Plastic part meshing method and device, electronic equipment and storage medium

Abstract

This application relates to a method, apparatus, electronic device, and storage medium for mesh generation of plastic parts, comprising: generating a point cloud dataset based on the geometric data of the plastic part, and determining a target substrate region and a target feature region based on the point cloud dataset; generating a first mid-surface geometry based on the target substrate region, and generating a mid-surface point cloud based on the first mid-surface geometry and the target feature region, and generating a second mid-surface geometry based on the mid-surface point cloud; meshing the second mid-surface geometry to obtain a feature mesh, and meshing the first mid-surface geometry based on the contour line between the target substrate region and the target feature region and the feature mesh to obtain a substrate mesh, and obtaining a final mesh based on the feature mesh and the substrate mesh. This solves the problems of limited mid-surface generation accuracy, insufficient mesh pass rate, and low automation in related technologies when meshing plastic parts with complex geometric features, and effectively improves the quality and efficiency of plastic part mesh generation.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a method, apparatus, electronic device, and storage medium for mesh division of plastic parts. Background Technology

[0002] Currently, plastic components are widely used in vehicle structures. In the computer-aided engineering (CAE) stage of vehicle structural design, the quality of mesh generation for plastic components directly affects the accuracy of analysis results and computational efficiency, thus impacting vehicle safety and comfort.

[0003] In related technologies, plastic parts are mainly meshed manually. Operators use commercial software such as HyperMesh to manually adjust the shape, size, and density of the mesh based on their experience and professional knowledge to meet the analysis requirements.

[0004] However, when performing mesh generation on plastic parts with complex geometric features, the relevant technologies suffer from problems such as limited accuracy in mid-surface generation, insufficient mesh qualification rate, and low degree of automation, which urgently need to be addressed. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and storage medium for mesh generation of plastic parts, which solves the problems of limited mid-surface generation accuracy, insufficient mesh qualification rate, and low degree of automation in related technologies when meshing plastic parts with complex geometric features. It can effectively improve the quality and efficiency of mesh generation of plastic parts.

[0006] The first aspect of this application provides a method for mesh generation in plastic parts, including the following steps: Obtain geometric data of plastic parts from multiple vehicle models, generate point cloud datasets based on the geometric data of plastic parts, and determine the target substrate region and target feature region based on the point cloud datasets; A first mid-surface geometry is generated based on the target substrate region, and a mid-surface point cloud is generated based on the first mid-surface geometry and the target feature region, and a second mid-surface geometry is generated based on the mid-surface point cloud. The second mid-surface geometry is meshed to obtain a feature mesh, and the first mid-surface geometry is meshed to obtain a substrate mesh based on the contour line between the target substrate region and the target feature region and the feature mesh. The final mesh is obtained based on the feature mesh and the substrate mesh.

[0007] Optionally, in some embodiments, determining the target substrate region and the target feature region based on the point cloud dataset includes: Feature extraction is performed on each point cloud in the point cloud dataset, and point feature mapping is performed based on the feature extraction results to obtain the classification label of each point cloud; The initial substrate region and initial feature region are determined based on the classification label of each point cloud. Multiple local sub-point clouds are determined based on the initial feature regions, and coarse registration and fine registration are performed on the multiple local sub-point clouds to obtain the registered feature regions. Determine the voxelized crossover ratio (VCR) between the registered feature region and the standard feature region, and determine whether the VCR is greater than or equal to a first preset threshold. If the voxel crossover ratio is greater than or equal to the first preset threshold, the registered feature region is taken as the target feature region, and the target substrate region is determined based on the target feature region and the initial substrate region.

[0008] Optionally, in some embodiments, after determining the voxelized cross-union ratio between the registered feature regions and the standard feature regions, the method further includes: If the voxelized crossover ratio is less than the first preset threshold, the face separation category of the registered feature region is determined based on the preset face identifier mapping relationship; Adjust the registered feature regions according to the surface separation category, and re-execute the step of determining the voxelized crossover ratio between the registered feature regions and the standard feature regions until the new voxelized crossover ratio is greater than or equal to the first preset threshold.

[0009] Optionally, in some embodiments, the voxelized crossover ratio is: ; in ， For voxelization crossover ratio; The set of voxels occupied by the registered feature regions; The set of voxels occupied by the standard feature region; The number of voxels occupied by both the registered feature region and the standard feature region; The total number of voxels occupied by at least one of the registered feature regions and the standard feature region.

[0010] Optionally, in some embodiments, generating a mid-surface point cloud based on the target feature region according to the first mid-surface geometry includes: Based on the geometry of the first mid-surface, multiple displacement field vectors corresponding to the target feature region are determined. Generate mid-surface point clouds based on multiple displacement field vectors.

[0011] A second aspect of this application provides a plastic part mesh dividing device, comprising: The segmentation module is used to acquire geometric data of plastic parts from multiple vehicle models, generate point cloud datasets based on the geometric data of plastic parts, and determine the target substrate region and target feature region based on the point cloud datasets. The first generation module is used to generate a first mid-surface geometry based on the target substrate region, and generate a mid-surface point cloud based on the first mid-surface geometry and the target feature region, and generate a second mid-surface geometry based on the mid-surface point cloud. The second generation module is used to mesh the second mid-surface geometry to obtain a feature mesh, and to mesh the first mid-surface geometry to obtain a substrate mesh based on the contour line between the target substrate region and the target feature region and the feature mesh, and to obtain the final mesh based on the feature mesh and the substrate mesh.

