Shell structure meshing method and device, electronic equipment and storage medium

Abstract

The present application relates to the technical field of computer simulation analysis, and particularly relates to a plate shell structure grid division method and device, electronic equipment and storage medium, the method comprising: obtaining a feature structure image of a to-be-detected plate shell; identifying the feature structure image based on a preset deep learning feature recognition model to obtain a region feature recognition result of the to-be-detected plate shell; determining a typical feature region and an atypical feature region of the to-be-detected plate shell according to the region feature recognition result, determining a first grid division strategy of the typical feature region, and performing grid division on the typical feature region based on the first grid division strategy, and performing grid division on the atypical feature region based on finite element software, thereby solving the problem of low finite element analysis grid quality qualification rate and low work efficiency caused by the excessively complex geometry of the plate shell structure in the related art, improving the typical feature region grid division quality and the efficiency of finite element analysis.

Description

Technical Field

[0001] This invention relates to the field of computer simulation analysis technology, and in particular to a method, apparatus, electronic device, and storage medium for mesh generation of plate and shell structures. Background Technology

[0002] In the field of finite element analysis, automatic mesh generation technology for plate and shell structures is a key technology, which is of great significance for improving analysis efficiency, reducing manual intervention, and ensuring the accuracy of analysis results.

[0003] In related technologies, the main term refers to the batchmesh function of mainstream finite element software (such as Hypermesh, ANSA, etc.). This function can automatically generate meshes for plate and shell structures by using predefined mesh quality standards (such as size, Jacobian, warpage, etc.), which effectively reduces manual intervention and greatly improves work efficiency.

[0004] However, for plate and shell structures with complex geometries, the batchmesh function struggles to generate high-quality meshes, leading to mesh distortion and reduced element quality. This has a series of adverse effects on finite element analysis: reduced accuracy of finite element analysis results, inability of iterative calculations to converge to the correct solution, misjudgment of structural performance by engineers, low mesh quality pass rate, and reduced work efficiency. These issues urgently need to be addressed. Summary of the Invention

[0005] This invention provides a method, apparatus, electronic device, and storage medium for mesh generation of plate and shell structures, in order to solve the problems of low mesh quality and low work efficiency in finite element analysis caused by the overly complex geometry of plate and shell structures in related technologies. It improves the mesh generation quality of typical characteristic regions and enhances the efficiency of finite element analysis.

[0006] To achieve the above objectives, a first aspect of the present invention provides a method for meshing a plate and shell structure, comprising the following steps: acquiring a feature structure image of a plate and shell to be detected; recognizing the feature structure image based on a preset deep learning feature recognition model to obtain a regional feature recognition result of the plate and shell to be detected; determining typical and atypical feature regions of the plate and shell to be detected based on the regional feature recognition result, determining a first meshing strategy for the typical feature regions, meshing the typical feature regions based on the first meshing strategy, and meshing the atypical feature regions based on finite element software.

[0007] Furthermore, in some embodiments, before recognizing the target structure image in the shell structure to be detected based on a pre-trained deep learning model, the method further includes: acquiring a target feature image dataset; dividing the target feature image dataset into a training set, a validation set, and a test set based on a preset partitioning ratio; constructing a target feature recognition neural network, inputting the training set into the target feature recognition neural network for training to obtain initial model parameters; based on the initial model parameters, inputting the validation set into the target neural network for performance evaluation, and adjusting the initial model parameters according to the performance evaluation results until the joint loss function of the test set converges to obtain optimal model parameters; based on the optimal model parameters, inputting the test set into the target neural network for model testing, and obtaining the preset deep learning feature recognition model when the test results meet preset requirements.

[0008] Furthermore, in some embodiments, the acquisition of the target feature image dataset includes: acquiring a feature structure image of the target plate and shell structure; processing the feature structure image of the target plate and shell structure based on a preset data processing method to obtain a normalized feature structure image; performing feature classification and annotation on the normalized feature structure image based on a pre-built feature classification library, and constructing a target feature image dataset based on the feature classification and annotation results.

[0009] Furthermore, in some embodiments, determining the first grid partitioning strategy for the typical feature region and partitioning the typical feature region into a grid based on the first grid partitioning strategy includes: determining the feature type of the typical feature region; determining the grid partitioning strategy corresponding to each type based on the feature type and a preset grid partitioning strategy table; and partitioning the typical feature region into a grid based on the corresponding grid partitioning strategy.

