Intelligent generation type design method of anti-collision beam

Abstract

The invention relates to the technical field of automobile design, in particular to an intelligent generation type design method of an anti-collision beam, which comprises the following steps: firstly, performing experimental analysis on the anti-collision beam to obtain a section image comprising the anti-collision beam and performance response data corresponding to the section image; performing image recognition and text extraction on the section image; fusing the recognized image and the extracted text by using a multi-modal multi-layer fusion model to obtain a multi-modal design variable; training the constructed conditional variation network by using the multi-modal design variables and the performance response data corresponding to the multi-modal design variables to obtain a conditional variation model; generating design variables by using the conditional variation model, and predicting performance response data corresponding to the variables; optimal performance response data are screened out from the obtained performance response data, and then the optimal design scheme of the anti-collision beam to be designed is obtained. According to the method, the optimal design scheme is determined by using the conditional variation network, and the efficiency and the precision of optimization design are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of automotive design, and particularly relates to an intelligent generative design method for a bumper beam. Background Art

[0002] The front bumper beam is an important safety component at the front end of an automotive body-in-white, and its performance directly affects the vehicle's collision safety. At present, the materials used for the front bumper beam are mainly high-strength steel, aluminum alloy, and carbon fiber reinforced composite (CFRP). CFRP and aluminum alloy front bumper beams are light in weight but high in price, and are mostly used in high-class models. High-strength steel front bumper beams have a low cost but a large mass, and are commonly used in economy models. Existing single-material front bumper beams have their own advantages and disadvantages, and their application ranges are limited. To meet the major technical requirements of lightweight vehicle body research and development in China, CFRP / steel hybrid structure front bumper beams have been proposed. They can give full play to the unique performance advantages of CFRP and steel, and can achieve low cost, high performance, and lightweight of the front bumper beam, with broad application prospects. At present, the research on the design scheme of CFRP / steel hybrid structure bumper beams is still in the exploratory stage. Most inventors tentatively set the shapes of the CFRP layer and the metal layer (such as using common "mouth" shapes, "Ji" shapes, "Mu" shapes, etc.), and the connection type between the two is mostly stacked by complete wrapping. It is difficult to effectively guarantee the performance of the bumper beam composed of the CFRP layer and the metal layer, and it is necessary to repeatedly try and adjust to determine a reasonable structural shape and connection type. This design method has great blindness and low design efficiency. Therefore, there is a lack of an intelligent design method in the existing field that can automatically determine the shapes and connection types of the CFRP layer and the metal layer according to the performance indicators (such as weight, collision peak force, collision intrusion amount, collision energy absorption) of the CFRP / steel hybrid structure bumper beam. At present, there is a technical blank in this field of research, which needs to be further overcome.

[0003] At present, the traditional design method for automotive structures adopts the concept of "optimizing structural parameters → meeting performance requirements", and uses optimization algorithms to repeatedly iterate and update design variables. It mainly combines surrogate models (such as response surface, Kriging, radial basis neural network, etc.) and intelligent optimization algorithms (genetic algorithm, particle swarm algorithm, etc.). Due to the complex characteristics of multi-modal, high-dimensional, strong coupling, and mixed superposition of discrete / continuous variables presented by the design variables of the CFRP / steel hybrid structure, and the highly non-linear performance response. It is difficult for traditional surrogate model methods to effectively construct the "multi-modal high-dimensional variable-response" mapping relationship, and it is difficult to form a two-way mapping between the two. To obtain a set response value, it is only possible to repeatedly optimize and iterate the design variables. The calculation is complex and the high-dimensional variables often make the optimization algorithm difficult to converge. If the response conditions change, it is necessary to re-iterate and optimize. And during optimization, it is difficult to constrain the design variables.

[0004] In recent years, deep generative models have been proposed in artificial intelligence. They can quickly create optimal designs based on objectives without the need for an iterative optimization process and can fuse multivariate data, providing a solution to this problem. Existing research has generated a large number of design variables. To select the optimal design method from this vast array of design variables, it is still necessary to conduct experiments or simulation analysis on the design variables to obtain performance index values, and then make a comprehensive decision based on the values ​​of the design variables and performance responses. This technical solution makes the calculation steps of deep generative design more complex and significantly reduces design efficiency. Therefore, it is necessary to improve the traditional deep generative design method to directly derive the optimal design variables and reduce the intermediate solution steps. To date, no relevant research has reported on this design method. Summary of the Invention

[0005] In view of this, the present invention aims to provide an intelligent generative design method for anti-collision beams, innovatively uses an artificial intelligence deep generative model in the integrated design of the structural shape and structural dimensions of carbon / steel composite structure anti-collision beams, and uses the proposed conditional variational network to determine the optimal design scheme, solving the problems of low computational efficiency, poor nonlinear solution accuracy, and inability to directly map and solve the optimal design variables according to the response quantity requirements of traditional design methods for optimizing design variables, realizing one-step end-to-end intelligent design, and providing new approaches and methods for the design and development of automotive parts.

