A vehicle body topology optimization method and system based on generative adversarial network (GAN)

Abstract

The present invention relates to a vehicle body topology optimization method and system based on a generative adversarial network (GAN), belonging to the field of deep learning applications in vehicles and computers. The method comprises obtaining historical vehicle body topology optimization data, removing redundant features, and performing data enhancement to obtain a historical vehicle body topology optimization dataset. The historical vehicle body topology optimization data includes vehicle load data, vehicle body structure boundary conditions, and vehicle body structure geometry data. A generative adversarial network model is constructed based on the historical vehicle body topology optimization dataset using adversarial binary mapping. The generative adversarial network model includes a generator, a discriminator, and physical constraints. The generative adversarial network model is progressively trained, with gradient penalties introduced during the progressive training, to obtain a vehicle body topology optimization model. Based on the vehicle body topology optimization model, an optimized vehicle body topology structure is generated, and binarization processing and validity checks are performed. This invention achieves vehicle body topology optimization using a generative adversarial network, with high computational efficiency, high accuracy, and strong practicality.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle and computer deep learning application technology, and specifically relates to a vehicle body topology optimization method and system based on a generative adversarial network (GAN). Background Art

[0002] Currently, in order to improve fuel efficiency or battery life, the automotive industry is moving towards lightweight, energy-saving and environmentally friendly development. The optimized design of the vehicle body structure has become crucial. Topology optimization is a type of structural optimization, which generally refers to achieving optimal performance by optimizing material distribution within a given design space. Topology optimization is also an effective structural design solution for automobile body design problems. It can achieve the goal of lightweighting the vehicle body by finding the optimal material distribution while meeting the structural strength and stiffness requirements. However, existing topology optimization methods, such as the finite element method, are usually computationally intensive and time-consuming, and the optimization results are prone to falling into local optimality.

[0003] Generative adversarial networks (GANs) are a deep learning method that consists of a generator and a discriminator that interact to efficiently generate optimization results that are close to real data. Therefore, using GANs for topological optimization of vehicle body structures can significantly improve computational efficiency and achieve a more optimal overall structure. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a vehicle body topology optimization method and system based on generative adversarial network (GAN).

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] A vehicle body topology optimization method based on a generative adversarial network (GAN), wherein the implementation of the vehicle body topology optimization method comprises the following steps:

[0007] Acquire historical vehicle body topology optimization data, remove redundant features, and perform data enhancement to obtain a historical vehicle body topology optimization dataset, wherein the historical vehicle body topology optimization data includes vehicle load data, vehicle body structure boundary conditions, and vehicle body structure geometry data;

[0008] Constructing a generative adversarial network model based on the historical vehicle body topology optimization dataset through adversarial binary mapping, wherein the generative adversarial network model includes a generator, a discriminator, and physical constraints;

[0009] Progressively training the generative adversarial network model to obtain a vehicle body topology optimization model, wherein the progressive training introduces a gradient penalty;

[0010] A vehicle body topology optimization structure is generated based on the vehicle body topology optimization model, and binarization processing and legality checking are performed.

[0011] Preferably, the data enhancement is specifically: using a rotation matrix to perform data enhancement on the historical vehicle body topology optimization data, and the expression of the rotation matrix is: Where θ is the rotation angle, θ∈[-20°,20°].

[0012] Preferably, the generator specifically includes:

[0013] The historical vehicle body topology optimization dataset is input into the generator network to obtain the vehicle body topology optimization input vector. A deep convolutional generative network is used to extract feature maps through an alternating structure of convolutional layers and deconvolutional layers and output a binary topological layout.

