Topology Optimization-Based Hub Style Generation Method, Device and Server

Abstract

The present invention provides a method, apparatus and server for generating a wheel hub style based on topology optimization, relating to the technical field of artificial intelligence computer-aided design, including: obtaining an original data set; performing topology optimization processing on an initial wheel hub style based on a wheel hub reference style to reduce the flexibility of the initial wheel hub style and the difference between the initial wheel hub style and the wheel hub reference style, and determining a first optimized wheel hub style; screening the optimized wheel hub style through a preset L1 loss function to determine a second optimized wheel hub style, and obtaining a new style ratio corresponding to the second optimized wheel hub style; when the new style ratio is less than a preset ratio threshold, determining the second optimized wheel hub style as a target wheel hub style, and performing style evaluation processing on the target wheel hub style to generate a visual evaluation result. The present invention can significantly improve the diversity and engineering performance of the generated wheel hub styles.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence computer-aided design, and in particular to a method, device and server for generating hub styles based on topology optimization. Background Art

[0002] The generation of automobile hub styles is an important link in automobile styling design. For the design of hub styles, first of all, the engineering performance of the hub needs to be ensured. In addition, for users, the aesthetic design of the hub is also an essential factor. At present, the prior art proposes that the hub styles can be designed manually by CAD designers or new hub styles can be automatically generated by autoencoders. Among them, the diversity of the hub styles designed manually is relatively low, and the automatic design of the encoder can only focus on the engineering performance of the hub and cannot create aesthetic designs, thus unable to achieve the balance between the aesthetics and engineering performance of hub design. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method, device and server for generating hub styles based on topology optimization, which can significantly improve the diversity and engineering performance of the generated hub styles.

[0004] In a first aspect, an embodiment of the present invention provides a method for generating hub styles based on deep learning and topology optimization. The method includes: obtaining an original data set, where the original data set includes a preset number of hub reference styles based on existing vehicle hub styles; based on the hub reference styles, performing topology optimization processing on an initial hub style to reduce the compliance of the initial hub style and the difference between the initial hub style and the hub reference styles, and determining a first hub optimized style, where the compliance is used to represent the engineering performance, and the smaller the compliance, the better the engineering performance of the hub style; through a preset L1 loss function, performing screening processing on the hub optimized style to determine a second hub optimized style, and obtaining the new style ratio corresponding to the second hub optimized style, where the new style ratio is the ratio of the number of new hub styles in the current iteration to the total number of hub styles in the previous iteration; when the new style ratio is less than a preset ratio threshold, determining the second hub optimized style as the target hub style, and performing style evaluation processing on the target hub style to generate a visual evaluation result.

[0005] In one embodiment, based on the hub reference style, performing topology optimization on the initial hub style to reduce the flexibility of the initial hub style and the difference between the initial hub style and the hub reference style, and determining the first hub optimized style includes: performing data parsing on the hub reference style to determine the displacement vector, the overall stiffness matrix, and the density vector included in the hub reference style, and determining an optimization model based on the displacement vector, the overall stiffness matrix, and the density vector; using the optimization model to perform topology optimization on the initial hub style to determine the first hub optimized style, where the optimization model includes: a flexibility term and a difference term, and the optimization model is:

[0006]

[0007] Wherein, is the design variable, representing the density of the element ; is the density vector, is the displacement vector, T is the transpose of the matrix, is the overall stiffness matrix, is the flexibility term, is the difference term, is the hub reference style, is the L1 norm, is the similarity parameter, The larger it is, the closer the optimal design is to the hub reference style. For a smaller , the hub reference style is ignored and optimized to minimize compliance, is the volume fraction, is the material volume, is the design domain volume.

[0008] In one embodiment, the flexibility term in the optimization model is:

[0009]

[0010] Wherein, is the displacement vector, K is the overall stiffness matrix, is the flexibility, is the element displacement vector, is the element stiffness matrix, is the volume fraction, is the number of elements, is the density of the element, is the material volume, is the design domain volume.

[0011] In one embodiment, after the step of obtaining the new style ratio corresponding to the second optimized hub style, it includes: when the new style ratio is not less than the preset ratio threshold, using a preset generation model to perform auxiliary design processing on the second optimized hub style to determine the third optimized hub style.

[0012] In one embodiment, the generation model includes a preset generative adversarial network, and the preset generative adversarial network includes: a generator and a discriminator. Before the step of using the preset generation model to perform auxiliary design processing on the second optimized hub style to determine the third optimized hub style, it includes: obtaining a random noise vector and sending the random noise vector to the generator to provide randomness for hub style generation and determining random hub styles of different styles; sending the random hub styles to the discriminator so that the generator and the discriminator perform adversarial training to determine the target generative adversarial network, and using the target generative adversarial network to perform auxiliary design processing on the second optimized hub style to determine the third optimized hub style.

