Through hole design method and system based on AI
By adopting AI-based through hole design methods in ship structure design and using machine learning models to automatically output through hole codes, the problems of complexity and low efficiency of through hole design are solved, and a more efficient and accurate design process is achieved.
Patent Information
- Application Number
- CN202510235826.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-28
- Publication Date
- 2025-06-06
AI Technical Summary
In ship structural design, the through hole design is complex, large workload and low efficiency due to the diversification of the intersecting forms of profiles and plates, and it is difficult to meet structural strength requirements.
Using the AI-based through hole design method, we create a parameterized template library in the three-dimensional design software, generate a training data set, select a suitable machine learning algorithm to train the model, output the through hole code, and call the template in the three-dimensional model to create the through hole.
It significantly reduces the difficulty of through-hole design, improves design efficiency and accuracy, reduces design costs, and realizes the elements of digital design of ships.
Smart Images
Figure CN120105596A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of ship structure design, and in particular, relates to an AI-based through-hole design method and system. Background Art
[0002] With the rapid development of digital design technology in the shipbuilding industry, digital design technology based on three-dimensional models has been widely used in the shipbuilding field. In the hull structure, in order to ensure the continuity of the longitudinal profiles of the hull and meet the structural strength requirements, it is necessary to design profile through holes. However, due to the diversity of intersections between profiles and plates (especially non-orthogonal situations), the contour design of through holes is relatively cumbersome. On the one hand, in order to meet all geometric scenarios where profiles and plates intersect, basic library designers need to refer to the standard node diagram of through holes and define personalized parameters for each type, resulting in a sharp increase in the configuration of through hole types, greatly increasing the workload and difficulty of basic library design; on the other hand, when structural designers apply through holes, they also need to refer to the standard node diagram and select the corresponding through holes for layout in combination with the through hole configuration, which increases the design cost and fails to reflect the digital design elements of ships.
[0003] Therefore, it is urgent to improve the design method of through holes to reduce the design difficulty and improve the design efficiency. Summary of the invention
[0004] In view of the shortcomings of the prior art mentioned above, the purpose of the present application is to provide an AI through-hole design method and system to solve the problems of difficulty and low efficiency in through-hole design in hull structure design based on three-dimensional models, so as to improve the overall construction efficiency of the ship.
[0005] In a first aspect, the present application provides an AI-based through-hole design method, comprising at least the following steps:
[0006] S1: Based on the standard node diagram of the through hole, a through hole parametric template library is created in the 3D design software;
[0007] S2: Based on the two-dimensional design drawing of the existing structure, the through-hole images are captured to generate a training data set;
[0008] S3: Select appropriate machine learning algorithms and train machine learning models;
[0009] S4: Based on the machine learning model trained in step S3, the two-dimensional design drawing of the structure is uploaded, and the machine learning model outputs the through-hole code;
[0010] S5: Compare the 2D design drawing with the 3D model to generate a machine learning algorithm validation database;
[0011] S6: Repeat steps S4 and S5 for the existing product data to optimize the algorithm and retrain the machine learning model;
[0012] S7: Based on the three-dimensional design software, according to the through-hole code recognized by the machine learning model, the template of step S1 is called in the three-dimensional model to create a through-hole.
[0013] In an optional implementation manner, step S1 includes:
[0014] S11: Based on the through-hole standard node diagram, encoding the through-hole category of each profile in the through-hole standard node diagram to obtain a through-hole code;
[0015] S12: sorting out the key size parameters required for each through hole code in step S11 and the structural relationships corresponding to the key size parameters to form a comparison table;
[0016] S13: according to step S12, a through-hole parameterized template is created corresponding to each through-hole code to form a through-hole parameterized template library.
[0017] In an optional implementation manner, step S2 includes:
[0018] S21: for different product types, functional areas, structural forms and through-hole types, intercepting a number of through-hole images of the two-dimensional design drawings respectively;
[0019] S22: labeling the image obtained in step S21, where the label data is a through hole code;
[0020] S23: Generate a training data set based on the labeled through-hole images in step S22.