[0010] Furthermore, in some embodiments, after meshing the typical feature regions based on the first meshing strategy and meshing the atypical feature regions based on finite element software, the method further includes: performing quality assessment on each meshing result of the plate and shell structure based on a preset mesh quality detection method; if the quality assessment result meets the preset mesh quality detection standard, then the meshing result of the plate and shell structure is output; otherwise, the meshing result of the plate and shell structure is meshed and optimized until it meets the preset mesh quality detection standard.

[0011] According to the plate and shell structure meshing method provided in the embodiments of the present invention, the feature structure image of the plate and shell to be detected is first acquired, and the feature structure image is intelligently recognized by a preset deep learning feature recognition model, and the regional feature recognition result of the plate and shell to be detected is output. Based on the recognition result, the typical feature regions and atypical feature regions of the plate and shell to be detected are accurately distinguished. A dedicated first meshing strategy is matched for the typical feature regions and high-precision mesh generation is performed. At the same time, the atypical feature regions are efficiently and automatically divided using finite element software. This solves the problem of low mesh quality qualification rate and low work efficiency in related technologies due to the overly complex geometry of the plate and shell structure, improves the meshing quality of typical feature regions, and improves the efficiency of finite element analysis.

[0012] To achieve the above objectives, a second aspect of the present invention provides a plate and shell structure meshing device comprising: an acquisition module for acquiring a feature structure image of a plate and shell to be detected; an identification module for identifying the feature structure image based on a preset deep learning feature recognition model to obtain a region feature recognition result of the plate and shell to be detected; and a partitioning module for determining typical and atypical feature regions of the plate and shell to be detected based on the region feature recognition result, determining a first meshing strategy for the typical feature regions, partitioning the typical feature regions into meshes based on the first meshing strategy, and partitioning the atypical feature regions into meshes based on finite element software.

[0013] Furthermore, in some embodiments, before recognizing the target structure image in the shell structure to be detected based on a pre-trained deep learning model, the recognition module is further configured to: acquire a target feature image dataset; divide the target feature image dataset into a training set, a validation set, and a test set based on a preset division ratio; construct a target feature recognition neural network, input the training set into the target feature recognition neural network for training to obtain initial model parameters; based on the initial model parameters, input the validation set into the target neural network for performance evaluation, and adjust the initial model parameters according to the performance evaluation results until the joint loss function of the test set converges to obtain optimal model parameters; based on the optimal model parameters, input the test set into the target neural network for model testing, and obtain the preset deep learning feature recognition model when the test results meet preset requirements.

[0014] Furthermore, in some embodiments, the recognition module is also used to: acquire a feature structure image of the target shell structure; process the feature structure image of the target shell structure based on a preset data processing method to obtain a normalized feature structure image; perform feature classification and annotation on the normalized feature structure image based on a pre-built feature classification library, and construct a target feature image dataset based on the feature classification and annotation results.

[0015] Furthermore, in some embodiments, the partitioning module is specifically used for: determining the feature type of the typical feature region; determining the grid partitioning strategy corresponding to each type based on the feature type and a preset grid partitioning strategy table; and performing grid partitioning on the typical feature region based on the corresponding grid partitioning strategy.

[0016] Furthermore, in some embodiments, after meshing the typical feature regions based on the first meshing strategy and meshing the atypical feature regions based on finite element software, the meshing module is further configured to: perform quality assessment on each meshing result of the plate and shell structure based on a preset mesh quality detection method; if the quality assessment result meets the preset mesh quality detection standard, then output the meshing result of the plate and shell structure; otherwise, perform mesh correction and optimization on the meshing result of the plate and shell structure until it meets the preset mesh quality detection standard.

[0017] According to the plate and shell structure mesh generation device provided in the embodiment of the present invention, the feature structure image of the plate and shell to be detected is first acquired, and the feature structure image is intelligently recognized by a preset deep learning feature recognition model, and the regional feature recognition result of the plate and shell to be detected is output. Based on the recognition result, the typical feature regions and atypical feature regions of the plate and shell to be detected are accurately distinguished. A dedicated first mesh generation strategy is matched for the typical feature regions and high-precision mesh generation is performed. At the same time, the atypical feature regions are efficiently and automatically divided using finite element software. This solves the problem of low mesh quality qualification rate and low work efficiency in related technologies due to the overly complex geometry of the plate and shell structure, improves the mesh generation quality of typical feature regions, and improves the efficiency of finite element analysis.