[0014] Preferably, the output of the binary topological layout includes:

[0015] The vehicle body topology optimization input vector is convolved to extract the load-sensitive features, which is mathematically described as C pq (r) =f(∑m∑n(W mn r ×X(p+m)(q+n) (r-1) +v r ), where C pg (r) is the load-sensitive eigenvalue at position (p, q) in the feature map after the rth convolution, X(p+m)(q+n) (r-1) is the load-sensitive eigenvalue in the local convolution region of the feature map of the previous layer, W mn r is the generator convolution kernel weight, f is the activation function, v r is the generator bias term;

[0016] Introducing the attention mechanism, dynamically weighted key areas, mathematically described as α pq =Softmax(W a ·C pq +b a ), where α pq is the dynamic weighted value, W a is the weight matrix, b a is the attention mechanism bias term;

[0017] The deconvolution operation is used to upsample and reconstruct the feature map, which is mathematically described as X (l+1) =f(WT l ·C l +b l ), where WT l is the deconvolution weight, Cl is the feature map output by the convolutional layer, b l is the deconvolution bias term;

[0018] The convolution and deconvolution operations are repeated to output the binary topological layout.

[0019] Preferably, the discriminator includes:

[0020] The binary topological layout and the real structural layout are input into the discriminator respectively, and features are extracted through multi-layer convolution and pooling operations, which is mathematically described as Y (l+1) =f(WR l ·Y l +b′ l ), where Y (l+1) is the feature map of the discriminator layer l+1, Y l is the feature map of the discriminator layer l, WR l is the discriminator convolution kernel weight, b′ l is the discriminator bias term, f is the activation function, and the discriminator loss function of the discriminator is Γ D = -E[logD(x|B)]-E[log(1-D(z|B))], where E is the expectation, D(x|B) is the probability that the discriminator considers the true structural layout x to be true, and D(z|B) is the probability that the discriminator considers the binary topological layout z to be true;

[0021] Output discriminator probability.

[0022] Preferably, the physical constraints specifically include performance loss constraints and volume constraints, and the performance loss constraints include stress constraints and deformation constraints.

[0023] Preferably, the mathematical description of the stress constraint term is Λ physics1 =λ1‖Stress(G output )-σ max ‖ 2 , where Stress is the actual stress, G output is the binary topological layout, σ max is the maximum stress allowed by the material, λ1 is the stress weight coefficient; the mathematical description of the deformation constraint term is Λ physics2 =λ2||Displacement(G output )|| 2 , where Displacement is the key point deformation variable and λ2 is the deformation weight coefficient. The mathematical description of the performance loss constraint is Λ physics =λ1‖Stress(G output )-σ max ‖ 2+λ2‖Displacement(G output )‖ 2 ;

[0024] The mathematical description of the volume constraint is Among them, λ3 is the volume weight coefficient, V target is the target material volume ratio, H is the number of grids in the height direction of the generated body structure diagram, and W is the number of grids in the width direction of the generated body structure diagram.

[0025] Preferably, the progressive training of the generative adversarial network model includes:

[0026] Using progressive training, starting from a low-resolution binary topology layout, and gradually increasing the resolution to the preset target size;

[0027] Introducing WGAN-GP, the gradient penalty of WGAN-GP is Among them, λ GP is the penalty coefficient, is the gradient of the discriminator with respect to the input sample.

[0028] Preferably, the binarization processing is to perform threshold segmentation on the vehicle body topology optimization structure to extract continuous material areas; and the legality check includes connectivity analysis and minimum size constraint.

[0029] A vehicle body topology optimization system based on a generative adversarial network (GAN), for executing the above-mentioned vehicle body topology optimization method, comprising a data acquisition module, a model building module, a model training module, and a structure inspection module;

[0030] The data acquisition module is used to acquire historical vehicle body topology optimization data, remove redundant features and perform data enhancement to obtain a historical vehicle body topology optimization data set, wherein the historical vehicle body topology optimization data includes vehicle load data, vehicle body structure boundary conditions and vehicle body structure geometry data;

[0031] The model building module is used to build a generative adversarial network model based on the historical vehicle body topology optimization dataset through adversarial binary mapping, and the generative adversarial network model includes a generator, a discriminator and physical constraints;

[0032] The model training module is used to progressively train the generative adversarial network model to obtain a vehicle body topology optimization model, wherein the progressive training introduces gradient penalty;

[0033] The structure checking module is used to generate a vehicle body topology optimization structure based on the vehicle body topology optimization model, and implement binarization processing and legality checking.

