Sheet metal part structure performance analysis method based on multi-scale modeling

Abstract

The invention provides a sheet metal part structure performance analysis method based on multi-scale modeling, and the method comprises the following steps: S1, carrying out the high-precision scanning of a solid sheet metal part, obtaining the point cloud data of the sheet metal part, and carrying out the preprocessing; s2, importing the preprocessed point cloud data into reverse engineering software for surface reconstruction, generating a three-dimensional digital model of the sheet metal part, and performing feature recognition and parameterization processing to extract critical dimension parameters; s3, performing grid division on the parameterized three-dimensional digital model, generating finite element grid models with different densities by adopting a multi-scale modeling method, and performing grid encryption in a key region; and S4, defining load and boundary conditions for the multi-scale model, selecting material attribute parameters, and performing structural performance simulation analysis on the multi-scale model under multiple working conditions by using a finite element analysis method. According to the invention, the performance indexes of strength, rigidity, deformation and the like of the sheet metal part under different load and boundary conditions can be comprehensively evaluated.

Description

Technical Field

[0001] The invention relates to a sheet metal part structural performance analysis method, in particular to a sheet metal part structural performance analysis method based on multi-scale modeling, and belongs to the technical field of sheet metal part structural analysis. Background Art

[0002] Sheet metal parts are a type of component widely used in modern manufacturing, and are mainly manufactured through stamping, bending, stretching and other processes of metal sheets. For sheet metal products such as electrical equipment cabinets and electronic chassis, structural performance analysis is very important for evaluating their reliability and safety. In many cases, such products are updated quickly, the original design data is incomplete, and structural analysis cannot be performed based on the original design documents. On the other hand, such products may encounter design changes, maintenance and renovation in actual use, resulting in differences in their structure from the original design. In these cases, it is challenging to conduct structural performance analysis of sheet metal parts based on reverse engineering.

[0003] The structural performance analysis of existing technologies is usually based on three-dimensional modeling. Due to the rich local product features of such sheet metal parts, full-size high-precision modeling will result in too much model data, which can easily lead to great difficulties in subsequent finite element meshing and solving. While surface fitting and detail simplification operations can reduce the complexity of the model, for areas with concentrated stress and strain, such as welded joints, bolted connections, stiffeners, etc., this simplified model is difficult to accurately describe local geometric details and mechanical behaviors, affecting the accuracy of structural performance analysis of key parts.

[0004] Therefore, for sheet metal products such as electrical equipment cabinets and electronic chassis, an efficient and flexible multi-scale modeling method is urgently needed to analyze their structural performance. Summary of the invention

[0005] Based on the above background, the purpose of the present invention is to provide a sheet metal structural performance analysis method based on multi-scale modeling to solve the problems described in the background technology.

[0006] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0007] A sheet metal structural performance analysis method based on multi-scale modeling, the method comprising the following steps:

[0008] Step S1, using three-dimensional scanning technology to perform high-precision scanning on the physical sheet metal part, obtain point cloud data of the sheet metal part, and pre-process the point cloud data;

[0009] Step S2, importing the pre-processed point cloud data into reverse engineering software for surface reconstruction, generating a three-dimensional digital model of the sheet metal part, and performing feature recognition and parameterization processing on the three-dimensional digital model to extract key dimensional parameters;

[0010] Step S3, meshing the parameterized three-dimensional digital model, generating finite element mesh models of different densities using a multi-scale modeling method, and performing mesh encryption in key areas to obtain a multi-scale model of the sheet metal part;

[0011] Step S4: define loads and boundary conditions for the multi-scale model, select material property parameters, and use the finite element analysis method to perform structural performance simulation analysis on the multi-scale model under multiple working conditions.

[0012] Preferably, the step S1 specifically includes:

[0013] Step S11, selecting a suitable 3D scanning device, and determining scanning parameters according to the size, shape complexity and required accuracy of the sheet metal part;

[0014] Step S12, performing surface treatment on the solid sheet metal part and fixing the solid sheet metal part in an appropriate position;

[0015] Step S13, scanning the physical sheet metal part at multiple angles and in multiple directions by means of a three-dimensional scanning device to obtain point cloud data of the physical sheet metal part;

[0016] Step S14: preprocessing the acquired point cloud data, including point cloud filtering, point cloud registration, point cloud downsampling and point cloud smoothing.

[0017] Preferably, step S2 specifically includes:

[0018] Step S21, importing the pre-processed point cloud data into reverse engineering software;

[0019] Step S22, performing surface reconstruction on the imported point cloud data to generate a three-dimensional digital model of the sheet metal part;

[0020] Step S23, performing post-processing operations of model repair, model simplification and model smoothing on the reconstructed three-dimensional digital model;

[0021] Step S24, performing feature recognition on the post-processed three-dimensional digital model, extracting key geometric features such as plane, round hole, bend, and stamping of the sheet metal part, classifying, annotating, and organizing the recognized features, and establishing a complete feature tree;

[0022] Step S25, perform parameterization processing on the identified features, extract key dimensional parameters such as plane size, circular hole diameter, bending radius, punching depth, etc., convert the geometric information of the features into parameterized dimensional constraints and relationships, and establish a parameterized three-dimensional digital model.