[0012] Optionally, in some embodiments, the division of modules is specifically used for: Feature extraction is performed on each point cloud in the point cloud dataset, and point feature mapping is performed based on the feature extraction results to obtain the classification label of each point cloud; The initial substrate region and initial feature region are determined based on the classification label of each point cloud. Multiple local sub-point clouds are determined based on the initial feature regions, and coarse registration and fine registration are performed on the multiple local sub-point clouds to obtain the registered feature regions. Determine the voxelized crossover ratio (VCR) between the registered feature region and the standard feature region, and determine whether the VCR is greater than or equal to a first preset threshold. If the voxel crossover ratio is greater than or equal to the first preset threshold, the registered feature region is taken as the target feature region, and the target substrate region is determined based on the target feature region and the initial substrate region.

[0013] Optionally, in some embodiments, after determining the voxelized crossover ratio (CLORD) between the registered feature regions and the standard feature regions, the first generation module is further configured to: If the voxelized crossover ratio is less than the first preset threshold, the face separation category of the registered feature region is determined based on the preset face identifier mapping relationship; Adjust the registered feature regions according to the surface separation category, and re-execute the step of determining the voxelized crossover ratio between the registered feature regions and the standard feature regions until the new voxelized crossover ratio is greater than or equal to the first preset threshold.

[0014] Optionally, in some embodiments, the voxelized crossover ratio is: ; in, For voxelization crossover ratio; The set of voxels occupied by the registered feature regions; The set of voxels occupied by the standard feature region; The number of voxels occupied by both the registered feature region and the standard feature region; The total number of voxels occupied by at least one of the registered feature regions and the standard feature region.

[0015] Optionally, in some embodiments, the first generation module is specifically used for: Based on the geometry of the first mid-surface, multiple displacement field vectors corresponding to the target feature region are determined. Generate mid-surface point clouds based on multiple displacement field vectors.

[0016] A third aspect of this application 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 plastic part mesh division method described in the first aspect embodiment.

[0017] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the plastic part mesh division method described in the first aspect embodiment.

[0018] Therefore, the embodiments of this application have at least the following beneficial effects: (1) High-precision feature segmentation model: In this application embodiment, a deep learning model based on PointNet++ is constructed, which is trained using a mixed dataset and further combined with point cloud registration post-processing. The feature segmentation completeness of this model can reach at least 80.2%.

[0019] (2) High-fidelity mid-surface generation model: The embodiments of this application construct an end-to-end mid-surface point cloud generation model that integrates Transformer and ResNet architectures, which can achieve high-fidelity generation with an average mid-surface position deviation of less than 0.085mm.

[0020] (3) Improved efficiency and quality of plastic part mesh generation: This application embodiment is based on the HyperMesh software and integrates the above-mentioned AI module, which can realize the full-process automation from feature segmentation, mid-surface generation to final mesh generation. Compared with the batchmesh method of HyperMesh software in related technologies, this application embodiment can improve the mesh generation qualification rate to 93.7% and reduce the total working time by 85%.

[0021] (4) Independent and controllable engineering solutions: The embodiments of this application provide an efficient and accurate intelligent solution for mesh generation of plastic parts through the deep learning technology route. This solution not only helps to solve the automation bottleneck of mesh generation in current vehicle CAE simulation, but also lays a key technical foundation for building a domestic independent CAE software platform.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a method for mesh generation of plastic parts according to an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a plastic part according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a certain geometric category of a plastic part according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a feature segmentation result for a plastic part according to an embodiment of this application; Figure 5 This is a schematic diagram of a structure with a sudden change in circumferential dimensions according to an embodiment of this application; Figure 6 This is a schematic diagram of a point cloud registration process according to an embodiment of this application; Figure 7 This is a schematic diagram of a first mid-surface geometry provided according to an embodiment of this application; Figure 8 This is a schematic diagram of a second mid-surface geometry provided according to an embodiment of this application; Figure 9 This application provides a training and validation loss curve for a mid-surface generation model according to one embodiment of the present application. Figure 10 This is a structural schematic diagram of a feature of a certain plastic part according to an embodiment of this application; Figure 11 This is a comparative diagram showing the mid-surface generation results of the mid-surface generation method of this application and related technologies; Figure 12 This is a schematic diagram of a feature mesh structure according to an embodiment of this application; Figure 13 This is a schematic diagram of a substrate mid-surface cutting structure according to an embodiment of this application; Figure 14 This is a schematic diagram of a substrate mesh structure according to an embodiment of this application; Figure 15 This is a schematic diagram of a final mesh structure provided according to an embodiment of this application; Figure 16 This is a schematic diagram of the structure of a plastic part for a passenger vehicle according to an embodiment of this application; Figure 17A comparative schematic diagram showing the local mesh quality generated by the mesh generation method of this application and related technologies; Figure 18 This is a flowchart of a deep learning-based mesh generation process for plastic parts, according to one embodiment of this application. Figure 19 This is a block diagram of a plastic part grid dividing device according to an embodiment of this application; Figure 20 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application are described in detail below. Examples of these embodiments are shown 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 this application, and should not be construed as limiting this application.