[0034] The beneficial effects of the present invention are:

[0035] (1) Synergistic improvement of computational efficiency and accuracy: Through a progressive training strategy, the model is generated from low resolution to high resolution, shortening the training time and generating clearer structural details. Through embedded physical constraints, stress, deformation, and volume constraints are directly coupled in the GAN, reducing the number of a posteriori finite element verifications.

[0036] (2) Strong engineering practicality: By integrating physical constraints with data-driven methods, the maximum stress of the generated structure is ensured not to exceed the material limit. The volume constraint directly controls the material proportion through the probability threshold, avoiding the tedious process of manually adjusting the penalty factor in the traditional method. Through automatic legality checking, isolated areas are eliminated to ensure that the structure can be directly used for manufacturing, thereby improving practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] To facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0038] Figure 1 This is a flowchart of the steps of a vehicle body topology optimization method based on a generative adversarial network (GAN) of the present invention. DETAILED DESCRIPTION

[0039] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0040] The working principle and use process of the present invention:

[0041] See also Figure 1 , a car body topology optimization method based on generative adversarial network (GAN), including:

[0042] S1: Obtain historical vehicle body topology optimization data, remove redundant features and perform data enhancement to obtain a historical vehicle body topology optimization dataset. The historical vehicle body topology optimization data includes vehicle load data, vehicle body structure boundary conditions, and vehicle body structure geometry data. The historical vehicle body topology optimization dataset is expressed as D aug ={(L std ,B norm ,G std )},i=1,…,N, where D aug For the historical vehicle body topology optimization dataset, L std is the vehicle load dataset after data standardization, B norm is the normalized vehicle body structure boundary condition dataset, G std The vehicle body structure geometry dataset after data standardization;

[0043] S2: constructing a GAN model based on the historical vehicle body topology optimization dataset through adversarial binary mapping, wherein the GAN model includes a generator, a discriminator, and physical constraints;

[0044] S3: progressively training the GAN model to obtain a vehicle body topology optimization model, wherein the progressive training introduces a gradient penalty;

[0045] S4: Generate a vehicle body topology optimization structure based on the vehicle body topology optimization model, and perform binarization processing and legality check.

[0046] In this embodiment, redundant feature removal and data enhancement are performed to obtain a historical vehicle body topology optimization dataset, which can be specifically implemented by the following steps:

[0047] The historical vehicle body topology optimization data is data of the vehicle under different working conditions, wherein the vehicle load data includes structural response data or simulation data. The structural response data is data collected by sensors from the vehicle in working conditions such as vehicle collision tests, vehicle vibration tests, and vehicle static load tests. The simulation data is data obtained by simulation through computer simulation software such as ANSYS, ABAQUS, or LS-DYNA. The vehicle body structure boundary conditions are the support point positions, constraint types, and load application positions and directions of various parts of the vehicle body determined according to the actual conditions of the vehicle. The support point positions include suspension connection points and support positions, and the constraint types include fixed constraints and sliding constraints. The vehicle body structure geometry data is structural geometry information formed by importing and processing CAD files or finite element models of existing topology optimization structures, performing geometric cleaning, and removing redundant or repeated features.

[0048] The redundant feature removal includes data cleaning, data standardization and data normalization. The data cleaning includes identifying and removing data anomalies, and the data anomalies include historical body topology optimization data that obviously exceeds the normal range; the data standardization includes unifying the data scale of the vehicle load data and the body structure geometry data to the standard range through the standard deviation normalization method; the data normalization includes normalizing the boundary conditions of the body structure.

[0049] The data enhancement is specifically: using a rotation matrix to perform data enhancement on the historical vehicle body topology optimization data. The formula of the rotation matrix is: Here, θ is the rotation angle, θ∈[-20°, 20°]. Through data augmentation, the historical vehicle body topology optimization dataset is significantly expanded, which is conducive to improving the generalization ability of the GAN model.