[0013] In one embodiment, after the step of using the preset generation model to perform auxiliary design processing on the second optimized hub style to determine the third optimized hub style, it includes: performing secondary screening processing on the third optimized hub style through a preset L1 loss function to determine the fourth optimized hub style, ending the topological optimization of this round, and re-performing topological optimization processing on the fourth optimized hub style.

[0014] In one embodiment, the step of performing style evaluation processing on the target hub style to generate a visual evaluation result includes: quantitatively evaluating the innovativeness of the target hub style through a preset autoencoder, determining the structural innovativeness evaluation result of the target hub style according to the difference between the target hub style and the hub reference style, and performing a physical performance test on the target hub style to determine the feasibility evaluation result of the target hub style; integrating the structural innovativeness evaluation result and the feasibility evaluation result to determine the visual evaluation result.

[0015] Second aspect, an embodiment of the present invention further provides a hub style generation device based on topology optimization. The device includes: a data acquisition module that acquires an original data set, where the original data set includes a preset number of hub reference styles based on existing vehicle hub styles; a topology optimization module that performs topology optimization processing on an initial hub style based on the hub reference styles to reduce the flexibility of the initial hub style and the difference between the initial hub style and the hub reference styles, and determines a first hub optimized style, where flexibility is used to represent engineering performance, and the smaller the flexibility, the better the engineering performance of the hub style; a screening and filtering module that performs screening processing on the hub optimized style through a preset L1 loss function to determine a second hub optimized style, and obtains the new style ratio corresponding to the second hub optimized style, where the new style ratio is the ratio of the number of new hub styles in the current iteration to the total number of hub styles in the previous iteration; a hub style evaluation module that, when the new style ratio is less than a preset ratio threshold, determines the second hub optimized style as the target hub style, and performs style evaluation processing on the target hub style to generate a visual evaluation result.

[0016] Third aspect, an embodiment of the present invention further provides a server, including a processor and a memory. The memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method according to any one of the first aspect.

[0017] Fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions that, when called and executed by a processor, cause the processor to implement the method according to any one of the first aspect.

[0018] The embodiments of the present invention bring the following beneficial effects:

[0019] A hub style generation method, device and server based on topology optimization provided by the embodiments of the present invention. After acquiring the original data set, the method performs topology optimization processing on the initial hub style based on the hub reference styles to reduce the flexibility of the initial hub style and the difference between the initial hub style and the hub reference styles, and determines a first hub optimized style. Then, through a preset L1 loss function, the hub optimized style is screened to filter out similar styles, and a second hub optimized style is determined. The new style ratio corresponding to the second hub optimized style is obtained. When the new style ratio is less than a preset ratio threshold, the second hub optimized style is determined as the target hub style, and style evaluation processing is performed on the target hub style to generate a visual evaluation result. The embodiments of the present invention ensure the engineering performance of the generated hubs through topology optimization processing, and at the same time generate diverse hub styles by combining deep learning, which can significantly improve the diversity and engineering performance of the generated hub styles.

[0020] Other features and advantages of the present invention will be described in the following specification, and in part will be obvious from the specification, or can be learned by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, claims, and drawings.

[0021] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specifically provides preferred embodiments and, in conjunction with the accompanying drawings, the detailed description is as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0023] Figure 1 FIG. [X] is a schematic structural diagram of a wheel design area provided by an embodiment of the present invention;

[0024] Figure 2 FIG. [X] is a schematic flowchart of a method for generating a wheel hub style based on topology optimization provided by an embodiment of the present invention;

[0025] Figure 3 FIG. [X] is a specific flowchart of a method for generating a wheel hub style based on topology optimization provided by an embodiment of the present invention;

[0026] Figure 4 FIG. [X] is a schematic structural diagram of a device for generating a wheel hub style based on topology optimization provided by an embodiment of the present invention;

[0027] Figure 5 FIG. [X] is a schematic structural diagram of a server provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0029] Currently, the generation of automotive wheel styles is an important part of automotive styling design. For the design of wheel styles, it is first necessary to ensure the engineering performance of the wheels. In addition, for users, the aesthetic design of the wheels is also an essential factor. The prior art proposes that the wheel styles can be manually designed by CAD designers or new wheel styles can be automatically generated by autoencoders. Among them, the diversity of the manually designed wheel styles is relatively low, and the automatic design of the encoder can only focus on the engineering performance of the wheels and cannot create aesthetic designs, thus failing to achieve the balance between the aesthetics and engineering performance of wheel design. In addition, the existing generation models are limited when directly generating engineering designs. Firstly, the generation models require a large amount of data, but the training data accumulated by the industry for various designs are confidential. Secondly, the generation models cannot guarantee the engineering feasibility. Finally, mode collapse is one of the main problems of generative models, and low-quality designs can be improved through topology optimization.