[0006] To achieve the above object, the technical solution created by the present invention is implemented as follows: An intelligent generative design method for an anti-collision beam, comprising: S1: Perform experimental analysis on the anti-collision beam to obtain a data set. The data set includes a cross-sectional image of the anti-collision beam and performance response data corresponding to the cross-sectional image. S2: performing image recognition and text extraction on the cross-sectional images in the data set in step S1 to obtain the structural shape and structural dimensions respectively; fusing the structural shape and structural dimensions to obtain multimodal design variables; S3: Using the multimodal design variables obtained in step S2 as input and the performance response data corresponding to the multimodal design variables as output, the constructed conditional variational network is trained to obtain a conditional variational model; S4: extracting multimodal design variables from the anti-collision beam to be designed according to steps S1 and S2, inputting the obtained multimodal design variables into the conditional variational model obtained in step S3 to obtain corresponding performance response data; S5: Filter out the optimal performance response data based on the multimodal design variables and performance response data generated in step S4, and derive the optimal design scheme of the anti-collision beam to be designed based on the optimal performance response data.

[0007] Furthermore, in step S1, the anti-collision beam is experimentally analyzed, and the process of obtaining a data set includes: establishing a simulation model of the anti-collision beam to obtain a simulation data set; the simulation data set includes a cross-sectional image of the simulation model and its corresponding performance response; and performing an actual collision test on the anti-collision beam to obtain a real data set, the real data set includes a cross-sectional image of the anti-collision beam and its corresponding performance response value.

[0008] Furthermore, a simulation model of the anti-collision beam is established, and the process of obtaining a simulation data set includes: establishing an implicit parameterized model including the structural shape and structural dimensions of the anti-collision beam; determining sample points of different structural shapes and structural dimensions in the anti-collision beam based on the Hammersley experimental design method; obtaining cross-sectional images of the sample points; performing low-speed collision finite element simulation analysis on the sample points, and extracting the performance response value corresponding to each sample point from the simulation results; the simulation data set includes the cross-sectional image and the corresponding performance response value.

[0009] Furthermore, in step S2, the process of extracting the structural shape and structural dimensions of the cross-sectional image in the simulation data set includes: reading the structural dimensions and structural shape of the cross-section of the anti-collision beam from the Hamsley test design list obtained by the Hamsley test design method.

[0010] Furthermore, actual collision tests are conducted on the anti-collision beam. The process of obtaining a real data set includes: conducting multiple collision tests on the anti-collision beam, extracting the performance response values ​​measured in the experiments, and extracting a cross-sectional image of the anti-collision beam from the image results of the experiments; the real data set includes the cross-sectional image and the corresponding performance response values.

[0011] Furthermore, in step S2, the process of extracting the structural shape and structural size of the cross-sectional images in the real data set includes: using a convolutional neural model to identify the cross-sectional images in the real data set to obtain the corresponding structural shape; pre-processing the cross-sectional images in the real data set by edge detection and contour extraction, and then calibrating the scale and converting the length and thickness parameters to obtain the corresponding structural size.

[0012] Furthermore, in step S2, the process of fusing the structural shape and the structural size to obtain the multimodal design variables includes: integrating the structural shape of the cross-sectional image in the real data set and the structural shape of the cross-sectional image in the simulation data set to obtain a structural shape set; integrating the structural size extracted from the real data set and the structural size extracted from the simulation data set to obtain a structural size set; inputting the structural shape set and the structural size set into the multimodal multi-layer fusion model for fusion to obtain the multimodal design variables.

[0013] Furthermore, in the conditional variational network of step S3: the multimodal design variables are encoded using an encoder based on a multilayer perceptron, the input multimodal design variables are mapped to a latent space, and the mean and logarithmic variance of the latent space are obtained; the latent space is reparameterized and sampled based on the logarithmic variance and the mean to obtain latent variables; the latent variables and the set conditional performance responses are input into a decoder based on a mixed density network for decoding to obtain a mixed Gaussian parameter distribution; the mixed Gaussian parameter distributions are combined according to weights to obtain generated design variables; a Gaussian process regression model is used to perform performance prediction on the generated design variables to obtain corresponding performance response data.

[0014] Furthermore, in step S3, the conditional variational network is trained by jointly optimizing the loss function; the joint optimization loss function is: ; Among them, L represents the joint optimization loss function, represents the reconstruction loss, represents the network parameters of the decoder, X represents the multimodal design variables, Y represents the corresponding performance response data, and Z represents the latent variables. represents the KL divergence, represents the network parameters of the encoder, represents the negative feedback term, Represents the performance response data predicted by the conditional variation model obtained from the current training, Represents conditional performance response data, represents the KL divergence scaling coefficient, Represents the negative feedback proportional coefficient; the negative feedback proportional coefficient is obtained by the following formula: ; in, It represents the error between the predicted performance response data and the corresponding conditional performance response data, represents the error variance.