[0023] Preferably, the step S3 specifically includes:

[0024] Step S31, using an intelligent feature recognition method based on machine learning to geometrically clean and simplify the three-dimensional digital model;

[0025] Step S32, meshing the simplified three-dimensional digital model to generate a global scale finite element mesh model, and defining material property parameters of the sheet metal part;

[0026] Step S33, identifying stress concentration areas in the bending area, punching area and welding area of ​​the sheet metal part, performing local geometric extraction on the stress concentration area, generating an independent local sub-model, defining a local coordinate system, and establishing a geometric coupling relationship between the local sub-model and the global scale finite element mesh model;

[0027] Step S34, using a high-order unit type and a refined grid size to encrypt the grid of the local sub-model;

[0028] Step S35: perform error estimation, evaluate the quality and convergence of the current finite element mesh, and use an adaptive meshing strategy driven by an evolutionary algorithm based on key geometric feature information extracted by the intelligent feature recognition method to adjust the mesh density and unit size for areas with large errors, and repeat the error estimation and adaptive mesh encryption process until a preset convergence accuracy threshold is reached;

[0029] Step S36: embed the local sub-model into the global scale finite element mesh model to form a multi-scale model of the sheet metal part.

[0030] Preferably, the step S31 specifically includes:

[0031] Step S311, collecting a certain number of three-dimensional digital models of sheet metal parts of different types as original data sets; manually annotating the features of each three-dimensional digital model of the sheet metal part, identifying key features such as bending areas, punching areas and welding areas, and forming a data set with labels; converting the annotated three-dimensional digital model of the sheet metal part into a mesh model and performing mesh cleaning and repair; normalizing the mesh model, converting it into a voxel representation, and generating a three-dimensional voxel mesh;

[0032] Step S312: constructing a three-dimensional CNN network model for extracting multi-scale geometric features from a three-dimensional voxel grid, wherein the three-dimensional CNN network model includes two convolutional layers, two pooling layers, two fully connected layers, a multi-scale feature extraction fusion module, and an attention module;

[0033] Step S313, input the three-dimensional voxel grid annotated in step S311 into the three-dimensional CNN network model, train the model, and optimize the network parameters to minimize the feature prediction error; use the trained three-dimensional CNN network model to extract features from the three-dimensional digital model of the sheet metal part to obtain a feature vector representation of each voxel; convert the extracted voxel feature vector into a graph representation, with each voxel as a node of the graph, and establishing edge connections between adjacent voxels;

[0034] Step S314: construct a GNN network model for feature recognition and segmentation in the graph domain, wherein the GNN network model includes 2 graph convolution layers, 2 graph attention layers, 2 graph pooling layers, 1 global pooling layer and 1 fully connected layer;

[0035] Step S315, input the graph constructed in step S313 into the GNN network model, train the model, and optimize the network parameters to minimize the feature classification error; use the trained GNN network model to perform feature recognition and segmentation on the three-dimensional digital model of the sheet metal part to obtain the feature category of each voxel;

[0036] Step S316: According to the feature segmentation results of the GNN network model, the three-dimensional digital model is geometrically cleaned and simplified; for ordinary areas, the edge collapse algorithm is used to merge adjacent faces to simplify the mesh complexity; for key feature areas, the original geometric topological structure is retained without simplification; for symmetrical or periodic structures, the minimum repeating unit is identified and extracted, only the geometric information of the minimum repeating unit is retained, and periodic constraints are imposed on the unit boundaries to generate simplified partial models to reduce the scale of calculation.

[0037] Preferably, in the three-dimensional CNN network model, the first convolution layer is used to perform a convolution operation on the input data; the first pooling layer is used to perform a maximum pooling operation on the output of the first convolution layer to halve the size of the feature map; the multi-scale feature extraction fusion module is used to first perform convolution and pooling operations of three branches on the output of the first pooling layer in parallel, and then upsample the outputs of the three branches, and finally splice the upsampled results of the three branches in the feature dimension to obtain a fused multi-scale feature representation; the second convolution layer is used to perform a convolution operation on the multi-scale features; the attention module is used to perform an output of the second convolution layer. Feature recalibration and weight adjustment are performed to generate an attention weight map with the same size as the input feature map, and then the attention weight map is multiplied by the output of the second convolutional layer to obtain the feature map after attention enhancement; the second pooling layer is used to perform the maximum pooling operation on the feature map after attention enhancement to halve the size of the feature map; the first fully connected layer is used to flatten the output of the second pooling layer to obtain a high-dimensional feature vector, and map the feature vector to a high-dimensional hidden representation through a fully connected operation; the second fully connected layer is used to map the output of the first fully connected layer to a K-dimensional output vector through a fully connected operation, where K is the number of feature categories.