[0021] In an optional implementation, in step S3, a convolutional neural network machine learning algorithm is used to perform machine learning.
[0022] In an optional implementation, a preprocessing operation is performed on the intercepted through-hole image, and the preprocessing operation includes one or more operations of normalization, expansion or random shuffling.
[0023] In an optional implementation, the constructed convolutional neural network model includes two convolutional layers, two pooling layers, a fully connected layer and an output layer.
[0024] In an optional implementation, a gradient descent algorithm is used to generate a machine training model;
[0025] Evaluate the performance of machine trained models using preprocessed images.
[0026] In an optional implementation manner, step S5 includes:
[0027] S51: for the two-dimensional design drawing marked with the through hole code generated in step S4, extract the coordinate value of the center of gravity of the through hole relative to the hull coordinate system, and enter it into the database;
[0028] S52: extracting the through hole code and the coordinate value of the center of gravity of the through hole from the existing three-dimensional model, converting them into the new version of the through hole code according to the comparison table in step S12, and then entering them into the database;
[0029] S53: Compare the information generated in step S51 and step S52, determine the output result of the training model, and enter the determination result into the database.
[0030] In a second aspect, the present application provides an AI-based through-hole design system, comprising:
[0031] 3D model maintenance module: used to create 3D parametric templates and generate 3D models;
[0032] Two-dimensional image data extraction module: used to extract structured data from two-dimensional design drawings and process data;
[0033] Image tagging module: used to label image data;
[0034] Database module: used to store extracted structured data;
[0035] Machine learning model training module: used for machine learning algorithm selection, machine learning model training, machine learning model verification, and machine learning model testing;
[0036] Machine learning model result output module: used to output the recognition results of the machine learning model.
[0037] Compared with the prior art, the technical solution provided by this application has the following beneficial effects:
[0038] The technical solution provided by this application is based on AI technology for through-hole design, which greatly reduces the workload of basic library designers in creating through-hole template libraries, and also reduces the workload of structural designers in manually selecting through-hole types by comparing standard node diagrams. In the overall process of ship design and construction, it significantly reduces the difficulty of through-hole design and improves design efficiency and design accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 Shown is a step diagram of the through-hole design method based on AI provided in this application;
[0040] Figure 2 Shown is a diagram of the AI-based through-hole design system provided in this application. DETAILED DESCRIPTION
[0041] The following describes the implementation methods of the present application through specific examples. Those skilled in the art can easily understand other advantages and principles of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific implementation methods. The details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application.
[0042] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the restrictive conditions for the implementation of this patent, so they have no substantial technical significance. Any structural modification, change in proportional relationship or adjustment of size, without affecting the effects and purposes that can be achieved by this patent, should still fall within the scope of the technical content disclosed by this patent. At the same time, the terms such as "upper", "lower", "left", "right", "middle" and so on quoted in this specification are only for the convenience of description, and are not used to limit the scope of the implementation of this patent. The change or adjustment of their relative relationship should also be regarded as the scope of the implementation of this patent without substantial change in the technical content.
[0043] Embodiment 1:
[0044] This embodiment provides an AI-based through-hole design method, comprising the following steps:
[0045] S1: Based on the standard node diagram of the through hole, create a parametric template for the through hole in the 3D design software. The 3D design software can be SolidWorks, CATIA, etc. The template library includes the design parameters of the through holes of different profiles (such as hole diameter D, spacing L, inclination angle θ, etc.). The template can be quickly called through scripts or plug-ins, which can realize the standardization and automation of the through hole design and reduce the workload of repeated design.
[0046] Specifically, step S1 includes:
[0047] S11: Based on the standard node diagram of through holes, such as the GB / T ship structure node atlas, the through hole category of each profile in the standard node diagram of through holes is uniquely coded: [profile code]-[function code]-[serial number] (such as "Beam-A01-001"), and the through hole code is obtained to realize the standardized management of the through hole category.