[0015] Furthermore, the process of step S5 includes: The performance response data obtained by S4 are analyzed using the principal component analysis method to obtain the objective weight value of each performance response; the expert scoring method and the hierarchical analysis method are used to obtain the subjective weight value of each performance response, and the subjective and objective weight values ​​are solved by the zero-sum game method to obtain the comprehensive weight value; the grey correlation analysis method is used to obtain the grey correlation coefficient of each group of performance response schemes in the performance response data generated by S4, and the grey correlation coefficient is multiplied by the comprehensive weight value to obtain the grey correlation degree of each group of performance response schemes; all performance response schemes are sorted according to the size of the grey correlation degree, and the scheme with the largest grey correlation degree value is the optimal performance response scheme, and the corresponding relevant parameters of the anti-collision beam to be designed are the optimal design scheme of the anti-collision beam.

[0016] Compared with the prior art, the present invention can achieve the following beneficial effects: (1) The present invention creates an intelligent generative design method for the anti-collision beam, which uses an artificial intelligence deep generative model in the integrated design of the structural shape and structural dimensions of the carbon / steel composite structure anti-collision beam. On the basis of the traditional conditional variational network architecture, a lightweight Gaussian process regression model and a multi-criteria decision model are introduced, and the error feedback term is iterated into the network learning by reverse error propagation, thereby improving the generation accuracy. The proposed improved conditional variational network architecture can solve the optimal design scheme in one step, realize direct end-to-end intelligent design from conditional performance response to optimal design variables, greatly improve the design efficiency, provide a new technical approach and method for automobile structural design, and solve the problem that the traditional design method of optimizing design variables to obtain the optimal response quantity has low computational efficiency, poor nonlinear solution accuracy, and cannot directly map and solve the optimal design variables according to the response quantity requirements, thus providing a new approach and method for the design and development of automobile parts; (2) The present invention creates an intelligent generative design method for the anti-collision beam. Based on image recognition technology, CNN is used to identify and predict the structural shape of the carbon / steel composite anti-collision beam. Edge detection, contour extraction, proportion calibration and parameter conversion methods are used to identify and predict the structural dimensions of the carbon / steel composite anti-collision beam. This solves the technical problem of extracting and calibrating the shape and dimensions of the complex and diverse sections of the carbon / steel composite anti-collision beam, greatly improving the design efficiency. In addition, the present invention integrates simulation and test data to enhance the accuracy of the model, and multimodally fuses the image variables of the structural shape and the text variables of the structural dimensions to achieve an integrated collaborative design of the structural shape and structural dimensions. This solves the problem that the connection shape and dimensions of the CFRP layer and the metal layer of the carbon / steel composite anti-collision beam are difficult to determine directly and require repeated trial and error. This reduces the blindness of the design and improves the design efficiency. (3) The present invention creates an intelligent generative design method for the anti-collision beam, which introduces principal component analysis, expert scoring, hierarchical analysis and zero-sum game methods to obtain comprehensive weight values ​​for evaluating various performance responses, thereby ensuring the reliability and scientific nature of the weights and improving the accuracy of predicting performance responses. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which constitute part of the present invention, are intended to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are intended to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings: Figure 1 A schematic diagram of a flow chart of an intelligent generative design method for an anti-collision beam according to an embodiment of the present invention; Figure 2 A schematic diagram of the process of obtaining a simulation data set according to an embodiment of the present invention; Figure 3 A schematic diagram of the structural shape of the steel plate layer and the CFRP layer of the anti-collision beam according to an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating an example of the structural shape of the anti-collision beam according to an embodiment of the present invention; Figure 5 A schematic diagram of the process of acquiring a "multimodal design variable-performance response" data set according to an embodiment of the present invention; Figure 6 A schematic diagram of the framework of the conditional variational network described in the embodiment of the present invention.

[0018] Description of reference numerals: 1. Anti-collision beam; 101. Steel material layer; 102. CFRP layer. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0020] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other.

[0021] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second" and the like are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, features defined as "first", "second" and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0022] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "connected," and "connected" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to internal connections between two components. Those skilled in the art can understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0023] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0024] like Figure 1 As shown in Figure 2, the intelligent generative design method for anti-collision beams includes: S1: Perform experimental analysis on the anti-collision beam to obtain a data set. The data set includes a cross-sectional image of the anti-collision beam and performance response data corresponding to the cross-sectional image.