[0025] The following description, with reference to the accompanying drawings, outlines a method, apparatus, electronic device, and storage medium for meshing plastic parts according to embodiments of this application. Addressing the problems of limited mid-surface generation accuracy, insufficient mesh pass rate, and low automation in meshing plastic parts with complex geometric features, as mentioned in the background art, this application provides a method for meshing plastic parts. This method generates a point cloud dataset based on the geometric data of the plastic part, thereby determining the target substrate region and the target feature region. A first mid-surface geometry is generated based on the target substrate region, and a second mid-surface geometry is generated by combining it with the target feature region. The second mid-surface geometry is meshed to obtain a feature mesh, and the first mid-surface geometry is meshed using the contour line between the target substrate region and the target feature region to obtain a substrate mesh. Finally, the final mesh is obtained based on the feature mesh and the substrate mesh. This solves the problems of limited mid-surface generation accuracy, insufficient mesh pass rate, and low automation in meshing plastic parts with complex geometric features, effectively improving the quality and efficiency of plastic part meshing.

[0026] Specifically, Figure 1 A flowchart illustrating the plastic part mesh generation method provided in this application embodiment.

[0027] like Figure 1 As shown, the method for meshing plastic parts includes the following steps: In step S101, geometric data of plastic parts of multiple vehicle models are acquired, and a point cloud dataset is generated based on the geometric data of the plastic parts. The target substrate area and the target feature area are determined based on the point cloud dataset.

[0028] In some embodiments, determining the target substrate region and the target feature region based on the point cloud dataset includes: extracting features from each point cloud in the point cloud dataset and performing point feature mapping based on the feature extraction results to obtain a classification label for each point cloud; determining the initial substrate region and the initial feature region based on the classification label of each point cloud; determining multiple local sub-point clouds based on the initial feature region and performing coarse registration and fine registration processing on the multiple local sub-point clouds to obtain the registered feature region; determining the voxel cross-union ratio (VCR) between the registered feature region and the standard feature region, and determining whether the VCR is greater than or equal to a first preset threshold; if the VCR is greater than or equal to the first preset threshold, then the registered feature region is taken as the target feature region, and the target substrate region is determined based on the target feature region and the initial substrate region.

[0029] Among them, the geometric data of the plastic part is a set of three-dimensional geometric dimensions and structural parameters of the plastic part; the point cloud dataset is a set of three-dimensional point clouds on the surface of the plastic part; the target substrate region is the effective target region belonging to the substrate in the point cloud dataset; the target feature region is the effective target region belonging to the feature in the point cloud dataset; the classification label of each point cloud is the annotation information used to distinguish the substrate and the feature; the initial substrate region is the initially extracted substrate point cloud region; the initial feature region is the feature point cloud region to be registered obtained from the initial localization; the local sub-point cloud is a local feature subset extracted from the original point cloud; the registered feature region is the initial feature region after registration; the standard feature region is the pre-established standard feature point cloud region; the target substrate region is the pre-established standard substrate point cloud region; and the first preset threshold is the critical value used to determine the registration accuracy and the validity of the region.

[0030] Specifically, such as Figure 2 and Figure 3 As shown, Figure 2 This is a schematic diagram of the structure of a certain plastic part provided in one embodiment of this application. Figure 3 This application provides a schematic diagram of the structure of a certain geometric category of plastic parts according to one embodiment, wherein, Figure 3 (a) is a structural diagram of the "features" such as snaps and ribs in the geometry of plastic parts. Figure 3 (b) is a schematic diagram of the "substrate" structure in the geometry of the plastic part. The embodiments of this application can collect data from multiple mass-produced vehicle models, such as... Figure 2The geometric data of the plastic part shown is divided into triangular meshes with an average size of 1 mm using HyperMesh. To convert the plastic part geometric data into a point cloud dataset suitable for deep learning, this embodiment extracts the centroid of each triangular unit as the 3D coordinates of each point cloud and assigns the normal vector of each triangle to the normal vector of each point cloud. Simultaneously, the original geometric face number and its geometric position label corresponding to each point cloud are retained. Next, this embodiment randomly samples each 70 mm × 70 mm × 70 mm sliding window, unifying the number of points to 8192. Finally, each point cloud contains eight attributes: 3D coordinates, 3D normal vector, geometric face number, and geometric category. Through the above steps, this embodiment can construct a point cloud dataset containing 132,000 samples.

[0031] Furthermore, embodiments of this application can separate the initial substrate region from the initial feature region by inputting the point cloud dataset of the plastic part into a feature segmentation AI model. The segmentation result can be as follows: Figure 4 As shown, Figure 4 This is a schematic diagram of the structure of a feature segmentation result for a plastic part according to an embodiment of this application, wherein, Figure 4 (a) represents the initial substrate region in the plastic part. Figure 4 (b) represents the initial feature region in the plastic part. Specifically, this application embodiment can perform feature segmentation on a point cloud dataset using a PointNet++-based feature segmentation model. The input features are a point cloud dataset of 8192 points (coordinates and normal vectors), and the output features are the classification label (feature or substrate) for each point cloud. PointNet++ is an extended architecture of PointNet designed specifically for processing 3D point cloud data, capable of directly extracting and classifying features from unordered point cloud datasets. As shown in Table 1, the overall network architecture of this feature segmentation model mainly consists of three parts: an encoder, a decoder, and a prediction head. The encoder gradually downsamples to 16 points through farthest point sampling, increasing the feature dimension from 6 dimensions to 1024 dimensions. The decoder uses interpolation techniques and skip connection mechanisms to gradually restore the point cloud resolution to 8192 points. The prediction head maps the 128-dimensional point features output by the decoder to a 2-class probability distribution and generates the final classification label for each point cloud. Table 1 is a feature segmentation model network architecture table provided in one embodiment of this application.