[0048] S12: Sort out the key dimensional parameters required for each through-hole code in step S11 and the structural relationships corresponding to the key dimensional parameters (such as aperture D, spacing L, inclination angle θ, etc.) to form a comparison table, for example, the structural relationship formula L=2D+10mm; through unique coding and comparison tables, standardize the design to eliminate design ambiguity and reduce the risk of manual error operation.
[0049] S13: Create a parametric template for each through hole code according to step S12 to form a parametric template library for through holes. For example, when creating an adaptive template in CATIA, when D=50mm is input, L=110mm is automatically calculated and a hole array is generated. The parametric template supports one-click generation of complex through hole structures, which can be quickly called, and the design time is verified to be reduced by at least 20%.
[0050] S2: Based on the two-dimensional design drawings of the existing structure, capture the through-hole images and generate a training data set. Capture the through-hole images from the existing two-dimensional design drawings, use image annotation tools (such as LabelImg, AutoLabel) to add through-hole code labels, generate training data sets, provide high-quality training data, and lay the foundation for machine learning models.
[0051] Specifically, step S2 includes:
[0052] S21: For different product types, functional areas, structural forms and through-hole types, several through-hole images of two-dimensional design drawings are captured respectively, for example, 1,000 400*400 pixel images are captured for each. Through images of different product types and functional areas, the generalization ability of the model is improved, and multi-dimensional classification data covers various design scenarios, improving the recognition accuracy of the model.
[0053] S22: Data labeling is performed on the image obtained in step S21, and the label data is the through-hole code; the through-hole code is used as a label to mark the through-hole contour in the image to provide accurate training data.
[0054] S23: Generate a training data set based on the labeled through-hole images in step S22. The accuracy of the training data is improved through precise label data, and the key point coordinate annotation provides a data basis for subsequent coordinate comparison to reduce verification errors. For example, use a Python script to batch capture the through-hole areas in AutoCAD / DWG drawings, save them in PNG format, and use the LabelStudio annotation tool to label the images with the through-hole code (such as "Hole-TypeA-001"), and divide the training set, validation set, and test set into a ratio of 8:1:1 to ensure balanced data distribution.
[0055] S3: Select a suitable machine learning algorithm and train the machine learning model. Optionally, use a convolutional neural network machine learning algorithm for machine learning and use the TensorFlow or PyTorch framework for model training. Convolutional neural networks are highly robust to noise and deformation, can effectively extract features, and have obvious advantages in processing image data. For example, select ResNet-18 as the CNN backbone network, and use transfer learning to optimize training efficiency; set hyperparameters, initial learning rate 0.001, batch size 32, and number of iterations 100; use the cross entropy loss function, and the Adam optimizer dynamically adjusts the learning rate.
[0056] In an optional embodiment, a preprocessing operation is performed on the intercepted through-hole image, and the preprocessing operation includes one or more operations of normalization, expansion or random shuffling. The normalization operation makes the data on the same scale to facilitate model processing. The expansion and random shuffling operations increase the diversity of the data and improve the generalization ability of the model. The random shuffling disrupts the order of the data to avoid overfitting.
[0057] In an optional implementation, the constructed convolutional neural network model includes two convolutional layers, two pooling layers, a fully connected layer and an output layer. By reasonably designing the number of layers, the model structure is simplified and the training efficiency is improved. The combination of the convolutional layer and the pooling layer effectively extracts image features, and the fully connected layer and the output layer implement the classification task. Specifically, the convolutional neural network includes:
[0058] Input layer: receives the through-hole image normalized to 224×224 pixels;
[0059] Feature extraction module: contains 4 convolutional layers and 2 maximum pooling layers, which are used to extract local features of the image;
[0060] Classification module: Contains the global average pooling layer and the Softmax output layer, and outputs the classification results of the through-hole code.
[0061] The multi-layer convolution structure captures image details (such as hole edges and shapes), improves classification accuracy, and replaces the fully connected layer with global average pooling, reducing the number of parameters by 30%, speeding up inference, and achieving the goal of lightweight design.