[0018] To achieve the above objectives, a third aspect of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the plate and shell structure mesh division method as described in the above embodiments.

[0019] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the plate and shell structure meshing method as described in the above embodiments.

[0020] A fifth aspect of the present invention provides a computer program product, including a computer program that is executed to implement the plate and shell structure meshing method as described in the above embodiments.

[0021] Therefore, the present invention has the following beneficial effects: (1) The present invention can achieve 100% typical feature mesh division of complex plate and shell structure mesh quality, that is, the typical feature mesh quality is 100%.

[0022] (2) The present invention can achieve 90% mesh quality of complex plate and shell structure mesh division, that is, the overall mesh quality of the plate and shell structure after mesh division is 90%.

[0023] (3) The present invention can effectively reduce the time required for manual repair of inferior grids. Attached Figure Description

[0024] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A flowchart of a plate and shell structure mesh generation method provided according to an embodiment of the present invention; Figure 2 This is a schematic diagram of typical feature acquisition results provided according to a specific embodiment of the present invention; Figure 3 A schematic diagram illustrating 21 typical feature labels and their meanings according to a specific embodiment of the present invention; Figure 4 This is a structural schematic diagram of 21 typical feature categories provided according to a specific embodiment of the present invention; Figure 5 This is a schematic diagram of a target feature image dataset provided according to a specific embodiment of the present invention; Figure 6 This is a schematic diagram of a Resnet50 network structure according to a specific embodiment of the present invention; Figure 7 This is a schematic flowchart of a plate and shell structure mesh generation method according to a specific embodiment of the present invention; Figure 8 A schematic diagram comparing typical feature mesh division results provided according to a specific embodiment of the present invention. Figure 9 This is a block diagram of a plate and shell structure mesh division device provided according to an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0025] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0026] The following description, with reference to the accompanying drawings, outlines a plate and shell structure meshing method, apparatus, electronic device, and storage medium according to embodiments of the present invention. First, the plate and shell structure meshing method according to embodiments of the present invention will be described with reference to the accompanying drawings.

[0027] Figure 1 This is a flowchart of a plate and shell structure meshing method provided according to an embodiment of the present invention.

[0028] like Figure 1 As shown, the mesh generation method for this plate and shell structure includes the following steps: In step S101, the feature structure image of the shell to be detected is obtained.

[0029] Among them, the feature structure image of the shell to be detected refers to the visualization of the target shell structure to be divided into grids, and the two-dimensional image of the standardized geometric structure with fixed shape that is prone to grid distortion is extracted.

[0030] Specifically, finite element software (such as Hypermesh, ANSA, etc.) is used to acquire images of typical structures in complex plate and shell structures. The background color, feature color, image size, and image format of the images are set consistently, and the display angle of the typical features is kept consistent.

[0031] In step S102, the feature structure image is identified based on a preset deep learning feature recognition model to obtain the regional feature recognition result of the shell to be detected.

[0032] Furthermore, in some embodiments, before recognizing the target structure image in the shell structure to be detected based on the pre-trained deep learning model, the method further includes: acquiring a target feature image dataset; dividing the target feature image dataset into a training set, a validation set, and a test set based on a preset partitioning ratio; constructing a target feature recognition neural network, inputting the training set into the target feature recognition neural network for training to obtain initial model parameters; based on the initial model parameters, inputting the validation set into the target neural network for performance evaluation, and adjusting the initial model parameters according to the performance evaluation results until the joint loss function of the test set converges to obtain the optimal model parameters; based on the optimal model parameters, inputting the test set into the target neural network for model testing, and obtaining a preset deep learning feature recognition model when the test results meet preset requirements.

[0033] In some embodiments, obtaining the target feature image dataset includes: obtaining the feature structure image of the target plate and shell structure; processing the feature structure image of the target plate and shell structure based on a preset data processing method to obtain a normalized feature structure image; performing feature classification and annotation on the normalized feature structure image based on a pre-built feature classification library, and constructing the target feature image dataset based on the feature classification and annotation results.