[0050] In this embodiment, the generator is specifically:

[0051] The historical vehicle body topology optimization dataset is input into the generator network to obtain the vehicle body topology optimization input vector. The deep convolutional generator network is used to extract feature maps through an alternating structure of convolutional and deconvolutional layers and output a binary topology layout:

[0052] The vehicle body topology optimization input vector is convolved to extract the load-sensitive features, which is mathematically described as C pq (r)=f(∑m∑n(W mn r ×X(p+m)(q+n) (r-1) +v r ), where C pg (r) is the load-sensitive eigenvalue at position (p, q) in the feature map after the rth convolution, X(p+m)(q+n) (r-1) is the load-sensitive eigenvalue in the local convolution region of the feature map of the previous layer, W mn r is the generator convolution kernel weight, f is the activation function (usually ReLU), v r is the generator bias term;

[0053] Introducing the attention mechanism, dynamically weighted key areas, mathematically described as α pq =Softmax(W a ·C pq +b a ), where α pq is the dynamic weight value, α pq The larger the value, the more important the position is. a is the learning weight matrix, b a is the attention mechanism bias term.

[0054] The deconvolution operation is used to upsample and reconstruct the feature map, which is mathematically described as X (l+1) =f(WT l ·C l +b l ), where WT l is the deconvolution weight, C l is the feature map output by the convolutional layer, b l is the deconvolution bias term;

[0055] Repeat the convolution and deconvolution operations to output a binary topological layout.

[0056] In this embodiment, the discriminator is specifically:

[0057] The binary topological layout and the real structural layout are input into the discriminator respectively, and features are extracted through multi-layer convolution and pooling operations, which is mathematically described as Y (l+1) =f(WR l·Y l +b' 1 ), where Y (l+1) is the feature map of the discriminator layer l+1, Y l is the feature map of the discriminator layer l, WR l is the discriminator convolution kernel weight, b' 1 is the discriminator bias term, f is the activation function (usually LeakyReLU);

[0058] The output discriminator probability indicates the possibility that the input binary topological layout is the true structural layout. The discriminator loss function of the discriminator is Γ D = -E[logD(x|B)]-E[LOG(1-D(z|B))], where E is the expectation, D(x|B) is the probability that the discriminator believes that the true structural layout x is true, and D(z|B) is the probability that the discriminator believes that the binary topological layout z is true. The first term encourages the discriminator to recognize the true structural layout, and the second term encourages the discriminator to see through the binary topological layout.

[0059] In this embodiment, the physical constraints are specifically:

[0060] Embedding the physical constraints, wherein the physical constraints include a performance loss constraint and a volume constraint, and the performance loss constraint includes a stress constraint term and a deformation constraint term;

[0061] The mathematical description of the stress constraint term is Λ physics1 =λ1||Stress(G output )-σ max || 2 , where Stress is the actual stress calculated by finite element simulation, G output is the binary topological layout, σ max is the maximum stress allowed by the material, λ1 is the stress weight coefficient, that is, if the actual stress of the generated structure exceeds the material's bearing limit, points will be deducted; the mathematical description of the deformation constraint term is Λ physics2 =λ2||Displacement(G output )|| 2 , where Displacement is the key point deformation variable and λ2 is the deformation weight coefficient. The mathematical description of the performance loss constraint is Λ physics =λ1||Stress(G output )-σ max || 2 +λ2||Displacement(G output )| | 2;

[0062] The mathematical description of the volume constraint is Among them, λ3 is the volume weight coefficient, V target is the target material volume ratio (for example, 50% of the material is required), H is the number of grids in the height direction of the generated body structure diagram, W is the number of grids in the width direction of the generated body structure diagram, ∑G output represents the sum of the probability values in each grid. Assuming that the average probability of each grid is 0.5 and the number of grids is 65536 (i.e. H·W), then ∑G output It is 65536×0.5=32768.

[0063] In this embodiment, the GAN model is progressively trained to obtain a vehicle body topology optimization model, which can be specifically implemented by the following steps:

[0064] S301: Use progressive training, starting from a low-resolution binary topology layout, and gradually increasing the resolution to the preset target size. For example, starting from 64×64 resolution, it is increased to 128×128 and 256×256 every 10,000 iterations.