[0025] In some embodiments, the process of performing experimental analysis on an anti-collision beam and obtaining a data set includes: establishing a simulation model of the anti-collision beam to obtain a simulation data set; and conducting an actual collision test on the anti-collision beam to obtain a real data set. The simulation data set includes cross-sectional images of the simulation model and their corresponding performance response values, and the real data set includes cross-sectional images of the anti-collision beam and their corresponding performance response values. The cross-sectional images in the simulation and real data sets represent the structural dimensions and shape of the anti-collision beam.

[0026] Specifically, the process of establishing a simulation model of the anti-collision beam and obtaining a simulation data set is as follows: Figure 2 As shown, including: S101: Establish an implicit parametric model including the structural shape and structural dimensions of the anti-collision beam. In the embodiments of the present invention, a professional implicit parametric modeling software, such as SFE / Concept software, is used to establish a parametric model of the carbon / steel composite structural anti-collision beam for vehicles. The parameters include the structural dimensions and structural shape of the anti-collision beam.

[0027] S102: Determine sample points with different structural shapes and structural dimensions in the anti-collision beam based on the Hammersley experimental design method. In the embodiments of the present invention, multidisciplinary optimization software, such as Isight and Hyperstudy, etc., is adopted. Based on the experimental design module, the Hammersley experimental design method is used to generate a large number of data points, and a large number of sample points with different structural shapes and structural dimensions are generated through the parametric model of the anti-collision beam.

[0028] S103: Obtain cross-sectional images of the sample points. The structural shapes of the steel plate layer and the CFRP (carbon fiber reinforced plastic) layer include "Ji" shape, "Kou" shape, "Ri" shape, "Mu" shape, half "Kou" shape, incompletely enclosed "Kou" shape, etc., as Figure 3 shown. At the same time, various structural shapes can be formed after the composite of the steel plate layer and the CFRP layer. For example, Figure 4 lists some possible formed structural shapes. Figure 4 It shows that in the structure of the anti-collision beam 1, there are various superimposing methods between the steel material layer 101 and the CFRP layer 102. What the method provided by the present invention needs to solve is how to directly determine the superimposing method between the steel material layer 101 and the CFRP layer 102 through end-to-end intelligent design, as well as the dimensions such as the thickness and length between the steel material layer 101 and the CFRP layer 102.

[0029] S104: Conduct low-speed collision finite element simulation analysis on the sample points, and extract the performance response values corresponding to each sample point from the simulation results. In the embodiments of the present invention, according to the requirements of national standards, low-speed impact simulation boundary conditions are given, and LS-DYNA software is called to conduct low-speed collision finite element simulation analysis on the sample points respectively, and a performance response data set of the total weight, peak force, energy absorption, and intrusion amount of the anti-collision beam is obtained.

[0030] S105: The simulation data set includes cross-sectional images and corresponding performance response values.

[0031] Specifically, the process of obtaining the real data set by conducting an actual collision test on the anti-collision beam includes: S111: Conduct multiple collision tests on the anti-collision beam, extract the experimentally measured performance response values, and extract a cross-sectional image of the anti-collision beam from the captured images of the experiments. In this embodiment of the present invention, based on a large number of low-speed collision tests accumulated during the historical development of CFRP / steel hybrid anti-collision beams, the measured performance response values ​​are extracted, and the cross-sectional morphology of the anti-collision beam is extracted from the captured images of the experiments.

[0032] S112: The real data set includes cross-sectional images and corresponding performance response values.

[0033] S2: Perform image recognition and text extraction on the cross-sectional images in the data set in step S1 to obtain the structural shape and structural dimensions respectively; fuse the structural shape and structural dimensions to obtain multimodal design variables. The multimodal design variables and the corresponding performance response data together constitute the "multimodal design variable-performance response" data set. The process of obtaining the "multimodal design variable-performance response" data set in step S2 is as follows: Figure 5 shown.

[0034] It should be noted that since the cross-sectional images in the simulation dataset are automatically generated by software using preset structural shapes and dimensions, image recognition of the simulation dataset is no longer required. Therefore, only text extraction of the simulation dataset is required to obtain the structural dimensions and shapes. Specifically, in some embodiments, the text extraction process for the simulation dataset includes reading the structural dimensions and shapes of the cross-sectional area of ​​the anti-collision beam from the Hamsler experimental design list obtained using the Hamsler experimental design method to complete text extraction from the simulation dataset. The structural dimensions of the cross-sectional area include actual measurable dimensions such as the thickness and length of the structure.

