Vehicle design scheme generation method and device, electronic equipment and storage medium
By introducing active learning and multi-objective optimization algorithms in vehicle design, the vehicle design scheme is automatically generated, and the problem of multiple and time-consuming manual intervention in the existing technology is solved, and efficient, general and flexible vehicle design optimization is achieved.
Patent Information
- Application Number
- CN202311206456.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-18
- Publication Date
- 2025-03-18
AI Technical Summary
The existing vehicle design optimization technology requires a lot of manual intervention, which is time-consuming, and lacks versatility and flexibility, making it difficult to obtain the optimal solution in a short period of time.
An intelligent vehicle design optimization system based on active learning is adopted, and multiple meta models for vehicle performance analysis and prediction are established by automatically generating training samples, and combining multi-objective optimization algorithms to automatically generate vehicle design solutions that meet user needs.
It reduces manual intervention, improves the generation efficiency of vehicle design schemes, and realizes multi-objective optimization of vehicle design, which is versatile and flexible.
Smart Images
Figure CN119646958A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of vehicle manufacturing, and in particular to a method, device, electronic device and storage medium for generating a vehicle design scheme. Background Art
[0002] Among the related technologies, vehicle optimization design refers to the technologies and methods that improve the performance, safety, fuel economy and environmental performance of vehicles by improving the design and manufacturing process of vehicles. With the rapid development of technologies such as artificial intelligence, big data, and cloud computing, the field of vehicle design optimization technology is also gradually becoming intelligent. By using intelligent algorithms, simulation technologies and other means, the vehicle design process can be intelligently optimized to improve design efficiency and product performance. The industry development in the field of vehicle design optimization technology is characterized by diversification, intelligence and greenness, providing strong support for the sustainable development of the vehicle industry. Summary of the invention
[0003] In order to overcome the problems existing in the related art, the present disclosure provides a method, device, electronic device and storage medium for generating a vehicle design scheme.
[0004] According to a first aspect of an embodiment of the present disclosure, a method for generating a vehicle design solution is provided, comprising:
[0005] Obtaining a design optimization objective and a corresponding first design constraint condition;
[0006] Determining a vehicle performance prediction metamodel that matches the design optimization goal, wherein the vehicle performance prediction metamodel is used to generate vehicle performance data according to the vehicle design solution;
[0007] Generate a first multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the first design constraint;
[0008] Based on the first multi-objective optimization workflow, a first vehicle design solution is generated.
[0009] Optionally, the vehicle performance prediction meta-model is generated by the following steps:
[0010] Acquiring a plurality of design variable data corresponding to the vehicle performance prediction meta-model;
[0011] Simulating and post-processing the plurality of design variable data based on a vehicle finite element model to generate a plurality of vehicle performance data;
[0012] Using the plurality of design variable data and the plurality of vehicle performance data as training samples, training an initial prediction meta-model;
[0013] When the prediction accuracy of the initial prediction meta-model is greater than a set accuracy threshold, the vehicle prediction meta-model is generated.
[0014] Optionally, the using the plurality of design variable data and the plurality of vehicle performance data as training samples to train an initial prediction meta-model comprises:
[0015] generating the training samples according to the plurality of design variable data and the plurality of vehicle performance data;
[0016] Determining the sample distribution of the training samples;
[0017] Filtering target training samples that meet the sample distribution condition from the training samples;
[0018] The initial prediction meta-model is trained based on the target training samples.
[0019] Optionally, the using the plurality of design variable data and the plurality of vehicle performance data as training samples to train an initial prediction meta-model comprises:
[0020] generating a plurality of sample feature vectors according to the plurality of design variable data and the plurality of vehicle performance data;
[0021] Filtering out a plurality of target sample feature vectors that match a preset feature vector from the plurality of sample feature vectors;
[0022] The design variable data and vehicle performance data corresponding to the plurality of target sample feature vectors are used as training samples to train the initial prediction meta-model.
[0023] Optionally, the simulating and post-processing the plurality of design variable data based on the vehicle finite element model to generate a plurality of vehicle performance data includes:
[0024] Simulating the plurality of design variable data according to the vehicle finite element model to generate a plurality of simulation results;
[0025] The plurality of simulation results are analyzed based on a preset evaluation algorithm to generate the plurality of vehicle performance data.
[0026] Optionally, generating a first multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the first design constraint condition includes:
[0027] Based on the vehicle performance prediction meta-model, generating a performance prediction function;
[0028] Generate a multi-objective optimization condition according to the performance prediction function, the design optimization goal and the first design constraint condition;
[0029] According to the multi-objective optimization conditions, the first multi-objective optimization workflow is generated by the multi-objective optimization algorithm.
[0030] Optionally, the method comprises:
[0031] Verifying the first vehicle design solution according to the vehicle performance prediction meta-model to generate first vehicle performance data;
[0032] generating a second design constraint condition based on the first vehicle performance data when the first vehicle performance data does not meet the set performance requirement;
[0033] generating a second multi-objective optimization workflow based on the multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the second design constraint;
[0034] Based on the second multi-objective optimization workflow, a second vehicle design solution is generated.