[0065] S302: Introduce WGAN-GP, the gradient penalty of WGAN-GP is Among them, λ GP is the penalty coefficient, usually set to 10, is the gradient of the discriminator with respect to the input sample; the purpose of adding gradient penalty is to prevent the training process from being too harsh and causing training collapse.

[0066] In this embodiment, a vehicle body topology optimization structure is generated based on the vehicle body topology optimization model, and binarization processing and legality check are performed. This can be specifically implemented by the following steps:

[0067] The binarization process is to perform threshold segmentation on the vehicle body topology optimization structure and extract continuous material areas, that is, convert them into clear "material retention" and "material removal" areas; the legality check includes connectivity analysis and minimum size constraints. The connectivity analysis is to remove isolated material points and voids, and the minimum size constraint is to ensure that the minimum structural size meets manufacturing requirements.

[0068] A vehicle body topology optimization system based on a generative adversarial network (GAN), comprising a data acquisition module, a model building module, a model training module, and a structure inspection module;

[0069] The data acquisition module is used to acquire historical vehicle body topology optimization data, remove redundant features and perform data enhancement to obtain a historical vehicle body topology optimization data set. The historical vehicle body topology optimization data includes vehicle load data, vehicle body structure boundary conditions and vehicle body structure geometry data. The historical vehicle body topology optimization data set is expressed as D aug ={(L std,B norm ,G std )},i=1,…,N, where D aug For the historical vehicle body topology optimization dataset, L std is the vehicle load dataset after data standardization, B norm is the normalized vehicle body structure boundary condition dataset, G std The vehicle body structure geometry dataset after data standardization;

[0070] The model building module is used to build a GAN model through adversarial binary mapping according to the historical vehicle body topology optimization dataset, wherein the GAN model includes a generator, a discriminator and physical constraints;

[0071] The model training module is used to progressively train the GAN model to obtain a vehicle body topology optimization model, wherein the progressive training introduces a gradient penalty;

[0072] The structure checking module is used to generate a vehicle body topology optimization structure based on the vehicle body topology optimization model, and perform binarization processing and legality checking.

[0073] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device.

[0074] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0075] The program code included in the computer-readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF or the like, or any suitable combination thereof. The computer program code for performing the operation of the present invention can be written in one or more programming languages or a combination thereof, and the programming language includes an object-oriented programming language such as Java, Smalltalk, C++, and also includes a conventional procedural programming language such as "C" language or similar programming language. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, utilizing an Internet service provider to connect through the Internet).

[0076] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to equivalent embodiments using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A vehicle body topology optimization method based on generative adversarial networks, characterized in that: The implementation of the vehicle body topology optimization method includes the following steps: Acquire historical vehicle body topology optimization data, remove redundant features, and perform data enhancement to obtain a historical vehicle body topology optimization dataset, wherein the historical vehicle body topology optimization data includes vehicle load data, vehicle body structure boundary conditions, and vehicle body structure geometry data; Constructing a generative adversarial network model based on the historical vehicle body topology optimization dataset through adversarial binary mapping, wherein the generative adversarial network model includes a generator, a discriminator, and physical constraints; The generator specifically includes: Inputting the historical vehicle body topology optimization dataset into a generator network to obtain a vehicle body topology optimization input vector, using a deep convolutional generator network to extract feature maps through an alternating structure of convolutional layers and deconvolutional layers and output a binary topology layout; The output of the binary topology layout includes: The vehicle body topology optimization input vector is convolved to extract the load-sensitive features, which is mathematically described as C pq (r) =f(∑m∑n(W mn r ×X(p+m)(q+n) (r-1) +v r ), where C pq (r) is the load-sensitive eigenvalue at position (p, q) in the feature map after the rth convolution, X(p+m)(q+n) (r-1) is the load-sensitive eigenvalue in the local convolution region of the feature map of the previous layer, W mn r is the generator convolution kernel weight, f is the activation function, v r is the generator bias term; Introducing the attention mechanism, dynamically weighted key areas, mathematically described as α pq =Softmax(W a ·C pq +b a ), where α pq is the dynamic weighted value, W a is the weight matrix, b a is the attention mechanism bias term; The deconvolution operation is used to upsample and reconstruct the feature map, which is mathematically described as X (l+1) =f(WT l ·C l +b l ), where WT l is the deconvolution weight, C l is the feature map output by the convolutional layer, b l is the deconvolution bias term; Repeating convolution and deconvolution operations to output the binary topological layout; The discriminator comprises: The binary topological layout and the real structural layout are input into the discriminator respectively, and features are extracted through multi-layer convolution and pooling operations, which is mathematically described as Y (l+1) =f(WR l ·Y l +b′ l ), where Y (l+1) is the feature map of the discriminator layer l+1, Y l is the feature map of the discriminator layer l, WR l is the discriminator convolution kernel weight, b′ l is the discriminator bias term, f is the activation function, and the discriminator loss function of the discriminator is Γ D = -E{logD(x|B)]-E[log(1-D(z|B))], where E is the expectation, D(x|B) is the probability that the discriminator considers the true structural layout x to be true, and D(z|B) is the probability that the discriminator considers the binary topological layout z to be true; output the discriminator probability; The generative adversarial network model is progressively trained to obtain a vehicle body topology optimization model, and the progressive training introduces gradient penalty; A vehicle body topology optimization structure is generated based on the vehicle body topology optimization model, and binarization processing and legality checking are performed.