[0038] Preferably, in the GNN network model, the first graph convolution layer is used to perform graph convolution operations on the input data; the first graph attention layer is used to perform feature aggregation and update on the output of the first graph convolution layer; the first graph pooling layer is used to perform pooling operations on the output of the first graph attention layer; the second graph convolution layer is used to perform graph convolution operations on the output of the first graph pooling layer; the second graph attention layer is used to perform feature aggregation and update on the output of the second graph convolution layer; the second graph pooling layer is used to perform pooling operations on the output of the second graph attention layer; the global pooling layer is used to perform global pooling operations on the output of the second graph pooling layer, aggregate graph-level features into a vector representation, and obtain a global feature vector; the fully connected layer is used to map the output of the global pooling layer to the number of feature categories K through a fully connected operation, and obtain the probability distribution of each category.

[0039] Preferably, the step S36 specifically includes:

[0040] A position constraint relationship based on the coupling of node degrees of freedom is established, and the displacement coordination equation is applied on the interface between the local sub-model and the global-scale finite element mesh model to ensure the continuity of displacement and rotation. The load transfer mechanism between the local sub-model and the global-scale finite element mesh model is defined to achieve two-way interaction of multi-scale information, thereby forming a multi-scale model of sheet metal.

[0041] Preferably, the step S4 specifically includes:

[0042] Step S41, analyzing the static load, dynamic load and impact load encountered by the sheet metal in actual use, and determining the size, direction and action position of each load condition; applying the simplified and equivalent load condition to the corresponding position of the multi-scale model, defining boundary conditions such as displacement constraints and symmetry conditions, and simulating the connection and support mode between the sheet metal and the surrounding structure;

[0043] Step S42: Select an elastic-plastic model or an anisotropic model according to the material type of the sheet metal part, and input the elastic modulus, Poisson's ratio, yield strength and hardening curve of the material;

[0044] Step S43, selecting an implicit solver or an explicit solver to match different load conditions and material nonlinear characteristics, setting the time step, iterative convergence criterion and stress failure criterion, performing nonlinear finite element analysis on the multi-scale model, and obtaining the displacement field, stress field and strain field of the sheet metal under different conditions;

[0045] Step S44: Evaluate the strength, stiffness and stability of the sheet metal parts under different working conditions, and analyze the fracture failure mode and fatigue failure position of the sheet metal parts under different working conditions.

[0046] Compared with the prior art, the present invention has the following advantages:

[0047] The present invention provides a sheet metal structural performance analysis method based on multi-scale modeling, which supports structural performance simulation analysis under multiple working conditions by integrating three-dimensional scanning, reverse engineering, multi-scale modeling and finite element analysis. It can comprehensively evaluate the performance indicators such as strength, stiffness, deformation, etc. of sheet metal under different loads and boundary conditions. By defining key areas and performing mesh encryption, it can specifically analyze the local stress and strain concentration of sheet metal parts and evaluate potential damage and failure risks. DETAILED DESCRIPTION

[0048] The technical solution of the present invention is further described in detail below through specific embodiments. It should be understood that the implementation of the present invention is not limited to the following embodiments, and any form of modification and / or change made to the present invention will fall within the protection scope of the present invention.

[0049] In the present invention, unless otherwise specified, all parts and percentages are weight units, and the equipment and raw materials used can be purchased from the market or are commonly used in the art. The methods in the following embodiments, unless otherwise specified, are conventional methods in the art. The components or equipment in the following embodiments, unless otherwise specified, are universal standard parts or components known to those skilled in the art, and their structures and principles are known to those skilled in the art through technical manuals or conventional experimental methods.

[0050] The following detailed description is made to the embodiments of the present invention. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present invention. However, one or more embodiments may also be implemented by those skilled in the art without these specific details.

[0051] An embodiment of the present invention discloses a sheet metal structure performance analysis method based on multi-scale modeling, the method comprising the following steps:

[0052] Step S1, using 3D scanning technology to perform high-precision scanning on the physical sheet metal part, obtaining point cloud data of the sheet metal part, and preprocessing the point cloud data;

[0053] Step S2, importing the pre-processed point cloud data into reverse engineering software for surface reconstruction, generating a three-dimensional digital model of the sheet metal part, and performing feature recognition and parameterization processing on the three-dimensional digital model to extract key dimensional parameters;

[0054] Step S3, meshing the parameterized three-dimensional digital model, generating finite element mesh models of different densities using a multi-scale modeling method, and performing mesh encryption in key areas to obtain a multi-scale model of the sheet metal part;

[0055] Step S4: define loads and boundary conditions for the multi-scale model, select material property parameters, and use the finite element analysis method to perform structural performance simulation analysis on the multi-scale model under multiple working conditions.