[0034] As one possible implementation, this embodiment of the invention uses ANSA software to collect images of typical structures in the body-in-white of a passenger vehicle. All images have a black background, gray feature color, a size of 700 pixels x 700 pixels, and are in JPG format. The images are named with "stamp_" + "typical feature acquisition order". Figure 2 This is a schematic diagram of typical feature acquisition results provided according to a specific embodiment of the present invention. Figure 3 This is a schematic diagram illustrating 21 typical feature labels and their meanings according to a specific embodiment of the present invention. Figure 4 This is a structural diagram of 21 typical feature categories provided according to a specific embodiment of the present invention, combined with... Figure 2 , Figure 3 and Figure 4 ,in, Figure 2 (a) is a circle, that is, a circular boss. Figure 2 (b) is normal_long_feature, i.e., a regular long feature protrusion. Figure 2 (c) is a circle with a hole, i.e., a circular boss with a hole. Figure 2 (d) represents the vertical long feature, i.e., the vertical long feature protrusion. Then, the typical features are divided into 21 types according to the feature classification library. This invention divides the typical features in complex plate and shell structures into 21 standard geometric structural units. Each type of typical feature corresponds to a unique category label and structural definition. The shape, structural composition, hole type, and layout of each type of typical feature are clearly shown through visual geometric graphics. Finally, corresponding folders are created according to the typical feature category, with the folder name named after the typical feature category. The collected typical feature images are placed into the corresponding folders according to the category to construct the target feature image dataset. Figure 5 This is a schematic diagram of a target feature image dataset provided according to a specific embodiment of the present invention.

[0035] Furthermore, after obtaining the target feature image dataset, this invention uses a deep learning framework to identify images with 21 typical features. For example, this invention uses PyTorch to build a ResNet50 to train a deep learning model.

[0036] in, Figure 6 This is a schematic diagram of a ResNet50 network structure according to a specific embodiment of the present invention, as shown below. Figure 6 As shown, the ResNet50 consists of 5 stages. Its standard input size is a 224×224 3-channel RGB image (3, 224, 224), and the output feature map size is (2038, 7, 7). Stage 0 includes a 7×7 convolutional layer (7×7 kernel, 64 output channels, stride 2, ReLU activation function) and a 3×3 max-pooling layer (3×3 kernel, stride 2), with an output size of (64, 56, 56). Stage 1 includes 3 concatenated residual blocks and a Bit-Shift key (BT). NK1: 64, 56, 64, 1 (first bottleneck block, stride 1, no downsampling) has a structure of 1×1 convolution, 3×3 convolution, 1×1 convolution and residual structure, with an output size of (256, 56, 56). Two BTNK2 blocks: 256, 56 (standard bottleneck blocks, stride 1), with a structure of 1×1 convolution, 3×3 convolution, 1×1 convolution and residual structure, also with an output size of (256, 56, 56). Therefore, the output of Stage 1 is (256, 56, 56). Stage 2 includes four residual blocks concatenated and one BTNK1 block. The code snippet 256, 56, 128, 2 (bottleneck block with step size 2, first downsampling) increases the channel count from 256 to 512, resulting in an output size of (512, 28, 28). Three BTNK2 blocks (512, 28, standard bottleneck block, step size 1) also have an output size of (512, 28, 28). Therefore, Stage 2's output is (512, 28, 28). Stage 3 includes six residual blocks, one BTNK1 block (512, 28, 256, 2, bottleneck block with step size 2, second downsampling) with an output size of (1024, 14, 14), and five B... TNK2: 1024, 14 (standard bottleneck block, stride 1) has an output size of (1024, 14, 14), so the output of Stage 3 is (1024, 14, 14). In Stage 4, there are 3 residual blocks: one BTNK1: 1024, 14, 512, 2 (bottleneck block with stride 2, third downsampling) with an output size of (2048, 7, 7), and two BTNK2: 2048, 7 (standard bottleneck block, stride 1) with an output size of (2048, 7, 7). So the final output feature map size is (2048, 7, 7).

[0037] It should be noted that the ResNet-50 has a total of 16 residual blocks. Including the initial convolutional layer, the total number of layers is 50. After the final output (2048, 7, 7) feature map, it is followed by global average pooling (GAP) to compress it into a 2048-dimensional vector, and then the classification is completed through a fully connected layer (the output dimension is the number of categories, such as 21 categories in this invention).