[0030] Based on this, the method, device and server for generating wheel styles based on topology optimization provided by the embodiments of the present invention first ensure the engineering performance of the generated wheels through topology optimization, and then generate diverse wheel styles from limited previous data designs through a generation model combined with deep learning. Compared with the previous generation design methods, this framework has better aesthetics, diversity and robustness in generation design, and can significantly improve the diversity and engineering performance of the generated wheel styles.

[0031] To facilitate the understanding of this embodiment, first, a method for generating wheel styles based on topology optimization disclosed in the embodiments of the present invention will be introduced in detail. Before performing topology training, it is first necessary to collect wheel reference styles, collect previous designs in the market and industry as the reference designs for subsequent topology optimization, and then generate new designs through topology optimization based on the wheel reference styles. The optimization objectives are to minimize compliance (representing engineering performance) and the difference from the wheel reference styles (aesthetics and diversity). Among them, topology optimization is usually used in structural design, in which the design area is divided into elements, and the optimal material density of the elements is determined under the consideration of given loads and boundary conditions to minimize compliance. The embodiments of the present invention provide a structural schematic diagram of the wheel design area, as Figure 1 shown, the design domain and boundary conditions of the two-dimensional wheel design. The original design domain is an element of 128 × 128, and the wheel reference style domain is also 128 × 128 pixels (that is to say, the format of the input wheel reference style is uniformly 128 × 128 pixels, and this pixel refers to the sum of D1, D2 and D3). The outer circle D1 of the wheel is set as the non-design area to maintain the shape of the rim, the inner area D3 is set as the fixed boundary condition to connect parts, and the spoke D2 is the main component of the design domain.

[0032] Based on Figure 1Schematic diagram of the structure of the wheel design area shown. In the embodiments of the present invention, a method for generating a wheel hub style based on topology optimization will be introduced in detail. Refer to Figure 2 Schematic flow chart of a method for generating a wheel hub style based on topology optimization shown. This method mainly includes the following steps S202 to step S208:

[0033] Step S202, obtain the original data set. Among them, the original data set includes: a preset number of wheel hub reference styles based on existing vehicle wheel hub styles.

[0034] Step S204, based on the wheel hub reference style, perform topology optimization on the initial wheel hub style to reduce the compliance of the initial wheel hub style and the difference between the initial wheel hub style and the wheel hub reference style, and determine the first optimized wheel hub style. Among them, compliance is used to represent engineering performance. The smaller the compliance, the better the engineering performance of the wheel hub style. In one implementation, data parsing processing can be performed on the wheel hub reference style to determine the displacement vector, overall stiffness matrix, and density vector included in the wheel hub reference style, and an optimization model is determined based on the displacement vector, overall stiffness matrix, and density vector, and the initial wheel hub style is subjected to topology optimization using the optimization model to determine the first optimized wheel hub style. Among them, the optimization model includes: a compliance term and a difference term. The optimization model is:

[0035]

[0036] Among them, is the design variable, representing the density of the element of, is the density vector, is the displacement vector, T is the transpose of the matrix, is the overall stiffness matrix, is the compliance term, is the difference term, is the wheel hub reference style, is the L1 norm, is the similarity parameter, The larger, the closer the optimal design is to the wheel hub reference style. For a smaller , the wheel hub reference style is ignored and optimized to minimize compliance, is the volume fraction, is the material volume, is the design domain volume.

[0037] Furthermore, the compliance term in the optimization model is:

[0038]

[0039] Among them, is the displacement vector, K is the global stiffness matrix, is the flexibility, is the element displacement vector, is the element stiffness matrix, is the volume fraction, is the number of elements, is the density of the element, is the material volume, is the volume of the design domain.

[0040] In addition, in the above expressions, the density directly related to the Young's modulus can be expressed as:

[0041]

[0042] where is the penalty factor to ensure a black-and-white design, and is introduced to avoid numerical instability when the element density is zero.

[0043] Step S206, through a preset L1 loss function, screen the optimized hub styles to filter out similar styles, determine the second optimized hub style, and obtain the new style ratio corresponding to the second optimized hub style. The new style ratio is the ratio of the number of new hub styles in the current iteration to the total number of hub styles in the previous iteration. In one implementation, after calculating the ratio of the number of new designs in the current iteration to the total number of designs in the previous iteration, if it is less than the threshold specified by the user, exit the iterative design exploration and conduct a design scheme evaluation; otherwise, use the preset generation model to perform auxiliary design processing on the second optimized hub style to further expand the hub design styles. In practical applications, the threshold can be set to 0.3, and this threshold can be adjusted according to the number of hub styles to be generated.

[0044] Step S208, when the new style ratio is less than the preset ratio threshold, determine the second optimized hub style as the target hub style, and conduct a style evaluation process on the target hub style to generate a visual evaluation result. In one implementation, quantitatively evaluate the innovativeness of the target hub style through a preset autoencoder, determine the structural innovativeness evaluation result of the target hub style according to the difference between the target hub style and the hub reference style, and conduct a physical performance test on the target hub style to determine the feasibility evaluation result of the target hub style. Finally, integrate the structural innovativeness evaluation result and the feasibility evaluation result to determine the visual evaluation result.