[0056] This method provides a new solution for the structural performance analysis of sheet metal parts by integrating 3D scanning, reverse engineering, multi-scale modeling and finite element analysis. Through 3D scanning technology and reverse engineering software, the rapid conversion from physical sheet metal parts to digital models can be achieved, reducing the workload of manual modeling. The multi-scale modeling method uses grids of different densities in different areas to reduce the calculation scale of non-critical areas while ensuring the accuracy of key areas, thus balancing the calculation scale and analysis efficiency as a whole. This method supports structural performance simulation analysis under multiple working conditions, and can comprehensively evaluate the performance indicators of sheet metal parts such as strength, stiffness, deformation, etc. under different loads and boundary conditions. By defining key areas and performing mesh encryption, the local stress and strain concentration of sheet metal parts can be analyzed in a targeted manner to evaluate potential damage and failure risks.

[0057] Step S1 specifically includes:

[0058] Step S11, selecting a suitable 3D scanning device, such as a structured light 3D scanner, a laser scanner or a handheld scanner, and determining scanning parameters, such as scanning resolution, scanning distance and scanning angle, according to the size, shape complexity and required accuracy of the sheet metal part;

[0059] Step S12: Surface treatment of the solid sheet metal part, for example, spraying a sunscreen and attaching reflective marking points, to improve the surface scanning quality and feature recognition capability. Fix the solid sheet metal part in an appropriate position so that it remains stable during the scanning process;

[0060] Step S13, scanning the physical sheet metal part at multiple angles and in multiple directions by means of a three-dimensional scanning device to obtain point cloud data of the physical sheet metal part;

[0061] Step S14, preprocessing the acquired point cloud data, including point cloud filtering, point cloud registration, point cloud downsampling and point cloud smoothing. Point cloud filtering can adopt statistical filtering method or radius filtering method, point cloud registration can adopt ICP algorithm, point cloud downsampling can adopt uniform sampling method or curvature sampling method, and point cloud smoothing can adopt moving least squares method.

[0062] In steps S11-S14, this embodiment uses three-dimensional scanning technology to quickly and efficiently obtain high-precision point cloud data of the physical sheet metal part. The pre-processed point cloud data can better reflect the geometric features and detail information of the physical sheet metal part.

[0063] Step S2 specifically includes:

[0064] Step S21, importing the pre-processed point cloud data into reverse engineering software such as Geomagic Design X, CATIA, SolidWorks, etc.;

[0065] Step S22: reconstruct the surface of the imported point cloud data to generate a three-dimensional digital model of the sheet metal part. Use the surface fitting tools such as mesh fitting and NURBS surface fitting of the reverse engineering software to convert the discrete point cloud data into a continuous surface model. According to the geometric characteristics and shape complexity of the sheet metal part, select the least square method, B-spline surface and other fitting algorithms to obtain the three-dimensional digital model;

[0066] Step S23, performing post-processing operations of model repair, model simplification and model smoothing on the reconstructed three-dimensional digital model. Model repair is to repair defects such as small holes, gaps, and self-intersections in the model, model simplification is to remove unnecessary small protrusions, holes and other detail features, and model smoothing is to smooth the surface to reduce noise and jaggedness;

[0067] Step S24, performing feature recognition on the post-processed three-dimensional digital model, extracting key geometric features such as plane, round hole, bend, and stamping of the sheet metal part, classifying, annotating, and organizing the recognized features, and establishing a complete feature tree;

[0068] Step S25, perform parameterization processing on the identified features, extract key dimensional parameters such as plane size, circular hole diameter, bending radius, punching depth, etc., convert the geometric information of the features into parameterized dimensional constraints and relationships, and establish a parameterized three-dimensional digital model.

[0069] In steps S21-S25, this embodiment converts the preprocessed point cloud data into a parameterized three-dimensional digital model, realizes the conversion from discrete data to a continuous model, and provides a digital foundation for subsequent finite element analysis and structural optimization.

[0070] Step S3 specifically includes:

[0071] Step S31: Use an intelligent feature recognition method based on machine learning to perform geometric cleaning and simplification on the three-dimensional digital model, detect and extract key geometric features of the sheet metal parts, retain the main geometric features and topological relationships of the sheet metal parts, remove detail features that have little impact on structural performance, and for symmetrical or periodic structures, identify and extract the minimum repeating unit to generate a simplified partial model to reduce the calculation scale;

[0072] Step S32, meshing the simplified three-dimensional digital model to generate a global scale finite element mesh model, and defining material property parameters of the sheet metal part;

[0073] Step S33, identifying stress concentration areas in the bending area, punching area and welding area of ​​the sheet metal part, performing local geometric extraction on the stress concentration area, generating an independent local sub-model, defining a local coordinate system, and establishing a geometric coupling relationship between the local sub-model and the global scale finite element mesh model;