[0035] In step S2 of the embodiment of the present invention, the process of performing image recognition and text extraction on the cross-sectional image in the real data set to obtain the structural shape and structural size respectively includes: S211: Using a convolutional neural model to identify cross-sectional images in a real dataset, the corresponding structural shapes are obtained. In an embodiment of the present invention, the convolutional neural model used is a trained LeNet network model. The LeNet network structure includes a convolutional layer, a pooling layer, and a fully connected layer. A cross-entropy loss function is specified, and a gradient descent method is used to optimize the parameters of the trained LeNet network to obtain a trained LeNet network model. The output of the LeNet network model is then compressed using an embedding method to obtain an image feature vector representing the structural shape, thereby completing image recognition of the real dataset. During the training of the LeNet network, the cross-sectional images in the real dataset are normalized. At the same time, the image dataset is divided into a training set, a validation set, and a test set. The training set is used to train the LeNet network, and the validation set is used to update the LeNet network parameters through a backpropagation algorithm to improve the model accuracy. Finally, the test set is used to evaluate and adjust the model.

[0036] S212: After preprocessing the cross-sectional images in the real dataset for edge detection and contour extraction, calibration and length and thickness parameter conversion are performed to obtain the structural dimensions of the anti-collision beam cross section, completing text extraction from the real dataset. In this embodiment of the present invention, edge detection and contour extraction can directly utilize existing image recognition techniques, such as the Canny algorithm for edge detection and the Sobel algorithm for edge extraction, to obtain the structural dimensions of the anti-collision beam cross section.

[0037] In some embodiments, the process of fusing structural shapes and structural dimensions using a multimodal multi-layer fusion model to obtain multimodal design variables includes: integrating the structural shapes of the cross-sectional images in the real data set with the structural shapes reflected by the cross-sectional images in the simulation data set to obtain a structural shape set; integrating the structural dimensions extracted from the real data set with the structural dimensions extracted from the simulation data set to obtain a structural dimension set; and inputting the structural shape set and the structural dimension set into the multimodal multi-layer fusion model for fusion to obtain multimodal design variables. In an embodiment of the present invention, the multimodal multi-layer fusion model adopts the deep neural network model disclosed in the invention patent application with Chinese patent publication number CN110674677A, publication date January 10, 2020, and patent name "A Deep Neural Network for Face Anti-Spoofing with Multimodal Multi-layer Fusion". During use, the structural shape set and the structural dimension set are input into this deep neural network model to output multimodal design variables.

[0038] S3: Take the multimodal design variables obtained in step S2 as input and the performance response data corresponding to the multimodal design variables as output, train the constructed conditional variational network, and obtain a conditional variational model.

[0039] This invention builds a conditional variational network based on the conditional variational autoencoder (cVAE), a deep learning model consisting of an encoder and a decoder. The encoder extracts key latent variables from the input data and maps the input data to a probability distribution in the latent space. The encoder's core goal is to construct the latent space of the input data through dimensionality reduction, preserving key information while identifying hidden patterns. The decoder uses these latent variables to reconstruct the original input data, restoring the latent variables to the same dimensionality as the input data. Compared to the unsupervised learning of the variational autoencoder (VAE), the conditional variational autoencoder (cVAE) introduces a supervisory element, enabling the model to learn labeled conditional information and enhancing the controllability of the decoder's output. Based on the trained neural network framework, the decoder controls the input of the supervisory element, namely the conditional performance response indicator, to obtain the required design variables, thus achieving reverse design. The invention then iteratively trains the conditional variational network until a performance indicator dataset with an accuracy of over 95% is generated and the iterations converge. At this point, the training of the conditional variational network is completed, resulting in a conditional variational model.

[0040] In some embodiments, in the conditional variational network of step S3, an encoder based on a multi-layer perceptron is used to encode the multimodal design variables, and the input multimodal design variables are mapped to the latent space to obtain the probability distribution of the latent space, specifically, the mean and logarithmic variance of the latent space are obtained; the latent space is reparameterized and sampled based on the logarithmic variance and the mean to obtain the latent variables; the latent variables and the set conditional performance responses are input into a decoder based on a mixed density network for decoding to obtain a mixed Gaussian parameter distribution, where the conditional performance response refers to the performance response that needs to meet certain conditions, which can be a fixed value or a fixed range of values. The conditional performance response is adaptively selected and adjusted according to the actual situation; the mixed Gaussian parameter distribution is combined according to the weights to obtain the generated design variables; the Gaussian process regression model (GPR) performs performance prediction on the generated design variables to obtain corresponding performance response data. The GPR model is a lightweight machine learning prediction model that is used to input design variables into the model and quickly predict performance response values. To ensure that the GPR model meets the accuracy requirements, pre-training is required. 80% of the data is extracted from the multimodal design variable and performance response data set obtained from simulation and experiments as the test set, and the remaining 20% ​​is used as the validation set. The model can only be used after the accuracy reaches 95% or above after testing.