[0032] Table 1

[0033] It should be understood that in Table 1, FPS stands for Farthest Point Sampling; MLP stands for Multi-Layer Perceptron; BN stands for Batch Normalization; ReLU stands for Rectified Linear Unit; Argmax stands for Argument of the Maximum; and B stands for Batch Size, which is the number of samples simultaneously input into the feature segmentation model.

[0034] Furthermore, embodiments of this application can evaluate the performance of the feature segmentation model using two metrics: mean Intersection over Union (mIoU) and Feature Segmentation Accuracy (FSA). The mean Intersection over Union (mIoU) can be: ; ; in, Average within each instance IoU ; In the first i The number of categories that actually exist in each instance (ignoring categories that do not exist in the instance); For the first i Category in each instance a The intersection and union ratio; i Assign instance number; a Category labels; The average intersection-union ratio of instances; The total number of object instances in the test set; For the first i The average crossover / union ratio of the instances.

[0035] The Intersection over Union (IoU) of a single class can be: ; in, For single-class intersection and union ratio ( IoU ); For category a The number of true positive points; For category a The number of false positives; For category a The number of false negatives; a For category labels.

[0036] The feature segmentation accuracy (FSA) can be: ; in, For feature segmentation accuracy; The total number of features that are correctly and completely segmented; This represents the total number of geometric features.

[0037] Furthermore, the performance evaluation of the feature segmentation model in this embodiment is shown in Table 2. As shown in Table 2, the FSA is 80.2% and the mIoU is 84.6%. This result demonstrates that the feature segmentation model has good robustness and generalization ability for feature segmentation; however, its output may still have incomplete feature segmentation. Therefore, this embodiment can introduce a post-processing method based on point cloud registration to improve the feature segmentation completeness. Table 2 is a performance comparison table of different point cloud dataset models provided in one embodiment of this application.

[0038] Table 2

[0039] Furthermore, the post-processing method based on point cloud registration includes the following three steps: (1) Input preprocessing. In this embodiment, the initial feature region obtained from preliminary segmentation by the feature segmentation model is first loaded. Then, based on the abrupt size changes in the circumferential contour of the point cloud, it is automatically divided into multiple local sub-point clouds. For example... Figure 5 As shown, Figure 5 This is a schematic diagram of a circumferential size change structure provided in one embodiment of this application. If N size change points are identified, the initial feature region is correspondingly divided into N independent local sub-point clouds.

[0040] (2) Multi-stage point cloud registration. For each local sub-point cloud, this embodiment of the application can traverse a predefined feature library and employ a two-stage registration strategy to perform multi-stage point cloud registration. For example... Figure 6 As shown, Figure 6 This is a schematic diagram illustrating a point cloud registration process according to an embodiment of this application. This embodiment first uses the Random Sample Consensus algorithm (RANSAC) for coarse registration to solve the translation problem, then uses the Iterative Closest Point algorithm (ICP) for fine registration to solve the rotation problem; and evaluates the quality of the registered feature regions using the Voxel Intersection over Union (VoU) to quantify the spatial overlap between the registered feature regions and predefined features. The VoU can be: ; in, The cross-union ratio is the voxelized crossover ratio, which ranges from [0, 1]. The closer the value is to 1, the higher the spatial overlap between the two point clouds. The set of voxels occupied by the registered feature regions; The set of voxels occupied by the standard feature region; The number of voxels occupied by the registered feature region and the standard feature region (intersection). The total number of voxels (union) occupied by at least one of the registered feature regions and the standard feature region.

[0041] In step S102, a first mid-surface geometry is generated based on the target substrate region, and a mid-surface point cloud is generated based on the first mid-surface geometry based on the target feature region, and a second mid-surface geometry is generated based on the mid-surface point cloud.

[0042] In some embodiments, generating a mid-surface point cloud based on the target feature region based on the first mid-surface geometry includes: determining multiple displacement field vectors corresponding to the target feature region based on the first mid-surface geometry; and generating a mid-surface point cloud based on the multiple displacement field vectors.

[0043] Wherein, the first mid-surface geometry is the mid-surface geometry corresponding to the substrate region; the mid-surface point cloud is the basic point cloud for constructing the second mid-surface geometry; the second mid-surface geometry is the mid-surface geometry corresponding to the feature region; and the displacement field vector is a vector used to represent the pose offset of the target feature region.

[0044] Specifically, in related technologies, when processing complex features of plastic parts, such as the Midsurface module of HyperMesh, incomplete mid-surface generation and large positional deviations often occur due to the discontinuity of geometric topology and abrupt changes in thickness, severely restricting the stability and simulation accuracy of automated processing. To address this, this application presents an end-to-end deep learning mid-surface point cloud generation model. This model takes the point cloud data of the original surface as input and directly maps the displacement field vector of each point cloud to the mid-surface position by predicting it. Specifically, the input point cloud contains the three-dimensional coordinates and unit normal vector of each point cloud (a total of 6-dimensional features), and the output consists of multiple displacement field vectors, which are superimposed point by point to obtain the mid-surface point cloud. The mid-surface point cloud can be: ; in, Dot the clouds in the middle; The input point cloud, i.e. the set of points on the original surface of the plastic part, is an N×6 matrix, where N is the number of points. Each point contains 6-dimensional features: the first 3 dimensions are the three-dimensional coordinates of the point, and the last 3 dimensions are the unit normal vector at the point. This is a slicing operation on the input point cloud matrix. `:` indicates retrieving all rows (i.e., all points), and `:3` indicates retrieving the first 3 columns. Therefore... Indicates from The portion extracted from the data, containing only the three-dimensional coordinates of each point; The displacement field predicted by the mid-surface generation model is an N×3 matrix that predicts a three-dimensional displacement vector for each point in the input point cloud. This displacement vector indicates the direction and distance required for the point to move from the original surface to the corresponding mid-surface position.