[0045] Specifically, it is possible to learn the designs in the dataset and train an autoencoder to construct a loss function that can help quantify the innovation degree of new designs, so as to compare with previous designs. And in the stage of evaluating design schemes, comprehensively evaluate the design schemes obtained through iterative design exploration, which not only includes the novelty evaluation of design schemes, but also covers actual physical characteristics such as the volume and compliance of the design. Finally, make trade-offs between different design attributes, draw the design schemes on each design attribute axis, and select the optimal design according to the relative importance of these attributes.

[0046] The above-mentioned hub style generation method based on topology optimization provided by the embodiments of the present invention ensures the engineering performance of the generated hubs through topology optimization processing. At the same time, combined with deep learning to generate diverse hub styles, it can significantly improve the diversity and engineering performance of the generated hub styles.

[0047] In one implementation manner, the embodiments of the present invention also provide a hub style optimization scheme when the new style ratio is not less than a preset ratio threshold: refer to the following (A) to (B):

[0048] (A) When the new style ratio is not less than the preset ratio threshold, use a preset generation model to perform auxiliary design processing on the second hub optimized style to determine the third hub optimized style. Among them, the generation model includes a preset generative adversarial network, and the preset generative adversarial network includes: a generator and a discriminator. In one implementation manner, a random noise vector can be obtained and sent to the generator to provide randomness for hub style generation, determine random hub styles of different styles, and then send the random hub styles to the discriminator, so that the generator and the discriminator perform adversarial training to determine the target generative adversarial network, and use the target generative adversarial network to perform auxiliary design processing on the second hub optimized style to determine the third hub optimized style.

[0049] In one implementation manner, a generation model can be used to generate new designs based on the current iterative design and use them as new hub reference styles. The generation model is mainly implemented through a generative adversarial network (GAN):

[0050]

[0051] Among them, is random noise, is random noise is the distribution of is the generator, with an input of z, and there is a weight parameter , is the discriminator, with an input of image , and its weight parameter is For a generative adversarial network (GAN), the training process involves two adversarial models, so its objective function includes a process of maximizing (max) and minimizing (min).

[0052] Specifically, first, a random noise vector is sampled from a normal distribution or a uniform distribution as the input of the generator. This noise vector provides randomness for generating hub images of different styles. Among them, the generator is a deep neural network model. After receiving the input noise it generates a simulated hub image through the learned weight parameters. The goal of the generator is to generate an image as close as possible to the real data distribution so that the discriminator cannot easily distinguish it.

[0053] The discriminator is responsible for receiving input images, including real hub images and images generated by the generator, and outputting a probability value to determine whether the image comes from the real data distribution. By maximizing the output probability of real images and minimizing the output probability of generated images at the same time, the discriminator gradually improves its ability to distinguish fake images.

[0054] After that, the generator and the discriminator are optimized through adversarial training. The generator tries to minimize the discrimination error of the discriminator for the generated images, while the discriminator tries to maximize the discrimination accuracy for real and generated images. As the training progresses, the hub images generated by the generator gradually approach the distribution of real images and can finally generate a variety of different styles and highly realistic hub images.

[0055] (B) By presetting the L1 loss function, the optimized style of the third hub is secondarily screened to filter out similar styles, determine the optimized style of the fourth hub, and end the topological optimization of this round. Then, the topological optimization process is carried out again for the optimized style of the fourth hub. In one implementation, the per-pixel L1 distance can be used as a criterion, and the boundary value is set to to filter out similar designs and reduce the computational cost.

[0056] See Figure 3 the specific process schematic diagram of a hub style generation method based on topological optimization shown. The embodiment of the present invention also provides a specific implementation manner for generating hub styles. Specifically, see the following (1) to (7):

[0057] (1) Collection of wheel hub reference styles (original data sets): First, 1,000 existing wheel hub designs were collected from the market and industry as benchmarks. These wheel hub reference styles were input into the system in the form of images and used to set initial conditions in subsequent topology optimization. This step provided a variety of design references for subsequent optimization and helped generate solutions that met engineering requirements and were aesthetically pleasing.

[0058] (2) Topology optimization: Based on the collected wheel reference styles, topology optimization is performed. The goal of topology optimization is not limited to improving the mechanical properties of the wheel hub, but also to ensure the diversity of design styles. Optimization is achieved by minimizing design flexibility (i.e., improving structural stiffness) and controlling the difference between the design and the reference image.