[0074] Step S34, using a high-order unit type and a refined grid size to encrypt the grid of the local sub-model;

[0075] Step S35: perform error estimation, evaluate the quality and convergence of the current finite element mesh, and use an adaptive meshing strategy driven by an evolutionary algorithm based on key geometric feature information extracted by the intelligent feature recognition method to adjust the mesh density and unit size for areas with large errors, and repeat the error estimation and adaptive mesh encryption process until a preset convergence accuracy threshold is reached;

[0076] Step S36, embed the local sub-model into the global-scale finite element mesh model, establish a position constraint relationship based on the coupling of node degrees of freedom, apply a displacement coordination equation on the interface between the local sub-model and the global-scale finite element mesh model to ensure the continuity of displacement and rotation, and define a load transfer mechanism between the local sub-model and the global-scale finite element mesh model to achieve two-way interaction of multi-scale information, thereby forming a multi-scale model of the sheet metal.

[0077] In steps S31-S36, this embodiment introduces machine learning and evolutionary algorithms in key links to achieve intelligent processing of the entire process from geometric processing to mesh generation. It can adaptively identify key geometric features at different scales and adaptively encrypt local meshes based on feature recognition results.

[0078] Wherein, step S31 specifically includes:

[0079] Step S311, collecting a certain number of three-dimensional digital models of sheet metal parts of different types as original data sets; manually annotating the features of each three-dimensional digital model of the sheet metal part, identifying key features such as bending areas, punching areas and welding areas, and forming a labeled data set; converting the annotated three-dimensional digital model of the sheet metal part into a mesh model and performing mesh cleaning and repair to eliminate noise and topological errors; normalizing the mesh model, scaling it to a uniform size range, and converting it into a voxel representation to generate a 64×64×64 three-dimensional voxel mesh;

[0080] Step S312, construct a three-dimensional CNN network model for extracting multi-scale geometric features from a three-dimensional voxel grid, the three-dimensional CNN network model includes 2 convolution layers, 2 pooling layers, 2 fully connected layers, a multi-scale feature extraction fusion module and an attention module; the input data is a 64×64×64 three-dimensional voxel grid, the first convolution layer performs a 3×3×3 convolution operation on the input data, and outputs 32 feature maps; the first pooling layer performs a 2×2×2 maximum pooling operation on the output of the first convolution layer to halve the size of the feature map; the multi-scale feature extraction fusion module first performs three branches of convolution and pooling operations on the output of the first pooling layer in parallel, the first branch performs a 3×3×3 convolution operation and a 2×2×2 maximum pooling operation, the second branch performs a 3×3×3 convolution operation and a 4×4×4 maximum pooling operation, and the third branch performs a 3×3×3 convolution operation and an 8×8×8 maximum pooling operation. The first convolutional layer performs a 3×3×3 convolution operation on the multi-scale features and outputs 128 feature maps. The attention module performs feature recalibration and weight adjustment on the output of the second convolutional layer to generate an attention weight map with the same size as the input feature map, and then multiplies the attention weight map with the output of the second convolutional layer to obtain the feature map after attention enhancement. The second pooling layer performs a 2×2×2 maximum pooling operation on the feature map after attention enhancement to halve the size of the feature map. The first fully connected layer flattens the output of the second pooling layer to obtain a high-dimensional feature vector, and maps the feature vector to a 1024-dimensional hidden representation through a fully connected operation. The second fully connected layer maps the output of the first fully connected layer to a K-dimensional output vector through a fully connected operation, where K is the number of feature categories.

[0081] Step S313, input the three-dimensional voxel grid annotated in step S311 into the three-dimensional CNN network model, train the model, and optimize the network parameters to minimize the feature prediction error; use the trained three-dimensional CNN network model to extract features from the three-dimensional digital model of the sheet metal part to obtain a feature vector representation of each voxel; convert the extracted voxel feature vector into a graph representation, with each voxel as a node of the graph, and establishing edge connections between adjacent voxels;

[0082] Step S314, construct a GNN network model for feature recognition and segmentation in the graph domain, the GNN network model includes 2 graph convolution layers, 2 graph attention layers, 2 graph pooling layers, 1 global pooling layer and 1 fully connected layer; the input data is the node feature matrix converted from the output of the three-dimensional CNN network model, the first graph convolution layer performs a graph convolution operation on the input data, thereby updating the feature representation of each node, and the output feature dimension is 256; the first graph attention layer performs feature aggregation and update on the output of the first graph convolution layer, and the output feature dimension is 256; the first graph pooling layer performs a pooling operation on the output of the first graph attention layer, thereby reducing the resolution and computational complexity of the graph, and the pooling The ratio is 0.5; the second graph convolution layer performs a graph convolution operation on the output of the first graph pooling layer, and the output feature dimension is 128; the second graph attention layer performs feature aggregation and update on the output of the second graph convolution layer, and the output feature dimension is 128; the second graph pooling layer performs a pooling operation on the output of the second graph attention layer to further reduce the resolution and computational complexity of the graph, and the pooling ratio is 0.5; the global pooling layer performs a global pooling operation on the output of the second graph pooling layer, aggregates the graph-level features into a vector representation, and obtains a 128-dimensional global feature vector; the fully connected layer maps the output of the global pooling layer to the number of feature categories K through a fully connected operation, and obtains the probability distribution of each category;