[0045] Furthermore, such as Figure 7 and Figure 8 As shown, Figure 7 This is a schematic diagram of a first mid-surface geometry provided in one embodiment of this application. Figure 8 This is a schematic diagram of a second mid-surface geometry provided as an embodiment of this application. Embodiments of this application can generate structures such as... using the inherent functions of HyperMesh software. Figure 7 The first mid-surface geometry (i.e., the mid-surface of the substrate) is shown, and the mid-surface point cloud corresponding to the target feature region is generated from the first mid-surface geometry. Then, the geometry reconstruction function of HyperMesh is used to convert the mid-surface point cloud into a shape like... Figure 8 The second mid-surface geometry is shown. Specifically, embodiments of this application can use a Transformer encoder as the backbone network to capture common local abrupt features in plastic parts. The self-attention mechanism of the Transformer encoder can establish long-range dependencies between points, thereby overcoming the feature fragmentation problem caused by the limited receptive field of convolutional networks in related technologies. In addition, to further alleviate the gradient vanishing problem in deep network training, embodiments of this application can integrate the residual connection structure of ResNett into the Transformer encoder to enhance information transmission and training stability. The detailed structure of the mid-surface generation model is shown in Table 3. The mid-surface generation model mainly includes a feature embedding layer, an 8-layer Transformer encoder (each layer contains multi-head self-attention and residual connections), and an output layer. The input point cloud is mapped to a 1024-dimensional feature space through the embedding layer, and then multi-level global geometric features are extracted by the encoder. Finally, the displacement field is regressed by the output layer to complete the mid-surface point cloud reconstruction. Table 3 is a network architecture table of a mid-surface generation model provided by an embodiment of this application.

[0046] Table 3

[0047] It should be understood that B is the batch size, N is the number of points, Linear is a fully connected layer, LayerNorm represents layer normalization, MultiHeadAttention represents a multi-head attention mechanism, and Residual Connection represents a residual connection.

[0048] Furthermore, to simultaneously ensure the global shape consistency and local geometric smoothness of the generated mid-surfaces, this embodiment of the application can use a composite loss function consisting of global distance loss and local distance loss to train the mid-surface generation model. Specifically, the model is trained on an NVIDIA RTX 4090 GPU with a batch size of 16 and the AdamW optimizer selected, for a total of 200 training epochs. The dataset is divided into 800 training samples and 200 validation samples. The loss variation during training can be as follows: Figure 9 As shown, Figure 9 The training and validation loss curves of a mid-surface generation model provided in one embodiment of this application show that both the training and validation losses steadily decrease without significant overfitting. By the 67th round, the validation set loss drops to 0.085 mm, meeting the engineering accuracy requirement (deviation ≤ 0.1 mm). The composite loss function can be: ; in, For the total loss, The overall distance loss is used to constrain the global shape matching between the predicted point cloud and the target point cloud; This is a local distance loss used to improve the local smoothness and continuity of mid-surface point clouds; =0.2 , =0.8 All are weighting coefficients.

[0049] in, It can be: ; in, This represents the overall distance loss; For predicting point clouds; To predict any point in the point cloud; To predict the number of point clouds; For the target point cloud; For any point in the target point cloud; The target number of point clouds; The squared Euclidean distance between the predicted point cloud and the target point cloud is given.

[0050] in, It can be: ; in, This is a local distance loss; A random sample subset (sampling ratio 50%); For the first i One sampling point; In order to in the target point cloud Let be the centroid of the three nearest neighbors in the neighborhood; The unit normal vector of the neighborhood plane; The perpendicular distance from the point to the plane.

[0051] Furthermore, in this embodiment of the application, the predicted mid-surface point cloud can be imported into HyperMesh software. Through its surface reconstruction function, the predicted mid-surface point cloud is first converted into a triangular mesh, and then a complete and continuous second mid-surface geometry is generated.

[0052] Furthermore, such as Figure 10 and Figure 11 As shown, Figure 10 This is a structural schematic diagram of a feature of a certain plastic part provided in one embodiment of this application. Figure 11 This diagram illustrates a comparison of the mid-surface generation results of the method described in this application and related technologies. Embodiments of this application can select, for example... Figure 10 The complex plastic part features shown were subjected to mid-surface generation tests, and the results were compared with the Midsurface function of HyperMesh software. Figure 11 As shown, the second mid-surface geometry generated by the mid-surface generation model proposed in this application completely covers the target feature region, has good integrity, and high positional accuracy. In contrast, the result generated by HyperMesh software has problems such as the inability to generate stepped surfaces and missing connections.

[0053] In step S103, the second mid-surface geometry is meshed to obtain a feature mesh, and the first mid-surface geometry is meshed to obtain a substrate mesh based on the contour line between the target substrate region and the target feature region and the feature mesh. The final mesh is obtained based on the feature mesh and the substrate mesh.

[0054] The final mesh is the mesh model of the overall plastic part after the feature mesh and the substrate mesh are merged.