[0059] Specifically, the design area is divided into a 128 × 128 element grid, and different design areas and boundary conditions are set. The outer ring of the wheel is defined as a non-design area to ensure the stability of the rim shape, and the inner ring area is set as a fixed boundary for connecting components. During the optimization process, the spoke area of the wheel is the core of the design, and the adjustment of the element density determines the material distribution in this area.

[0060] During this stage, specific mathematical models (such as flexibility minimization and hub reference style difference control) are used to constrain the design and the material distribution of each unit is calculated through finite element analysis (FEA) to obtain a preliminary design that meets the structural performance requirements.

[0061] (3) Design screening, filtering similar designs: In order to avoid generating redundant or similar designs, the system will automatically screen after each optimization. In one embodiment, the pixel-by-pixel L1 distance can be used as the screening criterion, and the threshold is set to 10. -3 , filtering out solutions that are too similar to the previous design and retaining designs with large differences. This process significantly reduces the redundancy of subsequent calculations.

[0062] (4)New design ratio calculation to determine if the threshold is reached: The system evaluates the ratio of new designs generated in the current iteration by calculating the ratio of the new design solutions to the total number of solutions in the previous iteration. Since the design process usually involves a large number of parameter adjustments and multi-dimensional optimizations, especially when generating new designs, the system needs to balance innovation and exploration costs. Therefore, if the ratio of new designs generated during the iteration is too high, it may mean that the design space has not been fully covered and there are still many unexplored areas, and further design assistance is required through the generative model. However, if this ratio is too low, it may indicate that the design has stabilized, and further iteration may not yield significant results, leading to a waste of resources. Through extensive experiments and empirical analysis, the ratio threshold for new designs is determined to be 0.3. This threshold serves as a reasonable critical point to balance the requirements of system efficiency and design diversity. If the ratio of new designs is lower than the set threshold, it is considered that the design space has been fully explored, and further iteration is stopped and the final evaluation stage is entered. If the ratio is higher than the threshold, iteration continues to generate more solutions.

[0063] (5)Generative model-assisted design to generate new wheel hub styles: To further enhance the diversity and innovation of the design solutions, this step uses a generative adversarial network (GAN) to generate new designs. The GAN model generates new wheel hub design images from random noise. Among them, the generator G generates new design images based on the input random noise, while the discriminator D is responsible for distinguishing between the generated images and the real images. After multiple rounds of adversarial training, the generator gradually learns to generate designs similar to the real data distribution. Eventually, GAN can generate a variety of wheel hub design images with different styles and a high degree of realism.

[0064] (6)Secondary design screening to filter out similar designs: Among the new designs generated by the generative adversarial network, screening is carried out again to ensure design diversity. The L1 distance per pixel is still used to judge the differences, effectively ensuring that the generated design solutions are diverse while maintaining structural rationality.

[0065] (7)Design evaluation: At this stage, the innovation of the design is quantitatively evaluated through an autoencoder. By calculating the loss function and comparing the differences between the newly generated designs and the previous designs, the contribution of the design in terms of structural innovation is evaluated. In addition, the design solutions also need to undergo strict physical performance tests, including material volume, flexibility, stress resistance, etc., to ensure their feasibility in practical applications.

[0066] In summary, the above solution combines topology optimization and generative adversarial networks. Through multiple rounds of iterative design and intelligent screening, it effectively improves the design efficiency and reduces manual intervention. During the design evaluation process, the system conducts multi-dimensional analysis on each design attribute, comprehensively considering the physical performance and aesthetics of the design. By setting the attribute weights, the optimal design is selected. Specifically: during the design evaluation process, the system first assigns initial weights to each design attribute. These weights are set based on the historical data and design goals in the hub reference style collection stage. To improve flexibility, the system also introduces adjustable parameters to reflect the weight relationship between different attributes and dynamically adjusts them during the topology optimization and generative adversarial network iterations. For example, the system can adjust the weights according to the design effect of the wheel hub after topology optimization and test the impact of different attribute combinations on physical performance and design style. After the design generation and evaluation are completed, the system collects the performance of each design scheme through a feedback mechanism, especially the balance between physical performance and aesthetics, and evaluates the rationality of the current weight setting.

[0067] For the hub style generation method based on topology optimization provided in the foregoing embodiments, the embodiments of the present invention provide a hub style generation device based on topology optimization. Refer to Figure 4 the structural schematic diagram of a hub style generation device based on topology optimization shown in

[0068] A data acquisition module 402, which acquires an original data set. Among them, the original data set includes: a preset number of hub reference styles based on existing vehicle hub styles;

[0069] A topology optimization module 404, which performs topology optimization processing on the initial hub style based on the hub reference style to reduce the flexibility of the initial hub style and the difference between the initial hub style and the hub reference style, and determines the first hub optimized style. Among them, flexibility is used to represent engineering performance, and the smaller the flexibility, the better the engineering performance of the hub style;

[0070] A screening and filtering module 406, which performs screening processing on the hub optimized style through a preset L1 loss function to filter out similar styles, determines the second hub optimized style, and obtains the new style ratio corresponding to the second hub optimized style. Among them, the new style ratio is the ratio of the number of new hub styles in the current iteration to the total number of hub styles in the previous iteration;

[0071] A hub style evaluation module 408, when the new style ratio is less than a preset ratio threshold, determines the second hub optimized style as the target hub style, and performs style evaluation processing on the target hub style to generate a visual evaluation result.