[0083] Step S315, input the graph constructed in step S313 into the GNN network model, train the model, and optimize the network parameters to minimize the feature classification error; use the trained GNN network model to perform feature recognition and segmentation on the three-dimensional digital model of the sheet metal part to obtain the feature category of each voxel;

[0084] Step S316: According to the feature segmentation results of the GNN network model, the three-dimensional digital model is geometrically cleaned and simplified; for ordinary areas, the edge collapse algorithm is used to merge adjacent faces to simplify the mesh complexity; for key feature areas, the original geometric topological structure is retained without simplification; for symmetrical or periodic structures, the minimum repeating unit is identified and extracted, only the geometric information of the minimum repeating unit is retained, and periodic constraints are imposed on the unit boundaries to generate simplified partial models to reduce the scale of calculation.

[0085] In steps S311-S316, this embodiment uses deep learning technology to construct an intelligent feature recognition method based on three-dimensional CNN and GNN, and automatically extracts and identifies key geometric features of sheet metal parts through it, as well as guides the model simplification process. Considering that electrical equipment cabinets and electronic chassis usually have the characteristics of large size, multiple functional areas and components, relatively regular geometric structure, a large number of symmetric and repetitive structures, and relatively concentrated distribution of key features such as bending areas, punching areas, and welding areas, this embodiment introduces a multi-scale feature fusion mechanism in the three-dimensional CNN network to better capture geometric features at different scales, and introduces an attention mechanism to better focus on the key feature areas of sheet metal parts. The GNN network and the three-dimensional CNN network form a collaborative feature extraction and recognition framework.

[0086] Wherein, step S35 specifically includes:

[0087] Step S351, extract the stress component on each unit of the current finite element mesh, use the Zienkiewicz-Zhu error estimation method, use the stress values ​​at the unit nodes to restore the smooth stress field, calculate the energy norm error between the restored stress field and the finite element solution on each unit, normalize the energy norm errors of all units, and obtain the relative error distribution;

[0088] Step S352: define the optimization goal of mesh division as minimizing the overall error and the number of network units, design a fitness function, use a genetic algorithm, randomly generate an initial mesh division scheme that meets the constraints as an initial population, repeat the selection operation, crossover operation and mutation operation, and continuously update the population until the maximum number of iterations is reached or the fitness converges, output the optimal mesh division scheme, and use it as the basis for mesh encryption;

[0089] Step S353, according to the Zienkiewicz-Zhu error estimation result, determine the cells with relative errors greater than 10% as candidate areas for mesh encryption, encrypt the cells and reduce the cell size to 5 mm, and further reduce the cell size to 2 mm in the key feature area to improve local accuracy;

[0090] Step S354: re-perform finite element analysis on the mesh after encryption and adjustment to obtain an updated stress field, and repeat steps S351-S353 until the overall relative error is less than 5% or the preset maximum number of iterations is reached.

[0091] In steps S351-S354, error estimation, mesh quality and computational efficiency are comprehensively considered, and rapid optimization of the finite element mesh is achieved through evolutionary algorithms and intelligent strategies, reducing the workload of manual adjustment and repeated trial calculations and improving the efficiency of finite element analysis.

[0092] Step S4 specifically includes:

[0093] Step S41, analyzing the static load, dynamic load and impact load encountered by the sheet metal in actual use, and determining the size, direction and action position of each load condition; applying the simplified and equivalent load condition to the corresponding position of the multi-scale model, defining boundary conditions such as displacement constraints and symmetry conditions, and simulating the connection and support mode between the sheet metal and the surrounding structure;

[0094] Step S42: Select an elastic-plastic model or an anisotropic model according to the material type of the sheet metal part, and input the elastic modulus, Poisson's ratio, yield strength and hardening curve of the material;

[0095] Step S43, selecting an implicit solver or an explicit solver to match different load conditions and material nonlinear characteristics, setting the time step, iterative convergence criterion and stress failure criterion, performing nonlinear finite element analysis on the multi-scale model, and obtaining the displacement field, stress field and strain field of the sheet metal under different conditions;

[0096] Step S44: Evaluate the strength, stiffness and stability of the sheet metal parts under different working conditions, and analyze the fracture failure mode and fatigue failure position of the sheet metal parts under different working conditions.