[0055] Specifically, such as Figures 12 to 15 As shown, Figure 12 This is a schematic diagram of a feature mesh structure provided in one embodiment of this application; Figure 13 This is a schematic diagram of a substrate mid-surface cutting structure provided in one embodiment of this application; Figure 14 This is a schematic diagram of a substrate mesh structure provided in one embodiment of this application; Figure 15This is a schematic diagram of the structure of a final mesh provided in one embodiment of this application. This embodiment can be encapsulated and integrated through the secondary development interface of HyperMesh software, thereby achieving full automation from feature segmentation to final mesh generation. Specifically, this embodiment can first perform the following steps on the geometry of the second mid-surface: Figure 12 The independent mesh division is shown; next, in order to ensure the compatibility of the substrate mesh and the feature mesh at the connection, this embodiment of the application needs to perform the first mid-surface geometry as shown in the figure. Figure 13 The cut shown is then performed on the geometry of the first mid-surface after the cut, as follows: Figure 14 The meshing is shown; finally, by moving the nodes of the feature mesh at the connection points and aligning them with the corresponding nodes of the substrate mesh, seamless integration is achieved, thus completing the process as shown. Figure 15 The final grid shown.

[0056] Furthermore, to verify the effectiveness of the methods in the embodiments of this application, the embodiments of this application may select as follows: Figure 16 The plastic part of a passenger vehicle shown is used as the test object. Figure 16 This document presents a schematic diagram of the structure of a plastic part for a passenger vehicle according to one embodiment of this application. A comparative analysis is conducted on the performance of the method in this embodiment and the batchmesh method of HyperMesh software in automated mesh generation. The mesh quality standards are shown in Table 4, which is a mesh quality standard table provided in one embodiment of this application.

[0057] Table 4

[0058] Furthermore, the comparison results of the automated mesh generation pass rate and working time are shown in Table 5. First, the mesh pass rate of the method in this embodiment is 93.7%, which is 8.4 percentage points higher than the 85.3% pass rate of the HyperMesh software batchmesh method. Second, the total working time required for this embodiment is 12 man-days, which is 85% less than the 80 man-days required by the HyperMesh software batchmesh method. In summary, the method in this embodiment not only significantly improves the mesh generation pass rate but also greatly reduces the reliance on manual intervention. Table 5 is a comparison table of the automatic mesh generation pass rate and working time provided by one embodiment of this application.

[0059] Table 5

[0060] Furthermore, such as Figure 17 As shown, Figure 17This diagram illustrates a comparison of the local mesh quality generated by the mesh generation method of this application and related technologies. The mesh generated by the batchmesh method in HyperMesh software suffers from problems such as element distortion, disordered mesh flow, and excessive aggregation of triangular elements. In contrast, the method in this application effectively avoids the above defects and generates a high-quality mesh with regular shape and precise connection.

[0061] Therefore, this application proposes and verifies a deep learning-based mesh generation method for plastic parts. This method, through the design of two core modules—feature segmentation and mid-surface generation—effectively solves the problems of low automation and insufficient mesh pass rate commonly found in commercial software such as HyperMesh and ANSA when processing complex plastic part features. Experimental results show that, compared with the batchmesh method of HyperMesh software in related technologies, the method of this application increases the mesh generation pass rate to 93.7% and reduces the total working time by 85%, significantly improving both mesh pass rate and efficiency.

[0062] Furthermore, to enable those skilled in the art to better understand the plastic part meshing method of this application, the following description, in conjunction with specific embodiments, addresses the possible situations that may arise when judging the registration accuracy of the feature regions after registration.

[0063] As one possible implementation, in some embodiments, after determining the voxelized intersection-union ratio (VUIR) between the registered feature region and the standard feature region, the method further includes: if the VUIR is less than a first preset threshold, determining the face separation category of the registered feature region based on a preset face identifier mapping relationship; adjusting the registered feature region according to the face separation category, and re-executing the step of determining the VUIR between the registered feature region and the standard feature region until the new VUIR is greater than or equal to the first preset threshold.

[0064] Among them, the preset surface identification mapping relationship is the correspondence between the pre-defined surface features and the identification information; the surface separation category is the classification result of dividing the registered feature regions according to geometric attributes; and the new voxelized intersection-union ratio is the voxelized intersection-union ratio recalculated after adjustment and optimization.

[0065] Specifically, in this embodiment, the face separation category can be divided into three categories for post-processing based on the face identifier mapping relationship between the registered feature region and the standard feature region: if the face separation category has "multiple faces", that is, it includes faces that do not belong to the feature, then the non-overlapping faces are classified back to the substrate, and the step of determining the voxel cross-union ratio between the registered feature region and the standard feature region is repeated; if the face separation category has "missing faces", that is, some feature faces are missing, then the corresponding faces are added from the adjacent region, and the step of determining the voxel cross-union ratio between the registered feature region and the standard feature region is repeated; if there are no non-overlapping faces, then it is determined that the feature has been correctly and completely segmented.

[0066] Therefore, this application proposes a deep learning-based method for mesh generation of plastic parts. By constructing a PointNet++ feature segmentation model and a Transformer-ResNet surface generation model, the method performs mesh generation of plastic parts, significantly improving the pass rate and efficiency of mesh generation, and has important theoretical and engineering application value.

[0067] Furthermore, to enable those skilled in the art to better understand the mesh generation method for plastic parts in this application, the following is combined with... Figure 18 Specific embodiments will be described below.

[0068] Figure 18 This application provides a flowchart of a deep learning-based plastic part mesh generation process as one embodiment.

[0069] like Figure 18 As shown, this deep learning-based mesh generation method for plastic parts includes the following steps: S1801 uses feature segmentation AI to segment the geometric data of plastic parts, generating target substrate area and target feature area.