[0072] The above-described hub style generation device based on topology optimization provided by the embodiments of the present application ensures the engineering performance of the generated hub through topology optimization processing, and at the same time combines deep learning to generate diverse hub styles, which can significantly improve the diversity and engineering performance of the generated hub styles.

[0073] In one implementation, when performing topology optimization processing on the initial hub style based on the hub reference style to reduce the flexibility of the initial hub style and the difference between the initial hub style and the hub reference style, and determine the first hub optimized style, the above topology optimization module 404 is further configured to: perform data parsing processing on the hub reference style to determine the displacement vector, the overall stiffness matrix, and the density vector included in the hub reference style, and determine an optimization model based on the displacement vector, the overall stiffness matrix, and the density vector; use the optimization model to perform topology optimization processing on the initial hub style to determine the first hub optimized style, where the optimization model includes: a flexibility term and a difference term, and the optimization model is:

[0074]

[0075] Wherein, is the design variable, representing the density of the element , is the density vector, is the displacement vector, T is the transpose of the matrix, is the overall stiffness matrix, is the flexibility term, is the difference term, is the hub reference style, is the L1 norm, is the similarity parameter, The larger it is, the closer the optimal design is to the hub reference style. For a smaller , the hub reference style is ignored and optimized to minimize compliance, is the volume fraction, is the material volume, is the design domain volume.

[0076] In one implementation, the above topology optimization module 404 is further configured to: the flexibility term in the optimization model is:

[0077]

[0078] Wherein, is the displacement vector, K is the overall stiffness matrix, is the flexibility, is the element displacement vector, is the element stiffness matrix, is the volume fraction, is the number of elements, For the density of the unit, is the material volume, is the design domain volume.

[0079] In one implementation, after the step of obtaining the new style ratio corresponding to the optimized style of the second hub, the above-mentioned screening and filtering module 406 is further configured to: when the new style ratio is not less than a preset ratio threshold, use a preset generation model to perform auxiliary design processing on the optimized style of the second hub to determine the optimized style of the third hub.

[0080] In one implementation, the generation model includes a preset generative adversarial network, and the preset generative adversarial network includes: a generator and a discriminator. Before the step of using the preset generation model to perform auxiliary design processing on the optimized style of the second hub to determine the optimized style of the third hub, the above-mentioned screening and filtering module 406 is further configured to: obtain a random noise vector and send the random noise vector to the generator to provide randomness for hub style generation, and determine random hub styles with different styles; send the random hub styles to the discriminator, so that the generator and the discriminator perform adversarial training to determine the target generative adversarial network, and use the target generative adversarial network to perform auxiliary design processing on the optimized style of the second hub to determine the optimized style of the third hub.

[0081] In one implementation, after the step of using the preset generation model to perform auxiliary design processing on the optimized style of the second hub to determine the optimized style of the third hub, the above-mentioned screening and filtering module 406 is further configured to: perform secondary screening processing on the optimized style of the third hub through a preset L1 loss function to filter out similar styles, determine the optimized style of the fourth hub, and end the topological optimization of this round, and re-perform topological optimization processing on the optimized style of the fourth hub.

[0082] In one implementation, when performing the step of evaluating the style of the target hub style to generate a visual evaluation result, the above-mentioned hub style evaluation module 408 is further configured to: quantitatively evaluate the innovation of the target hub style through a preset autoencoder, determine the structural innovation evaluation result of the target hub style according to the difference between the target hub style and the hub reference style, and perform a physical performance test on the target hub style to determine the feasibility evaluation result of the target hub style; integrate the structural innovation evaluation result and the feasibility evaluation result to determine the visual evaluation result.

[0083] The device provided by the embodiments of the present invention has the same implementation principle and the same technical effects as those of the foregoing method embodiments. For the sake of brief description, for the parts not mentioned in the device embodiments, reference may be made to the corresponding contents in the foregoing method embodiments.

[0084] An embodiment of the present invention provides a server. Specifically, the server includes a processor and a storage device; a computer program is stored on the storage device, and when the computer program is run by the processor, it executes the method described in any one of the above-described embodiments.

[0085] Figure 5 FIG. 4 is a schematic structural diagram of a server provided by an embodiment of the present invention. The server 100 includes: a processor 50, a memory 51, a bus 52, and a communication interface 53. The processor 50, the communication interface 53, and the memory 51 are connected through the bus 52; the processor 50 is configured to execute an executable module stored in the memory 51, such as a computer program.