2. The vehicle body topology optimization method according to claim 1, characterized in that: The data enhancement is specifically: using a rotation matrix to enhance the historical vehicle body topology optimization data. The expression of the rotation matrix is: Where θ is the rotation angle, θ∈[-20°,20°].

3. The vehicle body topology optimization method according to claim 1, characterized in that: The physical constraints specifically include performance loss constraints and volume constraints, and the performance loss constraints include stress constraints and deformation constraints.

4. The vehicle body topology optimization method according to claim 3, characterized in that: The mathematical description of the stress constraint term is Λ physics1 =λ1||Stress(G output )-σ max || 2 , where Stress is the actual stress, G output is the binary topological layout, σ max is the maximum stress allowed by the material, λ1 is the stress weight coefficient; the mathematical description of the deformation constraint term is Λ physics2 =λ2||Displacement(G output )|| 2 , where Displacement is the key point deformation variable and λ2 is the deformation weight coefficient. The mathematical description of the performance loss constraint is Λ physics =λ1||Stress(G output )-σ max || 2 +λ2||Displacement(G output )|| 2 ; The mathematical description of the volume constraint is Among them, λ3 is the volume weight coefficient, V target is the target material volume ratio, H is the number of grids in the height direction of the generated body structure diagram, and W is the number of grids in the width direction of the generated body structure diagram.

5. The vehicle body topology optimization method according to claim 1, characterized in that: The progressive training of the generative adversarial network model includes: Using progressive training, starting from a low-resolution binary topology layout, and gradually increasing the resolution to the preset target size; The gradient penalty is Among them, λ GP is the penalty coefficient, is the gradient of the discriminator with respect to the input sample.

6. The vehicle body topology optimization method according to claim 1, characterized in that: The binarization process is to perform threshold segmentation on the vehicle body topology optimization structure to extract continuous material areas; the legality check includes connectivity analysis and minimum size constraint.

7. A vehicle body topology optimization system based on generative adversarial networks, characterized in that: The system is applied to the vehicle body topology optimization method according to any one of claims 1 to 6, comprising a data acquisition module, a model construction module, a model training module and a structure inspection module; The data acquisition module is used to acquire historical vehicle body topology optimization data, remove redundant features and perform data enhancement to obtain a historical vehicle body topology optimization data set, wherein the historical vehicle body topology optimization data includes vehicle load data, vehicle body structure boundary conditions and vehicle body structure geometry data; The model building module is used to build a generative adversarial network model based on the historical vehicle body topology optimization dataset through adversarial binary mapping, and the generative adversarial network model includes a generator, a discriminator and physical constraints; The model training module is used to progressively train the generative adversarial network model to obtain a vehicle body topology optimization model, wherein the progressive training introduces gradient penalty; The structure checking module is used to generate a vehicle body topology optimization structure based on the vehicle body topology optimization model, and implement binarization processing and legality checking.

Images