[0070] S1802, through mid-surface generation AI, the first mid-surface geometry of the target substrate region is generated, and the second mid-surface geometry of the target feature region is generated.

[0071] S1803, the first mid-surface geometry is meshed to obtain the substrate mesh, and the second mid-surface geometry is meshed to obtain the feature mesh. The substrate mesh and the feature mesh are integrated to obtain the final mesh.

[0072] The plastic part meshing method proposed in this application can generate a point cloud dataset based on the geometric data of the plastic part, thereby determining the target substrate region and the target feature region; generate a first mid-surface geometry based on the target substrate region, and generate a second mid-surface geometry in combination with the target feature region; mesh the second mid-surface geometry to obtain a feature mesh, and mesh the first mid-surface geometry in combination with the contour line between the target substrate region and the target feature region to obtain a substrate mesh; finally, obtain the final mesh based on the feature mesh and the substrate mesh. This solves the problems of limited mid-surface generation accuracy, insufficient mesh pass rate, and low automation in related technologies when meshing plastic parts with complex geometric features, and can effectively improve the quality and efficiency of plastic part meshing.

[0073] Next, the plastic part grid dividing device proposed in the embodiments of this application is described with reference to the accompanying drawings.

[0074] Figure 19 This is a block diagram of a plastic part grid dividing device proposed in an embodiment of this application.

[0075] like Figure 19 As shown, the plastic part grid dividing device 10 includes: a dividing module 100, a first generating module 200, and a second generating module 300.

[0076] The system includes a segmentation module 100, which acquires geometric data of plastic parts from multiple vehicle models, generates a point cloud dataset based on the geometric data, and determines the target substrate region and the target feature region based on the point cloud dataset; a first generation module 200, which generates a first mid-surface geometry based on the target substrate region, generates a mid-surface point cloud based on the first mid-surface geometry and the target feature region, and generates a second mid-surface geometry based on the mid-surface point cloud; and a second generation module 300, which performs meshing on the second mid-surface geometry to obtain a feature mesh, performs meshing on the first mid-surface geometry based on the contour line between the target substrate region and the target feature region and the feature mesh to obtain a substrate mesh, and obtains the final mesh based on the feature mesh and the substrate mesh.

[0077] Optionally, in some embodiments, the segmentation module 100 is specifically used for: extracting features from each point cloud in the point cloud dataset, and performing point feature mapping based on the feature extraction results to obtain a classification label for each point cloud; determining an initial substrate region and an initial feature region based on the classification label of each point cloud; determining multiple local sub-point clouds based on the initial feature region, and performing coarse registration and fine registration processing on the multiple local sub-point clouds to obtain a registered feature region; determining the voxel cross-union ratio (VCR) between the registered feature region and the standard feature region, and determining whether the VCR is greater than or equal to a first preset threshold; if the VCR is greater than or equal to the first preset threshold, then using the registered feature region as the target feature region, and determining the target substrate region based on the target feature region and the initial substrate region.

[0078] Optionally, in some embodiments, after determining the voxelized intersection-union ratio (VUCR) between the registered feature region and the standard feature region, the first generation module 200 is further configured to: if the VUCR is less than a first preset threshold, determine the face separation category of the registered feature region based on a preset face identifier mapping relationship; adjust the registered feature region according to the face separation category, and re-execute the step of determining the VUCR between the registered feature region and the standard feature region until the new VUCR is greater than or equal to the first preset threshold.

[0079] Optionally, in some embodiments, the voxelized crossover ratio is: ; in, For voxelization crossover ratio; The set of voxels occupied by the registered feature regions; The set of voxels occupied by the standard feature region; The number of voxels occupied by both the registered feature region and the standard feature region; The total number of voxels occupied by at least one of the registered feature regions and the standard feature region.

[0080] Optionally, in some embodiments, the first generation module 200 is specifically used to: determine multiple displacement field vectors corresponding to the target feature region based on the first mid-surface geometry; and generate a mid-surface point cloud based on the multiple displacement field vectors.

[0081] It should be noted that the foregoing explanation of the plastic part mesh division method embodiment also applies to the plastic part mesh division device of this embodiment, and will not be repeated here.

[0082] The plastic part meshing apparatus proposed in this application can generate a point cloud dataset based on the geometric data of the plastic part, thereby determining the target substrate region and the target feature region; generate a first mid-surface geometry based on the target substrate region, and generate a second mid-surface geometry in combination with the target feature region; mesh the second mid-surface geometry to obtain a feature mesh, and mesh the first mid-surface geometry in combination with the contour line between the target substrate region and the target feature region to obtain a substrate mesh; finally, obtain the final mesh based on the feature mesh and the substrate mesh. This solves the problems of limited mid-surface generation accuracy, insufficient mesh pass rate, and low automation in related technologies when meshing plastic parts with complex geometric features, effectively improving the quality and efficiency of plastic part meshing.

[0083] Figure 20 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: Memory 2001, processor 2002, and computer program stored on memory 2001 and executable on processor 2002.

[0084] When processor 2002 executes the program, it implements the plastic part meshing method provided in the above embodiments.

[0085] Furthermore, the electronic device also includes: Communication interface 2003 is used for communication between memory 2001 and processor 2002.

[0086] Memory 2001 is used to store computer programs that can run on processor 2002.

