extraction module is used to extract profile feature data from CAD drawings;
[0030] The model training module integrates the CNN feature extraction submodule and the GAN model training submodule, and is used to train the profile feature extraction model and the 3D generation model.
[0031] The 3D model generation module is used to generate 3D models of ship profiles based on the trained GAN model.
[0032] The profile inspection module is used to compare the generated model with the manually modeled model and point out the differences for correction.
[0033] Compared with the prior art, the technical solution provided in this application has at least the following beneficial effects:
[0034] The ship profile modeling method and system provided in this application are based on AI technology. They can accurately locate and generate 3D models of ship profiles from structured data, achieving a fully automated process from data extraction and model training to model generation. By using AI algorithms to extract features and generate models, human errors and inconsistencies caused by differences in experience and fatigue are avoided, ensuring standardized and high-quality model output. Furthermore, repetitive work by engineers is reduced, lowering the time and labor costs of ship design and manufacturing. In addition, the technical solution of this application also lays a solid foundation for AI-based automatic modeling technology for other structural elements of ships. Attached Figure Description
[0035] Figure 1The diagram shown is a flowchart of the ship profile modeling method provided in Example 1.
[0036] Figure 2 The diagram shown is of the ship profile modeling system provided in Example 2. Detailed Implementation
[0037] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.
[0038] It should be noted that the illustrations provided in this embodiment are only schematic representations of the basic concept of the present invention. Therefore, the illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.
[0039] Example 1:
[0040] See Figure 1 This embodiment provides an AI-based method for modeling ship profiles. The following is a detailed description of the profile modeling method provided in this embodiment.
[0041] S1: Based on historical CAD drawings of profiles, summarize the modeling elements of profiles, extract the structured data of profiles, and build a profile database.
[0042] Specifically, the first step is to conduct business discussions with engineers to collect information on the overall classification of profiles, the 3D features required for modeling each type of profile, and other detailed features of the profiles. This involves analyzing the mapping relationship between all features and CAD 2D drawings to prepare for training the CAD profile feature extraction model. Understandably, ship profiles are not single structures, and the modeling difficulties for straight and twisted profiles are drastically different. Therefore, profiles are divided into two main categories: straight profiles and twisted profiles. Tables 1 and 2 respectively show the modeling elements for these two types of profiles, with Table 1 being a profile feature description table and Table 2 being a profile trajectory line description table.
[0043] Table 1: Profile Characteristics Table
[0044]
[0045] Table 2: Trajectory Line Table
[0046]
[0047] As shown in the table above, the characteristics of straight profiles include: boundaries, specifications and materials, end cuts, ball head direction, and trajectory lines. The characteristics of twisted profiles include: boundaries, specifications and materials, end cuts, ball head direction, trajectory lines, profile support surface angles, and sheet support surface angles.
[0048] This step summarizes the above modeling elements of the profile during execution, determines the mapping relationship between profile features and CAD drawings, identifies the profile's boundaries, specifications, materials, end cuts, ball joint direction, and trajectory lines, and classifies the trajectory lines as straight lines, arcs, or ellipses. For twisted profiles, it further identifies the angle between the profile support surface and the plate support surface. The above table clearly defines the classification of profiles and their feature system, ensuring that the system can accurately handle all types of profiles in ship design, especially for technically challenging twisted profiles. This ensures the comprehensiveness and relevance of CNN feature extraction and solves the problem of accurately capturing the spatial angles of twisted profiles.
[0049] Based on the description in the table above, profile information is collected. Then, a specific development plan is formed based on the collected information, and specific program development is carried out. During the development process, various drawing methods will be encountered, such as drawing errors and inconsistent annotation methods. Thus, design specifications are continuously formed during the algorithm iteration process, design standards are unified, and data quality is improved. Finally, while completing model training, the design specifications of profiles in CAD drawings are released. At the same time, the information in this step is also preparing the input for model training.
[0050] S2: Use convolutional neural networks to extract profile features from the profile database, including the geometric features and topological relationships of the profiles.
[0051] S201: Prepare profile training data based on the profile features in step S1. Each training data structure includes: profile cross-sectional shape image - profile end-cut shape image - profile external dimensions. The profile external dimensions can be physical parameters such as height and thickness.