[0097] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A sheet metal structural performance analysis method based on multi-scale modeling, characterized in that: The method comprises the following steps: Step S1, using three-dimensional scanning technology to perform high-precision scanning on the physical sheet metal part, obtain point cloud data of the sheet metal part, and pre-process the point cloud data; Step S2, importing the pre-processed point cloud data into reverse engineering software for surface reconstruction, generating a three-dimensional digital model of the sheet metal part, and performing feature recognition and parameterization processing on the three-dimensional digital model to extract key dimensional parameters; Step S3, meshing the parameterized three-dimensional digital model, generating finite element mesh models of different densities using a multi-scale modeling method, and performing mesh encryption in key areas to obtain a multi-scale model of the sheet metal part; Step S4: define loads and boundary conditions for the multi-scale model, select material property parameters, and use the finite element analysis method to perform structural performance simulation analysis on the multi-scale model under multiple working conditions.

2. The sheet metal structural performance analysis method based on multi-scale modeling according to claim 1, characterized in that: The step S1 specifically includes: Step S11, selecting a suitable 3D scanning device, and determining scanning parameters according to the size, shape complexity and required accuracy of the sheet metal part; Step S12, performing surface treatment on the solid sheet metal part and fixing the solid sheet metal part in an appropriate position; Step S13, scanning the physical sheet metal part at multiple angles and in multiple directions by means of a three-dimensional scanning device to obtain point cloud data of the physical sheet metal part; Step S14: preprocessing the acquired point cloud data, including point cloud filtering, point cloud registration, point cloud downsampling and point cloud smoothing.

3. The sheet metal structural performance analysis method based on multi-scale modeling according to claim 1, characterized in that: The step S2 specifically includes: Step S21, importing the pre-processed point cloud data into reverse engineering software; Step S22, performing surface reconstruction on the imported point cloud data to generate a three-dimensional digital model of the sheet metal part; Step S23, performing post-processing operations of model repair, model simplification and model smoothing on the reconstructed three-dimensional digital model; Step S24, performing feature recognition on the post-processed three-dimensional digital model, extracting key geometric features such as plane, round hole, bend, and stamping of the sheet metal part, classifying, annotating, and organizing the recognized features, and establishing a complete feature tree; Step S25, perform parameterization processing on the identified features, extract key dimensional parameters such as plane size, circular hole diameter, bending radius, punching depth, etc., convert the geometric information of the features into parameterized dimensional constraints and relationships, and establish a parameterized three-dimensional digital model.

4. The sheet metal structural performance analysis method based on multi-scale modeling according to claim 1, characterized in that: The step S3 specifically includes: Step S31, using an intelligent feature recognition method based on machine learning to geometrically clean and simplify the three-dimensional digital model; Step S32, meshing the simplified three-dimensional digital model to generate a global scale finite element mesh model, and defining material property parameters of the sheet metal part; Step S33, identifying stress concentration areas in the bending area, punching area and welding area of ​​the sheet metal part, performing local geometric extraction on the stress concentration area, generating an independent local sub-model, defining a local coordinate system, and establishing a geometric coupling relationship between the local sub-model and the global scale finite element mesh model; Step S34, using a high-order unit type and a refined grid size to encrypt the grid of the local sub-model; Step S35: perform error estimation, evaluate the quality and convergence of the current finite element mesh, and use an adaptive meshing strategy driven by an evolutionary algorithm based on key geometric feature information extracted by the intelligent feature recognition method to adjust the mesh density and unit size for areas with large errors, and repeat the error estimation and adaptive mesh encryption process until a preset convergence accuracy threshold is reached; Step S36: embed the local sub-model into the global scale finite element mesh model to form a multi-scale model of the sheet metal part.

5. The sheet metal structural performance analysis method based on multi-scale modeling according to claim 4 is characterized in that: The step S31 specifically includes: Step S311, collecting a certain number of three-dimensional digital models of sheet metal parts of different types as original data sets; manually annotating the features of each three-dimensional digital model of the sheet metal part, identifying key features such as bending areas, punching areas and welding areas, and forming a data set with labels; converting the annotated three-dimensional digital model of the sheet metal part into a mesh model and performing mesh cleaning and repair; normalizing the mesh model, converting it into a voxel representation, and generating a three-dimensional voxel mesh; Step S312: constructing a three-dimensional CNN network model for extracting multi-scale geometric features from a three-dimensional voxel grid, wherein the three-dimensional CNN network model includes two convolutional layers, two pooling layers, two fully connected layers, a multi-scale feature extraction fusion module, and an attention module; Step S313, input the three-dimensional voxel grid annotated in step S311 into the three-dimensional CNN network model, train the model, and optimize the network parameters to minimize the feature prediction error; use the trained three-dimensional CNN network model to extract features from the three-dimensional digital model of the sheet metal part to obtain a feature vector representation of each voxel; convert the extracted voxel feature vector into a graph representation, with each voxel as a node of the graph, and establishing edge connections between adjacent voxels; Step S314: construct a GNN network model for feature recognition and segmentation in the graph domain, wherein the GNN network model includes 2 graph convolution layers, 2 graph attention layers, 2 graph pooling layers, 1 global pooling layer and 1 fully connected layer; Step S315, input the graph constructed in step S313 into the GNN network model, train the model, and optimize the network parameters to minimize the feature classification error; use the trained GNN network model to perform feature recognition and segmentation on the three-dimensional digital model of the sheet metal part to obtain the feature category of each voxel; Step S316: According to the feature segmentation results of the GNN network model, the three-dimensional digital model is geometrically cleaned and simplified; for ordinary areas, the edge collapse algorithm is used to merge adjacent faces to simplify the mesh complexity; for key feature areas, the original geometric topological structure is retained without simplification; for symmetrical or periodic structures, the minimum repeating unit is identified and extracted, only the geometric information of the minimum repeating unit is retained, and periodic constraints are imposed on the unit boundaries to generate simplified partial models to reduce the scale of calculation.