[0038] Furthermore, since there are 21 categories, the last fully connected layer of ResNet50 needs to be modified to change the number of output features to 21, and the cross-entropy loss function should be selected as the loss function for the classification task, with SGD chosen as the optimizer. As one possible implementation, this embodiment of the invention divides the dataset into a training set and a test set. The training set is used for model training, and the test set is used for model testing. The division ratio can be determined according to the size of the dataset. First, two new folders are created, named train and test respectively. Then, subfolders are created for each category under these two folders. Subsequently, each category folder in the original dataset is traversed, and 80% of the images are randomly selected and placed into the subfolders corresponding to the train folder, and the remaining 20% ​​are placed into the subfolders corresponding to the test folder.

[0039] Furthermore, the image data undergoes preprocessing, including scaling, cropping, and normalization, to adapt to the input requirements of the deep learning model. This involves converting the raw data into a format that the model can understand and process, and creating a dataset and data loader capable of efficiently loading the data. First, images are loaded from files into memory using PIL or `ToPILImage` from `torchvision.transforms`. Since the input size of the ResNet50 model is 224x224 pixels, the `torchvision.transforms.Resize` method is used to resize each image to 224x224 pixels. PIL images are then converted to PyTorch tensors and normalized, using the mean and standard deviation from the ImageNet dataset as normalization parameters. Finally, custom dataset classes and data loaders are created using `torch.utils.data.Dataset` and `torch.utils.data.DataLoader`. The model is trained using the training set, calculating predicted values ​​through forward propagation and then calculating the loss function. Gradients are calculated using the backpropagation algorithm, and the optimizer is used to update the model's weights. The model and optimizer are initialized first. The model is transferred to the GPU and the optimizer is initialized. Then, training of ResNet50 begins. For each epoch, a batch of data and labels is loaded using the data loader. The data is input into the model to obtain the prediction results. The loss function is used to calculate the loss between the prediction results and the true labels. The gradient is calculated and the model parameters are updated. The loss and accuracy during the training process are recorded. The model state is saved after each epoch or when the validation loss reaches a certain threshold.

[0040] Furthermore, the trained model is tested using a test set to evaluate its performance on real-world data. If the model's performance on the test set meets the preset standard, a trained deep learning model is obtained. If the model's performance on the test set does not meet the preset standard, corresponding strategies (such as data augmentation, hyperparameter optimization, model architecture adjustment, optimizer adjustment, regularization, etc.) are adopted to optimize the model and retrain it until the model's performance on the test set meets the preset standard. Specifically, the best-performing model during training is loaded from the saved models. Test set data is loaded using a data loader, and forward propagation is performed on each batch of data in the test set to obtain prediction results. Indicators such as accuracy and confusion matrix on the test set are calculated. Finally, the model's performance is analyzed based on the test results, and optimization strategies are adopted to optimize the model's performance. These optimizations include adjusting data preprocessing steps, such as changing normalization parameters; adjusting model architecture, such as increasing or decreasing model depth; and adjusting hyperparameters, such as learning rate, batch size, or optimizer type, ultimately obtaining the preset deep learning feature recognition model.

[0041] In step S103, the typical and atypical feature regions of the shell to be detected are determined based on the regional feature recognition results, and the first mesh division strategy for the typical feature regions is determined. The typical feature regions are then meshed based on the first mesh division strategy, and the atypical feature regions are meshed based on finite element software.

[0042] Among them, the typical feature area refers to the structural area on the plate shell to be tested that has a fixed geometric shape, appears repeatedly, and is prone to mesh distortion, including the area where 21 types of standardized geometric structures such as circular bosses, perforated bosses, long features, gourd-shaped features, and polygonal bosses are located. The non-typical feature area refers to the flat or simple transition area without obvious local complex features. The first mesh division strategy refers to the special mesh division rules that are pre-formulated for each type of typical feature and meet the mesh quality standards.

[0043] In some embodiments, determining a first grid partitioning strategy for typical feature regions and partitioning the typical feature regions into grids based on the first grid partitioning strategy includes: determining the feature type of the typical feature regions; determining the grid partitioning strategy corresponding to each type based on the feature type and a preset grid partitioning strategy table; and partitioning the typical feature regions into grids based on the corresponding grid partitioning strategy.