[0062] In an optional implementation, a gradient descent algorithm is used to generate a machine training model, for example, Nesterov momentum acceleration, momentum coefficient 0.9, learning rate decay is reduced to 1 / 10 of the original every 20 rounds, and training is terminated if the validation set loss does not decrease for 5 consecutive rounds. The gradient descent algorithm can effectively optimize model parameters and speed up training; the performance of the machine training model is evaluated using preprocessed images, and the model performance is evaluated by preprocessing images to ensure the reliability of the model in practical applications.
[0063] S4: Based on the machine learning model trained in step S3, upload the two-dimensional design drawing of the structure, and the machine learning model outputs the through-hole code; input the two-dimensional design drawing into the trained model, the model outputs the through-hole code and saves it to the database, so as to realize the automatic generation of through-hole design and reduce manual intervention.
[0064] S5: Compare the 2D design with the 3D model to generate a machine learning algorithm verification database. Through coordinate projection comparison, the spatial consistency between the 2D design and the 3D model is ensured, reducing the manual verification workload by 20%, and the verification results are directly entered into the database for future reference.
[0065] Specifically, step S5 includes:
[0066] S51: For the two-dimensional design drawing marked with the through hole code generated in step S4, extract the center of gravity coordinate value of the through hole relative to the hull coordinate system and enter it into the database; for example, extract the center of gravity coordinate (X, Y) of the through hole in the two-dimensional design drawing and the corresponding through hole code.
[0067] S52: For the existing three-dimensional model, extract the through hole code and the through hole centroid coordinate value, and convert them into the new version of the through hole code according to the comparison table in step S12 and enter them into the database; for example, extract the centroid coordinates (X', Y', Z') and historical design code of the same through hole in the three-dimensional model.
[0068] S53: Compare the information generated in step S51 and step S52, determine the output result of the training model, and enter the determination result into the database. For example, compare the plane projection error of the two-dimensional and three-dimensional coordinates. If the error is ≤ 2mm and the code is consistent, then the verification is determined to be passed.
[0069] S6: Repeat steps S4 and S5 for the existing product data, optimize the algorithm and retrain the machine learning model; adjust the number of CNN convolution kernels and layers according to the error distribution in the verification database. If the recognition error rate of a specific through-hole category is greater than 5%, expand its training data separately and retrain. The above operations can improve model performance and reduce recognition error rate.
[0070] S7: Based on the three-dimensional design software, according to the through-hole code recognized by the machine learning model, the template of step S1 is called in the three-dimensional model to create a through-hole.
[0071] The through-hole design method provided in this embodiment includes the steps of creating a through-hole parametric template, capturing feature images to generate a training data set, training a machine learning model, outputting a through-hole code from the machine learning model, verifying the machine learning model, optimizing and iterating the machine learning model, and generating a through-hole model. From data collection to three-dimensional model generation, manual intervention is greatly reduced and design efficiency is improved; through a parametric template library and a model verification mechanism, the through-hole design is ensured to match the standard node diagram 100%. In addition, the technical solution of this embodiment supports rapid adaptation of multiple types of products, structural forms, and through-hole categories, greatly shortening the project development cycle. Take the design of a ship bulkhead as an example:
[0072] Original data: 200 AutoCAD 2D drawings, including 5 types of through holes (ventilation holes, cable holes, etc.).
[0073] Design time: reduced from 14 man-days to 11 man-days.
[0074] Material waste: The scrap rate of plates due to design errors is reduced by 15%.
[0075] Model accuracy: The first round of verification passed 92%, and reached 99% after 2 iterations of optimization.
[0076] Embodiment 2:
[0077] This embodiment provides an AI-based through-hole design system, including:
[0078] 3D model maintenance module: used to create 3D parametric templates and generate 3D models;
[0079] Two-dimensional image data extraction module: used to extract structured data from two-dimensional design drawings and process data;
[0080] Image tagging module: used to label image data;
[0081] Database module: used to store extracted structured data;
[0082] Machine learning model training module: used for machine learning algorithm selection, machine learning model training, machine learning model verification, and machine learning model testing;
[0083] Machine learning model result output module: used to output the recognition results of the machine learning model.