6. The sheet metal structural performance analysis method based on multi-scale modeling according to claim 5, characterized in that: In the three-dimensional CNN network model, the first convolution layer is used to perform a convolution operation on the input data; the first pooling layer is used to perform a maximum pooling operation on the output of the first convolution layer to halve the size of the feature map; the multi-scale feature extraction fusion module is used to first perform convolution and pooling operations of three branches on the output of the first pooling layer in parallel, then upsample the outputs of the three branches, and finally splice the upsampled results of the three branches in the feature dimension to obtain a fused multi-scale feature representation; The second convolutional layer is used to perform convolution operations on multi-scale features; the attention module is used to recalibrate and adjust the weights of the output of the second convolutional layer to generate an attention weight map with the same size as the input feature map, and then multiply the attention weight map with the output of the second convolutional layer to obtain the feature map after attention enhancement; The second pooling layer is used to perform the maximum pooling operation on the feature map after attention enhancement to halve the size of the feature map; the first fully connected layer is used to flatten the output of the second pooling layer to obtain a high-dimensional feature vector, and map the feature vector to a high-dimensional hidden representation through a fully connected operation; The second fully connected layer is used to map the output of the first fully connected layer to a K-dimensional output vector through a fully connected operation, where K is the number of feature categories.

7. The sheet metal structural performance analysis method based on multi-scale modeling according to claim 5, characterized in that: In the GNN network model, the first graph convolution layer is used to perform graph convolution operations on input data; the first graph attention layer is used to perform feature aggregation and update on the output of the first graph convolution layer; the first graph pooling layer is used to perform pooling operations on the output of the first graph attention layer; the second graph convolution layer is used to perform graph convolution operations on the output of the first graph pooling layer; the second graph attention layer is used to perform feature aggregation and update on the output of the second graph convolution layer; the second graph pooling layer is used to perform pooling operations on the output of the second graph attention layer; the global pooling layer is used to perform global pooling operations on the output of the second graph pooling layer, aggregate graph-level features into a vector representation, and obtain a global feature vector; the fully connected layer is used to map the output of the global pooling layer to the number of feature categories K through a fully connected operation, and obtain the probability distribution of each category.

8. The sheet metal structural performance analysis method based on multi-scale modeling according to claim 4 is characterized in that: The step S36 specifically includes: A position constraint relationship based on the coupling of node degrees of freedom is established, and the displacement coordination equation is applied on the interface between the local sub-model and the global-scale finite element mesh model to ensure the continuity of displacement and rotation. The load transfer mechanism between the local sub-model and the global-scale finite element mesh model is defined to achieve two-way interaction of multi-scale information, thereby forming a multi-scale model of sheet metal.

9. The sheet metal structural performance analysis method based on multi-scale modeling according to claim 1, characterized in that: The step S4 specifically includes: Step S41, analyzing the static load, dynamic load and impact load encountered by the sheet metal in actual use, and determining the size, direction and action position of each load condition; applying the simplified and equivalent load condition to the corresponding position of the multi-scale model, defining boundary conditions such as displacement constraints and symmetry conditions, and simulating the connection and support mode between the sheet metal and the surrounding structure; Step S42: Select an elastic-plastic model or an anisotropic model according to the material type of the sheet metal part, and input the elastic modulus, Poisson's ratio, yield strength and hardening curve of the material; Step S43, selecting an implicit solver or an explicit solver to match different load conditions and material nonlinear characteristics, setting the time step, iterative convergence criterion and stress failure criterion, performing nonlinear finite element analysis on the multi-scale model, and obtaining the displacement field, stress field and strain field of the sheet metal under different conditions; Step S44: Evaluate the strength, stiffness and stability of the sheet metal parts under different working conditions, and analyze the fracture failure mode and fatigue failure position of the sheet metal parts under different working conditions.