[0052] Specifically, step S201, preparing the profile training data, includes: vectorizing the profile images to generate grayscale images, such as images with a black background and white line outlines. Vectorization and grayscale conversion preserve the structural outline information of the profile while significantly reducing data redundancy and computational load. Normalizing the profile dimensions eliminates the influence of dimensions, removing deviations caused by different size units and orders of magnitude, such as the discrepancy caused by inconsistent length units like millimeters and inches. This allows the model to focus more on shape and proportion rather than absolute values, greatly improving the model's stability and generalization performance. This enables the model to adapt to drawing inputs of different standards and scales, ultimately resulting in a unified data structure for the convolutional neural network input.
[0053] S202: Since the performance of the AI model is highly dependent on the quality of the training data, according to the data structure in step S201, a convolutional neural network is used to extract at least the geometric features and topological relationships of the profile, such as the cross-section and end-cut shape.
[0054] S203: The above convolutional neural network is trained and optimized. The convolutional neural network employs a multi-scale structure and introduces an attention mechanism to improve the accuracy of profile feature extraction. This allows the model to learn autonomously and focus on the most critical feature regions for classification and generation tasks, such as the end chamfers of the profile or the connection points with other profiles. The profile feature output is a feature vector containing "curvature change sequence - cross-sectional shape classification - end chamfer classification". The "curvature change sequence" corresponds to the curvature values at different locations on the profile, digitally describing the dynamic center trajectory line of the profile. "Cross-sectional shape classification" defines its static cross-section, and "end chamfer classification" defines its end shape. The combination of these three provides sufficient feature description for the generation of 3D profiles, enabling the GAN generator to generate accurate models based on these explicit parameters.
[0055] S3: Based on the extracted profile features, a three-dimensional profile model is obtained by training a generative adversarial network.
[0056] S301: Based on the profile features extracted in step S2, design a Generative Adversarial Network (GAN): The generator takes the profile features extracted by the convolutional neural network as input and outputs a 3D voxel model of the profile; the discriminator takes the real 3D voxel model as input and outputs the authenticity probability and normality score.
[0057] S302: Rule Embedding and Adversarial Training: Physical feasibility loss, ship specification loss, and feature consistency loss are embedded in adversarial training. Model training is performed after embedding these rules. The generator generates candidate profile models, and the discriminator determines their authenticity. Iterative optimization continues until the output simultaneously satisfies both the distribution of real profile data and compliance with ship specification checks.
[0058] S303: Convert the voxel model into a solid model. The system iteratively corrects the converted solid model to ensure it conforms to the latest design specifications. It's understandable that voxel models are an output format in the field of AI, but they cannot be directly used in CAD / CAM production systems. Integration with software such as CATIA allows AI results to be implemented in actual workflows.
[0059] S304: Perform topology optimization in step S303. Topology optimization is a key preliminary step in computer-aided engineering (CAE) and additive manufacturing (AM). The directly generated model may have mesh defects or redundant details. The iterative correction function further enables the system to have the ability to learn and continuously improve itself, and can evolve with the release of new specifications.
[0060] Example 2:
[0061] See Figure 2 This embodiment provides an AI-based modeling system for ship profiles, which transforms the ship profile modeling method provided in Embodiment 1 into an executable algorithm. It integrates all functions from data input to model generation and verification, automatically outputs optimization schemes, and forms a complete solution.
[0062] Specifically, the AI-based ship profile modeling system provided in this embodiment includes the following functional modules.
[0063] The profile feature data extraction module is used to extract profile feature data from CAD drawings.
[0064] The model training module integrates the CNN feature extraction submodule and the GAN model training submodule, and is used to train the profile feature extraction model and the 3D generation model.
[0065] The 3D model generation module is used to generate 3D models of ship profiles based on the trained GAN model.
[0066] The profile inspection module is used to compare the generated model with the manually modeled model and point out the differences, assisting designers in checking whether the model has errors.