[0084] The major modules of the through-hole design system provided in this embodiment operate in a closed loop, achieving full process coverage of design-generation-verification-optimization, and supporting API docking with mainstream 3D design software (such as SolidWorks, CATIA) without switching platforms, thereby achieving full process automation of through-hole design and improving design efficiency and quality.
[0085] In summary, the AI-based through-hole design method and system technical solution provided in this application have high industrial utilization value because they effectively overcome various shortcomings in the existing technology.
[0086] The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.
Claims
1. A through hole design method based on AI, characterized in that: At least the following steps are included: S1: Based on the standard node diagram of the through hole, a through hole parametric template library is created in the 3D design software; S2: Based on the two-dimensional design drawing of the existing structure, the through-hole images are captured to generate a training data set; S3: Select appropriate machine learning algorithms and train machine learning models; S4: Based on the machine learning model trained in step S3, the two-dimensional design drawing of the structure is uploaded, and the machine learning model outputs the through-hole code; S5: Compare the 2D design drawing with the 3D model to generate a machine learning algorithm validation database; S6: Repeat steps S4 and S5 for the existing product data to optimize the algorithm and retrain the machine learning model; S7: Based on the three-dimensional design software, according to the through-hole code recognized by the machine learning model, the template of step S1 is called in the three-dimensional model to create a through-hole.
2. The AI-based through-hole design method according to claim 1, characterized in that: Step S1 includes: S11: Based on the through hole standard node diagram, encoding the through hole category of each profile in the through hole standard node diagram to obtain a through hole code; S12: sorting out the key size parameters required for each through hole code in step S11 and the structural relationships corresponding to the key size parameters to form a comparison table; S13: according to step S12, a through-hole parameterized template is created corresponding to each through-hole code to form a through-hole parameterized template library.
3. The AI-based through-hole design method according to claim 1, characterized in that: Step S2 includes: S21: for different product types, functional areas, structural forms and through-hole types, intercepting a number of through-hole images of the two-dimensional design drawings respectively; S22: labeling the image obtained in step S21, where the label data is a through hole code; S23: Generate a training data set based on the labeled through-hole images in step S22.
4. The AI-based through-hole design method according to claim 1, characterized in that: In step S3, a convolutional neural network machine learning algorithm is used to perform machine learning.
5. The AI-based through-hole design method according to claim 4, characterized in that: The captured through-hole image is preprocessed, wherein the preprocessing operation includes one or more operations of normalization, expansion or random shuffling.
6. The AI-based through-hole design method according to claim 5, characterized in that: The constructed convolutional neural network model includes two convolutional layers, two pooling layers, a fully connected layer and an output layer.
7. The AI-based through-hole design method according to claim 6, characterized in that: Use gradient descent algorithm to generate machine training model; Evaluate the performance of machine trained models using preprocessed images.
8. The AI-based through-hole design method according to claim 2, characterized in that: Step S5 includes: S51: for the two-dimensional design drawing marked with the through hole code generated in step S4, extract the coordinate value of the center of gravity of the through hole relative to the hull coordinate system, and enter it into the database; S52: For the existing 3D model, extract the through hole code and the through hole centroid coordinate value, and The comparison table in S12 is converted into the new version of the through hole code and entered into the database; S53: Compare the information generated in step S51 and step S52, determine the output result of the training model, and enter the determination result into the database.
9. An AI-based through-hole design system, characterized in that: include: 3D model maintenance module: used to create 3D parametric templates and generate 3D models; Two-dimensional image data extraction module: used to extract structured data from two-dimensional design drawings and process data; Image tagging module: used to label image data; Database module: used to store extracted structured data; Machine learning model training module: used for machine learning algorithm selection, machine learning model training, machine learning model verification, and machine learning model testing; Machine learning model result output module: used to output the recognition results of the machine learning model.