[0086] Among them, the memory 51 may include a high-speed random access memory (RAM, Random Access Memory), and may also include a non-volatile memory, such as at least one disk memory. Through at least one communication interface 53 (which can be wired or wireless), a communication connection is established between the system network element and at least one other network element. The Internet, wide area network, local area network, metropolitan area network, etc. can be used.

[0087] The bus 52 may be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 5 only a bidirectional arrow is shown in FIG. 4, but it does not mean that there is only one bus or one type of bus.

[0088] Among them, the memory 51 is used to store a program. After receiving an execution instruction, the processor 50 executes the program. The method executed by the device defined by the flow process disclosed in any one of the foregoing embodiments of the present invention can be applied to the processor 50 or implemented by the processor 50.

[0089] The processor 50 may be an integrated circuit chip with the ability to process signals. In the implementation process, each step of the above method may be completed by the integrated logic circuit of the hardware in the processor 50 or the instructions in the form of software. The above-mentioned processor 50 may be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it may also be a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field-programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention may be directly embodied as being executed and completed by the hardware decoding processor, or executed and completed by a combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 51, and the processor 50 reads the information in the memory 51 and combines its hardware to complete the steps of the above method.

[0090] The computer program product of the readable storage medium provided by the embodiments of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the method described in the foregoing method embodiments. For the specific implementation, reference can be made to the foregoing method embodiments, and details are not described herein again.

[0091] When the above-mentioned functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0092] Finally, it should be noted that the above-mentioned embodiments are only specific implementation manners of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments, or can easily think of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes, or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A hub style generation method based on deep learning and topology optimization, characterized in that, The method includes: Obtaining an original data set, where the original data set includes a preset number of hub reference styles based on existing vehicle hub styles; Based on the hub reference styles, performing topology optimization on an initial hub style to reduce the flexibility of the initial hub style and the difference between the initial hub style and the hub reference styles, and determining a first optimized hub style, where flexibility is used to represent engineering performance, and the smaller the flexibility, the better the engineering performance of the hub style; Through a preset L1 loss function, screening the optimized hub style to determine a second optimized hub style, and obtaining the corresponding new style ratio of the second optimized hub style, where the new style ratio is the ratio of the number of new hub styles in the current iteration to the total number of hub styles in the previous iteration, and the preset L1 loss function is used to quantify the innovation degree of the newly designed style, screen the optimized hub style to filter out similar styles, and evaluate the innovation degree and compliance of physical characteristics of the design solutions obtained through iterative design exploration during the evaluation design stage; When the new style ratio is less than a preset ratio threshold, determining the second optimized hub style as the target hub style, and performing style evaluation on the target hub style to generate a visual evaluation result; Among them, after the step of obtaining the new style ratio corresponding to the second optimized hub style, it includes: when the new style ratio is not less than the preset ratio threshold, using a preset generation model to perform assisted design on the second optimized hub style to determine a third optimized hub style; Among them, by calculating the ratio of the new design solution to the total number of solutions in the previous round of iteration, evaluating the new design ratio generated in the current iteration to balance the system innovation and exploration cost during the generation of new designs. Among them, if the generated new design ratio is higher than the ratio threshold during the iteration process, continue to iterate through the generation model to generate more solutions. If the generated new design ratio is not higher than the ratio threshold during the iteration process, it is determined that the design has tended to be stable, stop further iteration and enter the final evaluation stage; Among them, after the step of using a preset generation model to perform assisted design on the second optimized hub style to determine a third optimized hub style, it includes: through a preset L1 loss function, performing secondary screening on the third optimized hub style to determine a fourth optimized hub style, and ending the topology optimization of this round, and re-performing topology optimization on the fourth optimized hub style; Among them, the step of performing style evaluation on the target hub style to generate a visual evaluation result includes: quantitatively evaluating the innovation of the target hub style through a preset autoencoder, determining the structural innovation evaluation result of the target hub style according to the difference between the target hub style and the hub reference styles, and performing physical performance testing on the target hub style to determine the feasibility evaluation result of the target hub style; integrating the structural innovation evaluation result and the feasibility evaluation result to determine the visual evaluation result; Among them, when conducting quantitative evaluation, initial weights are assigned to various design attributes according to the historical data and design objectives in the hub reference style collection stage, and adjustable parameters are dynamically adjusted in the topology optimization and generative adversarial network iterations to reflect the weight relationship between different attributes through the adjustable parameters.