[0067] In summary, the ship profile modeling method and system provided in this application, leveraging AI technology, firstly extracts structured profile feature data automatically from CAD drawings, then uses a convolutional neural network to extract geometric features of the profile. This data is then input into a generative adversarial network for training, generating a 3D profile model. Based on this model and new profile data, a 3D profile model can be automatically created according to the designer's input, eliminating the designer's need to review drawings and create the model. Only standardized CAD 2D drawings are required, significantly reducing the time cost of repetitive operations in profile modeling and eliminating the personnel costs associated with repeated model modifications after completion. This revolutionizes the design process, opening a new avenue for AI-driven ship design and laying the foundation for AI-based automatic modeling of other ship structural elements. Therefore, the technical solution provided in this application has high industrial applicability due to its effective overcoming of the shortcomings of existing technologies.
[0068] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.
Claims
1. An AI-based method for modeling ship profiles, characterized in that, At least the following steps are included: S1: Based on the historical CAD drawings of profiles, summarize the modeling elements of profiles, extract the structured data of profiles, and build a profile database; S2: Use convolutional neural networks to extract profile features from the profile database, including the geometric features and topological relationships of the profiles; S3: Based on the extracted profile features, a three-dimensional profile model is obtained by training a generative adversarial network.
2. The AI-based ship profile modeling method according to claim 1, characterized in that, Profiles include straight profiles and twisted profiles; The features of the straight profile include: boundary, specifications and material, end cut, ball head direction and trajectory line; The characteristics of the twisted profile include: boundary, specifications and material, end cut, ball head direction, trajectory line, profile support surface and sheet support surface angle; Step S1 summarizes the modeling elements of the profile during the execution process, determines the mapping relationship between the profile features and the CAD drawings, identifies the profile's boundary, specifications, material, end cut, ball head direction and trajectory line, and classifies the trajectory line as a straight line, arc or ellipse; for twisted profiles, it further identifies the angle between the profile support surface and the sheet support surface.
3. The AI-based ship profile modeling method according to claim 1, characterized in that, Step S2 includes: S201: Prepare profile training data based on profile features. Each training data structure includes: profile cross-sectional shape image - profile end-cut shape image - profile external dimensions; S202: Based on the data structure in S201, a convolutional neural network is used to extract the geometric features and topological relationships of the profiles.
4. The AI-based ship profile modeling method according to claim 3, characterized in that, The preparation of profile training data in step S201 includes: vectorizing the profile image to generate a grayscale image; normalizing the profile dimensions to eliminate the influence of dimensions, so as to obtain a unified data structure for the input of the convolutional neural network.
5. The AI-based ship profile modeling method according to claim 3, characterized in that, It also includes the step of training a convolutional neural network: the convolutional neural network adopts a multi-scale structure and introduces an attention mechanism to improve the extraction accuracy of profile features.
6. The AI-based ship profile modeling method according to claim 3, characterized in that, The profile feature output is a feature vector containing "curvature change sequence - cross-sectional shape classification - end-cut classification".
7. The AI-based ship profile modeling method according to claim 1, characterized in that, Step S3 includes: S301: Design a generative adversarial network: The generator takes the profile features extracted by the convolutional neural network as input and outputs a 3D voxel model of the profile; the discriminator takes the real 3D voxel model as input and outputs the realism probability and normality score. S302: Rule Embedding and Adversarial Training: Embed physical feasibility loss, ship specification loss and feature consistency loss in adversarial training, and train the model.
8. The AI-based ship profile modeling method according to claim 7, characterized in that, Step S3 also includes S303: The step of converting the voxel model into a solid model. The system iteratively corrects the converted solid model to make it conform to the latest design specifications.
9. The AI-based ship profile modeling method according to claim 7, characterized in that, Step S3 also includes S304: Topology optimization steps for converting voxel models into solid models.
10. An AI-based modeling system for ship profiles, characterized in that, include: The profile feature data extraction module is used to extract profile feature data from CAD drawings; The model training module integrates the CNN feature extraction submodule and the GAN model training submodule, and is used to train the profile feature extraction model and the 3D generation model. The 3D model generation module is used to generate 3D models of ship profiles based on the trained GAN model. The profile inspection module compares the generated model with the manually modeled model and points out the differences for correction.