[0041] The conditional variational network provided by the embodiment of the present invention is as follows Figure 6 As shown, the encoder based on the multi-layer perceptron (MLP) encodes the multimodal design variables to obtain the mean of the multimodal design variables and log variance , represents the variance of the multimodal design variable, Represents the standard deviation of the multimodal design variable; based on the logarithmic variance and mean, reparameterized sampling is performed to obtain the latent variable z, that is, ,in, represents the noise variable that satisfies the standard normal distribution, that is, ; Combine the latent variable z with the set conditional performance response data , the input is decoded in the decoder based on the mixture density network (MDN) to obtain the mixed Gaussian parameter distribution, which includes the mixing weight , mean and variance ; Select the kth Gaussian distribution according to the mixed weight of the mixed Gaussian parameter distribution, and then extract a sample from the selected Gaussian distribution to obtain the generated design variable ; Gaussian process regression model for generating design variables Perform performance prediction to obtain the performance response data predicted by the conditional variation model currently trained .

[0042] In this embodiment of the present invention, an encoder based on a multi-layer perceptron is constructed using five fully connected layers. Each fully connected layer has 256 neurons, the activation function is ReLU, and residual connections are introduced. The multimodal design variables are input into the encoder, and the dimensionality is gradually reduced through the fully connected layers to learn the nonlinear relationship between the design space and the performance index, and the mean of the potential space distribution is output. and log variance By reparameterizing the sampling, the latent variable z is obtained, transforming the non-differentiable random sampling process into a differentiable computational process. This solves the problem of gradient backpropagation and facilitates optimization using gradient descent. Furthermore, this embodiment of the present invention constructs a decoder based on a mixture density network using five fully connected layers, each with 256 neurons and a Tanh activation function. The latent variable and a preset conditional performance response are input into the decoder, which outputs a mixed Gaussian parameter distribution. The decoder based on the mixture density network dynamically learns the parameters (mixing weights, mean, and variance) of multiple underlying probability distributions (such as Gaussian distributions) and linearly combines these distributions according to the weights to construct a multimodal probabilistic model, accurately fitting the complex distribution characteristics of the target data. Fully connected layers are the basic building blocks of multilayer perceptrons and mixture density networks. Each layer in a fully connected layer is fully connected, with the bottom layer being the input layer, the middle layer being the hidden layer, and the final layer being the output layer. The input of the fully connected layer is flattened into a one-dimensional vector, which is linearly transformed by multiplication of the weight matrix and the addition of bias terms.

[0043] In order to improve the matching accuracy between the performance response corresponding to the design variables generated by the conditional variational model and the conditional performance response, the present invention proposes a back propagation adjustment strategy. Specifically, in some embodiments, the generated design variables Obtain the corresponding performance response value through GPR regression model prediction , which is compared with the conditional performance response Error The decoder parameters are adjusted through backpropagation. Specifically, the conditional variational network is trained by jointly optimizing the loss function. In this invention, a negative feedback term is introduced on the basis of the existing maximum evidence lower bound (ELBO) to form a joint optimization loss function. Specifically, the maximum evidence lower bound (ELBO) is as follows: ; Among them, l represents the maximum evidence lower bound, represents the network parameters of the decoder, represents the network parameters of the encoder, represents the reconstruction loss, represents the KL divergence, X represents the multimodal design variables, Y represents the corresponding performance response data, and Z represents the latent variables.

[0044] The negative feedback term introduced is , It represents the negative feedback proportional coefficient and is obtained from the following formula: ; in, Represents the error variance. At this time, the joint optimization loss function is: ; Among them, L represents the joint optimization loss function, Represents the KL divergence scaling coefficient, KL divergence scaling coefficient Adopt dynamic scheduling strategy (initial value 0.1, linearly increase to 1.0), negative feedback proportional coefficient Adaptive adjustment based on error sample variance.

[0045] In an embodiment of the present invention, a multimodal design variable and performance response data set is used to train the conditional variation network. Specifically, 80% of the multimodal design variable and performance response data set is divided into a training set, and 20% is divided into a validation set. The training set is used to train the conditional variation network, and the validation set calculates the loss value of the joint optimization loss function. The training cutoff condition of the conditional variation network is that the loss value of the validation set in the multimodal design variable and performance response data set decreases by less than 0.5% for 10 consecutive training rounds. In an embodiment of the present invention, the Adam algorithm is used as the optimizer for training the conditional variation network, and the default parameters are , , the training batch size is 200 and the learning rate is 10 -8 .

[0046] S4: According to steps S1 and S2, multimodal design variables are extracted from the anti-collision beam to be designed, and the obtained multimodal design variables are input into the conditional variational model obtained in step S3 to obtain corresponding performance response data.

[0047] S5: Based on the multimodal design variables and performance response data generated in step S4, the optimal design variable-performance response data is screened out, and then the optimal design scheme of the anti-collision beam to be designed is obtained.