2. The method for generating a wheel hub style based on deep learning and topology optimization according to claim 1, wherein The step of performing topology optimization on the initial hub style based on the hub reference style to reduce the flexibility of the initial hub style and the difference between the initial hub style and the hub reference style, and determining the first optimized hub style includes: Performing data parsing on the hub reference style to determine the displacement vector, overall stiffness matrix, and density vector included in the hub reference style, and determining an optimization model based on the displacement vector, the overall stiffness matrix, and the density vector; Performing topology optimization on the initial hub style using the optimization model to determine the first optimized hub style, where the optimization model includes: a flexibility term and a difference term, and the optimization model is: Among them, is a design variable representing the density of the element, is the density vector, is the displacement vector, T is the transpose of the matrix, is the global stiffness matrix, is the compliance term, is the difference term, is the hub reference pattern, is the L1 norm, is the similarity parameter, The larger it is, the closer the optimal design is to the hub reference pattern. For less than the preset similarity parameter threshold, the hub reference pattern is ignored and optimized to minimize the compliance, is the volume fraction, is the material volume, is the design domain volume, is the number of elements.

3. The method for generating a wheel hub style based on deep learning and topology optimization according to claim 2, wherein The flexibility term in the optimization model is: Among them, is the displacement vector, K is the global stiffness matrix, is the flexibility, is the element displacement vector, is the element stiffness matrix, is the volume fraction, is the number of elements, is the density of the element, is the material volume, is the design domain volume.

4. The method for generating a wheel hub style based on deep learning and topology optimization according to claim 1, characterized in that, The generation model includes a preset generative adversarial network, and the preset generative adversarial network includes: a generator and a discriminator. Before the step of using the preset generation model to perform auxiliary design on the second optimized hub style to determine the third optimized hub style, it includes: Obtaining a random noise vector and sending the random noise vector to the generator to provide randomness for hub style generation and determining random hub styles of different styles; Sending the random hub styles to the discriminator to enable the generator and the discriminator to perform adversarial training to determine the target generative adversarial network, and using the target generative adversarial network to perform auxiliary design on the second optimized hub style to determine the third optimized hub style.

5. A hub style generation device based on topology optimization, characterized in that The device includes: A data acquisition module that acquires an original data set, where the original data set includes: a preset number of hub reference styles based on existing vehicle hub styles; A topology optimization module that performs topology optimization on the initial hub style based on the hub reference style to reduce the flexibility of the initial hub style and the difference between the initial hub style and the hub reference style, and determines the first optimized hub style, where flexibility is used to represent engineering performance, and the smaller the flexibility, the better the engineering performance of the hub style; A screening and filtering module that screens the optimized hub styles through a preset L1 loss function to determine the second optimized hub style and obtains the new style ratio corresponding to the second optimized hub style, where the new style ratio is the ratio of the number of new hub styles in the current iteration to the total number of hub styles in the previous iteration, and the preset L1 loss function is used to quantify the innovation degree of the newly designed style, screen the optimized hub styles to filter out similar styles, and evaluate the innovation degree and compliance of physical characteristics of the design solutions obtained through iterative design exploration in the evaluation design stage; The hub style evaluation module determines the second optimized hub style as the target hub style when the new style ratio is less than the preset ratio threshold, and performs style evaluation processing on the target hub style to generate a visual evaluation result; Among them, after the step of obtaining the new style ratio corresponding to the second optimized hub style, it includes: when the new style ratio is not less than the preset ratio threshold, using a preset generation model to perform auxiliary design processing on the second optimized hub style to determine the third optimized hub style; Among them, by calculating the ratio of the new design scheme to the total number of schemes in the previous round of iteration, the new design ratio generated in the current iteration is evaluated to balance the system innovation and exploration cost when generating a new design. Among them, if the new design ratio generated during the iteration is higher than the ratio threshold, continue to iterate through the generation model to generate more schemes. If the new design ratio generated during the iteration is not higher than the ratio threshold, it is determined that the design has tended to be stable, stop further iteration and enter the final evaluation stage; Among them, after the step of using a preset generation model to perform auxiliary design processing on the second optimized hub style to determine the third optimized hub style, it includes: performing secondary screening processing on the third optimized hub style through a preset L1 loss function to determine the fourth optimized hub style, and ending the topological optimization of this round, and re-performing topological optimization processing on the fourth optimized hub style; Among them, the step of performing style evaluation processing on the target hub style to generate a visual evaluation result includes: quantitatively evaluating the innovation of the target hub style through a preset autoencoder, determining the structural innovation evaluation result of the target hub style according to the difference between the target hub style and the hub reference style, and performing physical performance tests on the target hub style to determine the feasibility evaluation result of the target hub style; integrating the structural innovation evaluation result and the feasibility evaluation result to determine the visual evaluation result; Among them, when performing quantitative evaluation, initial weights are assigned to each design attribute according to the historical data and design objectives in the hub reference style collection stage, and the adjustable parameters are dynamically adjusted in topological optimization and generative adversarial network iteration to reflect the weight relationship between different attributes through the adjustable parameters.

6. A server, characterized in that, It includes a processor and a memory. The memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are called and executed by the processor, the computer-executable instructions cause the processor to implement the method according to any one of claims 1 to 4.

Images