[0044] Specifically, the geometric structure type of the current typical feature region is determined based on the feature recognition results. For example, if the current typical feature region is a circular boss, the meshing strategy corresponding to the circular boss in the meshing strategy table is the mesh flow direction adjustment method. If the current typical feature region is a square boss with holes, the meshing strategy corresponding to the square boss with holes in the meshing strategy table is the geometric cleanup method. For non-typical feature regions, this embodiment of the invention uses the batchmesh function of ANSA software to perform meshing. Finally, the meshing result of the shell to be detected is obtained from the meshing results of the typical feature regions and the meshing results of the non-typical feature regions.

[0045] Furthermore, in some embodiments, after meshing typical feature regions based on the first meshing strategy and meshing atypical feature regions based on finite element software, the method further includes: performing quality assessment on the meshing results of each plate and shell structure based on a preset mesh quality detection method; if the quality assessment result meets the preset mesh quality detection standard, the meshing result of the plate and shell structure is output; otherwise, the meshing result of the plate and shell structure is meshed and optimized until it meets the preset mesh quality detection standard.

[0046] Specifically, the completed mesh is subjected to quality inspection. If the warpage of the mesh is greater than a preset threshold or the aspect ratio exceeds a preset ratio threshold, the mesh is considered to fail to meet the quality inspection standards. In this case, the mesh needs to be corrected and optimized. Methods may include node smoothing optimization, local mesh reconstruction, mesh topology correction, geometric alignment correction, local refinement, and coarsening correction. Then, the quality assessment is re-executed until all meshes meet the preset quality indicators such as warpage and aspect ratio, and finally, a qualified plate and shell structure mesh is output.

[0047] To enable those skilled in the art to better understand the plate and shell structure mesh division method of the present invention, the following description is provided in conjunction with specific embodiments.

[0048] Figure 7 This is a schematic diagram of a plate and shell structure mesh generation method according to a specific embodiment of the present invention, as shown below. Figure 7 As shown, firstly, typical feature acquisition and labeling are performed to collect and classify typical features that are prone to distortion in the plate and shell structure, providing a data foundation for model training. Then, the deep learning model training and optimization phase begins. Through dataset partitioning, data preprocessing, model building, training, and testing optimization, a deep learning model with accurate typical feature recognition capabilities is obtained. Next, typical feature recognition and mesh generation are performed. Using a pre-defined mesh generation strategy specific to typical features, combined with the model recognition results, high-quality mesh generation of typical feature regions is completed. Finally, for non-typical feature regions, finite element software is used to automatically generate the mesh, and quality inspection and correction optimization are performed on the entire mesh until all meshes meet the preset standards, outputting the final mesh generation result for the plate and shell structure.

[0049] Furthermore, to verify the effectiveness of the plate and shell structure mesh generation method of the present invention, the following explanation will be provided through comparative results. Figure 8 This is a schematic diagram comparing typical feature mesh division results according to a specific embodiment of the present invention, such as... Figure 8 As shown, for the mesh generation results of the typical features of a circular boss, the existing technology uses the automatic generation function of traditional finite element software (such as batchmesh) to generate the mesh. The mesh element distribution is relatively sparse, and the element shape shows obvious irregular trapezoids or wedges in the edge area of ​​the circular boss. The mesh orientation is not regular enough, and the fit between the element edge and the circular contour is poor. The mesh generation method of the plate and shell structure of the present invention has a denser and more regular mesh element distribution. The core area adopts a structured mesh arrangement of approximately squares, the edge area fits the circular contour, the mesh flow lines are smoother, and the overall element shape is more uniform, which significantly improves the mesh quality of this typical feature area.

[0050] According to the plate and shell structure meshing method provided in the embodiments of the present invention, the feature structure image of the plate and shell to be detected is first acquired, and the feature structure image is intelligently recognized by a preset deep learning feature recognition model, and the regional feature recognition result of the plate and shell to be detected is output. Based on the recognition result, the typical feature regions and atypical feature regions of the plate and shell to be detected are accurately distinguished. A dedicated first meshing strategy is matched for the typical feature regions and high-precision mesh generation is performed. At the same time, the atypical feature regions are efficiently and automatically divided using finite element software. This solves the problem of low mesh quality qualification rate and low work efficiency in related technologies due to the overly complex geometry of the plate and shell structure, improves the meshing quality of typical feature regions, and improves the efficiency of finite element analysis.

[0051] Next, the plate and shell structure grid division device provided according to an embodiment of the present invention is described with reference to the accompanying drawings.

[0052] Figure 9 This is a block diagram of a plate and shell structure grid division device provided according to an embodiment of the present invention.