[0048] In the embodiment of the present invention, the process of step S5 includes: using the principal component analysis method to analyze the performance response data obtained in S4 to obtain the objective weight value of each performance response; using the expert scoring method and the hierarchical analysis method to obtain the subjective weight value of each performance response, and solving the subjective and objective weight values ​​through the zero-sum game method to obtain a comprehensive weight value. This comprehensive weight value takes into account both the objective distribution characteristics of the data and the preference orientation of practical applications, avoiding the defects of maximum and minimum values ​​in a single subjective or objective weight method, and ensuring the reliability and scientific nature of the weight. Using the gray correlation analysis method, the gray correlation coefficient of each group of performance response solutions in the performance response data generated in S4 is obtained, and the gray correlation coefficient is multiplied by the comprehensive weight value to obtain the gray correlation degree of each group of performance response solutions. All performance response solutions are sorted according to the gray correlation degree. The solution with the largest gray correlation degree value is the optimal performance response solution, and the corresponding relevant parameters of the anti-collision beam to be designed are the optimal design solution of the anti-collision beam.

[0049] The present invention combines a multi-criteria decision-making module based on principal component analysis and grey relational analysis with a GPR model to achieve one-step end-to-end intelligent design. End-to-end here refers to the input end to the output end. The back-propagation adjustment strategy is to ensure the accuracy of the conditional variation model generation. For the traditional conditional variation model architecture, the conditional performance response is used as the input end and input into the conditional variation model. Only design variables that meet the performance index can be generated. However, the optimal design solution cannot be determined based on the design variable values ​​alone. It is necessary to re-simulate or test the generated multiple sets of design variables to obtain performance indicators. The simulation or test performance indicators are compared with the conditional performance response to verify the generated accuracy. If the generated accuracy does not meet the standard, the conditional variation model needs to be repeatedly debugged until the generated accuracy meets the requirements. Then, a comprehensive decision is made based on the performance index and design variables to select the optimal design solution. The optimal design solution here is the output end. The traditional conditional variation model architecture is difficult to achieve one-step end-to-end functions and requires multiple auxiliary functions to achieve it. Moreover, these auxiliary functions cannot be embedded in the conditional variation model in a network learning manner, resulting in the need for repeated trial and error, and thus low efficiency and accuracy. The purpose of the present invention is to provide a one-step, end-to-end design model based on a conditional variational model. This innovatively integrates a lightweight GPR prediction model into the backend of the conditional variational model to rapidly predict the corresponding performance response from the generated design variables, eliminating the need for resimulation or experimental analysis of multiple sets of design variables. Furthermore, the present invention calculates the error based on the GPR-predicted performance response and the conditional performance response, and incorporates the error term as a negative feedback term into the maximum evidence lower bound to form a joint optimization loss function. Through iterative learning, the accuracy of the conditional variational model generation is guaranteed to be above 95%, eliminating the need for manual and repeated debugging of the conditional variational model parameters, greatly improving design efficiency. Furthermore, the present invention uses a multi-criteria decision-making method to select the generated design variables and the GPR-predicted performance response, directly outputting the optimal design solution. This eliminates the traditional steps of repeated comparison and manual selection of optimal design variables, greatly improving design efficiency. Therefore, the technical solution provided by the present invention implements a one-step, end-to-end design that automatically outputs the "optimal design solution" based on the input of the "conditional performance response."

[0050] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved. This is not limited herein.

[0051] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. An intelligent generative design method for an anti-collision beam, characterized in that: include: S1: performing an experimental analysis on the anti-collision beam to obtain a data set, where the data set includes a cross-sectional image of the anti-collision beam and performance response data corresponding to the cross-sectional image; S2: Perform image recognition and text extraction on the cross-sectional images in the data set in step S1 to obtain the structural shape and structural size respectively; The structural shape and structural size are integrated to obtain multimodal design variables; S3: Using the multimodal design variables obtained in step S2 as input and the performance response data corresponding to the multimodal design variables as output, the constructed conditional variational network is trained to obtain a conditional variational model; S4: extracting multimodal design variables from the anti-collision beam to be designed according to steps S1 and S2, inputting the obtained multimodal design variables into the conditional variational model obtained in step S3 to obtain corresponding performance response data; S5: Filter out the optimal performance response data based on the multimodal design variables and performance response data generated in step S4, and derive the optimal design scheme of the anti-collision beam to be designed based on the optimal performance response data.

2. The intelligent generative design method for an anti-collision beam according to claim 1, characterized in that: In step S1, the anti-collision beam is experimentally analyzed, and the process of obtaining the data set includes: Establishing a simulation model of the anti-collision beam and obtaining a simulation data set; the simulation data set includes a cross-sectional image of the simulation model and its corresponding performance response; And performing actual collision tests on the anti-collision beam to obtain a real data set, wherein the real data set includes a cross-sectional image of the anti-collision beam and its corresponding performance response value.