[0087] The memory 2001 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0088] If the memory 2001, processor 2002, and communication interface 2003 are implemented independently, then the communication interface 2003, memory 2001, and processor 2002 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized into address buses, data buses, control buses, etc. For ease of representation, Figure 20The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0089] Optionally, in a specific implementation, if the memory 2001, processor 2002, and communication interface 2003 are integrated on a single chip, then the memory 2001, processor 2002, and communication interface 2003 can communicate with each other through an internal interface.

[0090] The processor 2002 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0091] This application embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following... Figure 1 The method for dividing plastic parts into grids is shown.

[0092] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. 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.

[0093] 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 application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0094] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0095] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0096] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0097] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0098] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0099] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for mesh division of plastic parts, characterized in that, include: Obtain geometric data of plastic parts from multiple vehicle models, generate a point cloud dataset based on the geometric data of the plastic parts, and determine the target substrate region and the target feature region based on the point cloud dataset; A first mid-surface geometry is generated based on the target substrate region, and a mid-surface point cloud is generated based on the first mid-surface geometry and the target feature region, and a second mid-surface geometry is generated based on the mid-surface point cloud. The second mid-surface geometry is meshed to obtain a feature mesh, and the first mid-surface geometry is meshed to obtain a substrate mesh based on the contour line between the target substrate region and the target feature region and the feature mesh. The final mesh is obtained based on the feature mesh and the substrate mesh.

2. The method according to claim 1, characterized in that, The step of determining the target substrate region and the target feature region based on the point cloud dataset includes: Feature extraction is performed on each point cloud in the point cloud dataset, and point feature mapping is performed based on the feature extraction results to obtain the classification label of each point cloud; The initial substrate region and initial feature region are determined based on the classification label of each point cloud. Multiple local sub-point clouds are determined based on the initial feature region, and coarse registration and fine registration are performed on the multiple local sub-point clouds to obtain the registered feature region. Determine the voxelized crossover ratio (VCR) between the registered feature region and the standard feature region, and determine whether the VCR is greater than or equal to a first preset threshold. If the voxelized crossover ratio is greater than or equal to the first preset threshold, then the registered feature region is taken as the target feature region, and the target substrate region is determined based on the target feature region and the initial substrate region.

3. The method according to claim 2, characterized in that, After determining the voxel crossover ratio between the registered feature region and the standard feature region, the method further includes: If the voxelized intersection-union ratio is less than the first preset threshold, the face separation category of the registered feature region is determined based on the preset face identifier mapping relationship; Adjust the registered feature region according to the surface separation category, and re-execute the step of determining the voxelized crossover ratio between the registered feature region and the standard feature region until the new voxelized crossover ratio is greater than or equal to the first preset threshold.

4. The method according to claim 2, characterized in that, The voxelization crossover ratio is: ； in, The voxelization crossover ratio; The set of voxels occupied by the registered feature region; The set of voxels occupied by the standard feature region; The number of voxels simultaneously occupied by the registered feature region and the standard feature region; The total number of voxels occupied by at least one of the registered feature regions and the standard feature regions.

5. The method according to claim 1, characterized in that, The step of generating a mid-surface point cloud based on the first mid-surface geometry and the target feature region includes: Based on the geometry of the first mid-surface, multiple displacement field vectors corresponding to the target feature region are determined; The mid-surface point cloud is generated based on the multiple displacement field vectors.

6. A plastic part grid dividing device, characterized in that, include: The segmentation module is used to acquire geometric data of plastic parts from multiple vehicle models, generate a point cloud dataset based on the geometric data of the plastic parts, and determine the target substrate region and the target feature region based on the point cloud dataset. The first generation module is used to generate a first mid-surface geometry based on the target substrate region, and generate a mid-surface point cloud based on the first mid-surface geometry and the target feature region, and generate a second mid-surface geometry based on the mid-surface point cloud. The second generation module is used to perform meshing on the second mid-surface geometry to obtain a feature mesh, and to perform meshing on the first mid-surface geometry to obtain a substrate mesh based on the contour line between the target substrate region and the target feature region and the feature mesh, and to obtain a final mesh based on the feature mesh and the substrate mesh.

7. The apparatus according to claim 6, characterized in that, The partitioning module is specifically used for: Feature extraction is performed on each point cloud in the point cloud dataset, and point feature mapping is performed based on the feature extraction results to obtain the classification label of each point cloud; The initial substrate region and initial feature region are determined based on the classification label of each point cloud. Multiple local sub-point clouds are determined based on the initial feature region, and coarse registration and fine registration are performed on the multiple local sub-point clouds to obtain the registered feature region. Determine the voxelized crossover ratio (VCR) between the registered feature region and the standard feature region, and determine whether the VCR is greater than or equal to a first preset threshold. If the voxelized crossover ratio is greater than or equal to the first preset threshold, then the registered feature region is taken as the target feature region, and the target substrate region is determined based on the target feature region and the initial substrate region.

8. The apparatus according to claim 7, characterized in that, After determining the voxel cross-union ratio between the registered feature region and the standard feature region, the first generation module is further configured to: If the voxelized intersection-union ratio is less than the first preset threshold, the face separation category of the registered feature region is determined based on the preset face identifier mapping relationship; Adjust the registered feature region according to the surface separation category, and re-execute the step of determining the voxelized crossover ratio between the registered feature region and the standard feature region until the new voxelized crossover ratio is greater than or equal to the first preset threshold.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the plastic part meshing method as described in any one of claims 1-5.

10. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by the processor, it implements the plastic part mesh generation method as described in any one of claims 1-5.

Images