[0053] like Figure 9 As shown, the plate and shell structure mesh division device 10 includes: an acquisition module 100, an identification module 200, and a division module 300. The acquisition module 100 is used to acquire the feature structure image of the shell to be detected; the recognition module 200 is used to recognize the feature structure image based on a preset deep learning feature recognition model to obtain the regional feature recognition result of the shell to be detected; the division module 300 is used to determine the typical feature region and atypical feature region of the shell to be detected based on the regional feature recognition result, determine the first grid division strategy of the typical feature region, perform grid division of the typical feature region based on the first grid division strategy, and perform grid division of the atypical feature region based on finite element software.

[0054] Furthermore, in some embodiments, before recognizing the target structure image in the shell structure to be detected based on the pre-trained deep learning model, the recognition module 200 is further configured to: acquire a target feature image dataset; divide the target feature image dataset into a training set, a validation set, and a test set based on a preset division ratio; construct a target feature recognition neural network, input the training set into the target feature recognition neural network for training to obtain initial model parameters; based on the initial model parameters, input the validation set into the target neural network for performance evaluation, and adjust the initial model parameters according to the performance evaluation results until the joint loss function of the test set converges to obtain the optimal model parameters; based on the optimal model parameters, input the test set into the target neural network for model testing, and obtain a preset deep learning feature recognition model when the test results meet preset requirements.

[0055] Furthermore, in some embodiments, the identification module 200 is also used to: acquire a feature structure image of the target shell structure; process the feature structure image of the target shell structure based on a preset data processing method to obtain a normalized feature structure image; perform feature classification and annotation on the normalized feature structure image based on a pre-built feature classification library, and construct a target feature image dataset based on the feature classification and annotation results.

[0056] Furthermore, in some embodiments, the partitioning module 300 is specifically used for: determining the feature type of a typical feature region; determining the grid partitioning strategy corresponding to each type based on the feature type and a preset grid partitioning strategy table; and performing grid partitioning on the typical feature region based on the corresponding grid partitioning strategy.

[0057] Furthermore, in some embodiments, after meshing typical feature regions based on the first meshing strategy and meshing atypical feature regions based on finite element software, the meshing module 300 is further configured to: perform quality assessment on the meshing results of each plate and shell structure based on a preset mesh quality detection method; if the quality assessment result meets the preset mesh quality detection standard, then output the meshing result of the plate and shell structure; otherwise, perform mesh correction and optimization on the meshing result of the plate and shell structure until it meets the preset mesh quality detection standard.

[0058] It should be noted that the foregoing explanation of the plate and shell structure mesh division method embodiment also applies to the plate and shell structure mesh division device of this embodiment, and will not be repeated here.

[0059] According to the plate and shell structure mesh generation device provided in the embodiment of the present invention, the feature structure image of the plate and shell to be detected is first acquired, and the feature structure image is intelligently recognized by a preset deep learning feature recognition model, and the regional feature recognition result of the plate and shell to be detected is output. Based on the recognition result, the typical feature regions and atypical feature regions of the plate and shell to be detected are accurately distinguished. A dedicated first mesh generation strategy is matched for the typical feature regions and high-precision mesh generation is performed. At the same time, the atypical feature regions are efficiently and automatically divided using finite element software. This solves the problem of low mesh quality qualification rate and low work efficiency in related technologies due to the overly complex geometry of the plate and shell structure, improves the mesh generation quality of typical feature regions, and improves the efficiency of finite element analysis.

[0060] Figure 10 This is a schematic diagram of an electronic device provided according to an embodiment of the present invention. The electronic device may include: The memory 1001, the processor 1002, and the computer program stored on the memory 1001 and capable of running on the processor 1002.

[0061] When the processor 1002 executes the program, it implements the plate and shell structure mesh division method provided in the above embodiments.

[0062] Furthermore, electronic devices also include: Communication interface 1003 is used for communication between memory 1001 and processor 1002.

[0063] The memory 1001 is used to store computer programs that can run on the processor 1002.

[0064] The memory 1001 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0065] If the memory 1001, processor 1002, and communication interface 1003 are implemented independently, then the communication interface 1003, memory 1001, and processor 1002 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0066] Optionally, in a specific implementation, if the memory 1001, processor 1002, and communication interface 1003 are integrated on a single chip, then the memory 1001, processor 1002, and communication interface 1003 can communicate with each other through an internal interface.