3. The intelligent generative design method for an anti-collision beam according to claim 2, characterized in that: The process of establishing a simulation model of the anti-collision beam and obtaining a simulation data set includes: Establishing an implicit parameterized model including the structural shape and structural dimensions of the anti-collision beam; Determining sample points of different structural shapes and sizes in the anti-collision beam based on the Hammersley experimental design method; Acquiring a cross-sectional image of the sample point; Performing low-speed collision finite element simulation analysis on the sample points, and extracting performance response values ​​corresponding to each sample point from the simulation results; The simulation data set includes the cross-sectional image and the corresponding performance response value.

4. The intelligent generative design method for an anti-collision beam according to claim 3, characterized in that: In step S2, the process of extracting the structural shape and structural size of the cross-sectional image of the simulation data set includes: The structural dimensions and structural shape of the cross section of the anti-collision beam are read from the Hamsler experimental design list obtained by the Hamsler experimental design method.

5. The intelligent generative design method for an anti-collision beam according to claim 2, characterized in that: The process of conducting actual collision tests on anti-collision beams and obtaining a real data set includes: Conducting multiple collision tests on the anti-collision beam, extracting performance response values ​​measured in the experiments, and extracting a cross-sectional image of the anti-collision beam from the captured images of the experiments; The real data set includes cross-sectional images and corresponding performance response values.

6. The intelligent generative design method for an anti-collision beam according to claim 2 or 5, characterized in that: In step S2, the process of extracting the structural shape and structural size of the cross-sectional image of the real data set includes: Using a convolutional neural model to identify the cross-sectional images in the real data set to obtain the corresponding structural shape; After preprocessing the cross-sectional images in the real data set by edge detection and contour extraction, calibration scale and length and thickness parameter conversion are performed to obtain the corresponding structural dimensions.

7. The intelligent generative design method for an anti-collision beam according to claim 2, characterized in that: In step S2, the process of fusing the structural shape and structural dimensions to obtain multimodal design variables includes: Integrating the structural shapes of the cross-sectional images in the real data set and the structural shapes of the cross-sectional images in the simulation data set to obtain a structural shape set; Integrating the structural dimensions extracted from the real data set and the structural dimensions extracted from the simulation data set to obtain a structural dimension set; The structural shape set and the structural size set are input into the multimodal multi-layer fusion model for fusion to obtain the multimodal design variables.

8. The intelligent generative design method for an anti-collision beam according to claim 1, characterized in that: In the conditional variational network of step S3: Encoding the multimodal design variables using an encoder based on a multilayer perceptron, mapping the input multimodal design variables to a latent space, and obtaining a mean and a logarithmic variance of the latent space; reparameterizing and sampling the latent space based on the logarithmic variance and the mean to obtain latent variables; The latent variables and the set conditional performance responses are input into a decoder based on a mixture density network for decoding to obtain a mixed Gaussian parameter distribution; the mixed Gaussian parameter distribution is combined according to the weights to obtain the generative design variables; A Gaussian process regression model is used to perform performance prediction on the generated design variables to obtain corresponding performance response data.

9. The intelligent generative design method for an anti-collision beam according to claim 8, characterized in that: In step S3, the conditional variational network is trained by jointly optimizing the loss function; The joint optimization loss function is: ； Among them, L represents the joint optimization loss function, represents the reconstruction loss, represents the network parameters of the decoder, X represents the multimodal design variables, Y represents the corresponding performance response data, Z represents the latent variables, represents the KL divergence, represents the network parameters of the encoder, represents the negative feedback term, Represents the performance response data predicted by the conditional variation model obtained from the current training, Represents conditional performance response data, represents the KL divergence scaling coefficient, represents the negative feedback proportional coefficient; the negative feedback proportional coefficient is obtained by the following formula: ； in, It represents the error between the predicted performance response data and the corresponding conditional performance response data, represents the error variance.

10. The intelligent generative design method for an anti-collision beam according to claim 1, characterized in that: The process of step S5 includes: The performance response data obtained by S4 were analyzed using principal component analysis to obtain the objective weight value of each performance response; The expert scoring method and the hierarchical analysis method are used to obtain the subjective weight value of each performance response. The subjective and objective weight values ​​are solved through the zero-sum game method to obtain the comprehensive weight value; The grey relational analysis method is used to obtain the grey relational coefficient of each performance response scheme in the performance response data generated by S4, and the grey relational coefficient is multiplied by the comprehensive weight value to obtain the grey relational degree of each performance response scheme; All performance response schemes are sorted according to the gray correlation degree. The scheme with the largest gray correlation degree value is the optimal performance response scheme, and the corresponding relevant parameters of the anti-collision beam to be designed are the optimal design scheme of the anti-collision beam.

Images