[0067] The processor 1002 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention.

[0068] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described plate and shell structure meshing method.

[0069] In addition, embodiments of the present invention also provide a computer program product, including a computer program, which is executed to implement the above-described plate and shell structure meshing method.

[0070] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0071] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0072] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for mesh generation in a plate and shell structure, characterized in that, Includes the following steps: Acquire the feature structure image of the shell to be detected; The feature structure image is identified based on a preset deep learning feature recognition model to obtain the regional feature recognition result of the shell to be detected; Based on the regional feature recognition results, the typical and atypical feature regions of the shell to be detected are determined, and a first mesh division strategy for the typical feature regions is determined. The typical feature regions are then meshed based on the first mesh division strategy, and the atypical feature regions are meshed based on finite element software.

2. The method according to claim 1, characterized in that, Before identifying the target structure image in the shell structure to be detected based on a pre-trained deep learning model, the method further includes: Obtain the target feature image dataset; Based on a preset division ratio, the target feature image dataset is divided into a training set, a validation set, and a test set; Construct a target feature recognition neural network, and obtain the initial model parameters by inputting the training set into the target feature recognition neural network for training; Based on the initial model parameters, the validation set is input into the target neural network for performance evaluation, and the initial model parameters are adjusted according to the performance evaluation results until the joint loss function of the test set converges to obtain the optimal model parameters. Based on the optimal model parameters, the test set is input into the target neural network for model testing, and when the test results meet the preset requirements, the preset deep learning feature recognition model is obtained.

3. The method according to claim 2, characterized in that, The acquisition of the target feature image dataset includes: Obtain feature structure images of the target plate and shell structure; The feature structure image of the target plate and shell structure is processed based on a preset data processing method to obtain a normalized feature structure image. The normalized feature structure image is classified and labeled based on a pre-built feature classification library, and a target feature image dataset is constructed based on the feature classification and labeling results.

4. The method according to claim 1, characterized in that, The first grid partitioning strategy for determining the typical feature region, and the grid partitioning of the typical feature region based on the first grid partitioning strategy, includes: Determine the feature type of the typical feature region; Based on the feature type and the preset mesh partitioning strategy table, determine the mesh partitioning strategy corresponding to each type; The typical feature region is divided into grids based on the corresponding grid division strategy.

5. The method according to claim 1, characterized in that, After meshing the typical feature region based on the first meshing strategy and meshing the atypical feature region based on finite element software, the method further includes: The quality of each plate and shell structure mesh division result is assessed based on a preset mesh quality detection method. If the quality assessment result meets the preset mesh quality inspection standard, the mesh division result of the plate and shell structure is output; otherwise, the mesh division result of the plate and shell structure is corrected and optimized until it meets the preset mesh quality inspection standard.

6. A plate and shell structure grid dividing device, characterized in that, include: The acquisition module is used to acquire feature structure images of the shell to be detected; The recognition module is used to recognize the feature structure image based on a preset deep learning feature recognition model to obtain the regional feature recognition result of the shell to be detected. The segmentation module is used to determine the typical and atypical feature regions of the shell to be detected based on the regional feature recognition results, determine the first mesh segmentation strategy of the typical feature regions, perform mesh segmentation on the typical feature regions based on the first mesh segmentation strategy, and perform mesh segmentation on the atypical feature regions based on finite element software.

7. The apparatus according to claim 6, characterized in that, Before identifying the target structure image in the shell structure to be detected based on a pre-trained deep learning model, the identification module is further configured to: Obtain the target feature image dataset; Based on a preset division ratio, the target feature image dataset is divided into a training set, a validation set, and a test set; Construct a target feature recognition neural network, and obtain the initial model parameters by inputting the training set into the target feature recognition neural network for training; Based on the initial model parameters, the validation set is input into the target neural network for performance evaluation, and the initial model parameters are adjusted according to the performance evaluation results until the joint loss function of the test set converges to obtain the optimal model parameters. Based on the optimal model parameters, the test set is input into the target neural network for model testing, and when the test results meet the preset requirements, the preset deep learning feature recognition model is obtained.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, the processor executing the program to implement the plate and shell structure meshing method as described in any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the plate and shell structure meshing method as described in any one of claims 1-5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the plate and shell structure meshing method as described in any one of claims 1-5.

Images