[0035] According to a second aspect of an embodiment of the present disclosure, a device for generating a vehicle design scheme is provided, comprising:
[0036] An acquisition module is configured to acquire a design optimization target and a corresponding first design constraint condition;
[0037] a determination module configured to determine a vehicle performance prediction metamodel that matches the design optimization target, wherein the vehicle performance prediction metamodel is used to generate vehicle performance data according to a vehicle design solution;
[0038] A generating module configured to generate a first multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the first design constraint;
[0039] An execution module is configured to generate a first vehicle design solution based on the first multi-objective optimization workflow.
[0040] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, including:
[0041] a memory having a computer program stored thereon;
[0042] A processor is used to execute the computer program in the memory to implement the steps of the method for generating a vehicle design solution as described in any one of the first aspects of the present disclosure.
[0043] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which computer program instructions are stored. When the program instructions are executed by a processor, the steps of the method for generating a vehicle design solution provided in the first aspect of the present disclosure are implemented.
[0044] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects:
[0045] Through the above method, the design optimization target and the corresponding first design constraint are obtained, and the vehicle performance prediction metamodel matching the design optimization target is determined, wherein the vehicle performance prediction metamodel is used to generate vehicle performance data according to the vehicle design scheme, and the first multi-objective optimization workflow is generated based on the multi-objective optimization algorithm according to the vehicle performance prediction metamodel, the design optimization target and the first design constraint, and the first vehicle design scheme is generated based on the first multi-objective optimization workflow. Thus, based on the vehicle performance prediction metamodel, the design optimization target is optimized according to the first design constraint, and a vehicle design scheme that meets user needs is automatically generated. In the case of avoiding manual intervention, multi-objective optimization of vehicle design is intelligently performed, thereby improving the efficiency of generating vehicle design schemes.
[0046] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0048] Figure 1 The present invention is a flowchart of a method for generating a vehicle design scheme according to an exemplary embodiment.
[0049] Figure 2 The present invention is a flowchart of a method for generating a vehicle performance prediction meta-model according to an exemplary embodiment.
[0050] Figure 3 It is a schematic diagram of another method for generating a vehicle design solution according to an exemplary embodiment.
[0051] Figure 4 It is a block diagram of a device for generating a vehicle design solution according to an exemplary embodiment.
[0052] Figure 5 It is a block diagram of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION
[0053] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0054] It should be noted that all actions of acquiring signals, information or data in this application are carried out in compliance with the relevant data protection laws and policies of the country where they are located, and with the authorization given by the owner of the corresponding device.
[0055] Vehicle design is a complex process that requires consideration of multiple factors, such as performance, safety, comfort, economy, etc. The relevant vehicle design process usually requires a lot of manpower and time, and there is often a certain degree of human subjectivity and trial and error costs. Therefore, how to improve the efficiency and accuracy of vehicle design is an important issue. In the related technology, DoE (Design Of Experiments) experimental design is used to change the values of different design variables, and simulate and analyze the vehicle design to find a better design solution. However, this solution can only simulate the vehicle design, and cannot obtain the global optimal solution for the vehicle design.
[0056] With the development of artificial intelligence technology, automobile design optimization methods based on machine learning have gradually attracted attention. The vehicle design optimization scheme based on machine learning accumulates data through simulation, and then optimizes by establishing a predictive model through machine learning technology. However, this design scheme still requires a lot of manual intervention, requiring manual data sampling based on experimental design methods, manual calling of simulation software for simulation, manual post-processing to generate samples, and then training through machine learning models selected by engineering experience. After inputting relevant constraints, multi-objective optimization is performed to obtain the optimization scheme. Another optimization scheme is to combine the simulation process with the optimization algorithm, set the optimization goal through the optimization algorithm, use simulation to obtain various performance indicators, and iteratively select the optimal solution.
[0057] However, existing design solutions often require manual intervention and design experience, and require manual simulation of a large number of samples, which is very time-consuming. For machine learning solutions, whether it is the selection of predictive model algorithms or the selection and design of optimization algorithms, engineers' experience is required, and most processes cannot be automated. The entire workflow requires a lot of manual intervention to run normally. For simulation + optimization solutions, since a single simulation takes a long time and requires high computing power, it lacks flexibility and takes a long time. It is impossible to get the optimal solution in a short time. At the same time, once the constraints change, the new optimal result cannot be obtained immediately, and the time-consuming optimization needs to be restarted. In addition, existing design solutions are often designed for a specific component and lack versatility and generalization.
[0058] In view of this, this proposal designs a vehicle design intelligent optimization system based on active learning, aiming to introduce active learning workflow, reduce manual intervention, train a high-precision vehicle performance prediction model with very few samples, and then combine it with a multi-objective optimization algorithm to build an automated workflow for intelligent optimization of vehicle design. This vehicle design intelligent optimization system based on active learning can solve the problems that the current existing solutions are time-consuming, require a lot of manual intervention, and are not general and flexible enough. In this system, users only need to provide the vehicle finite element simulation model and related design variables and corresponding analysis items. The system can automatically generate training samples based on the active learning algorithm and establish multiple meta-models for vehicle performance analysis and prediction. Manual intervention is effectively reduced. Based on this meta-model, according to the optimization objectives and constraints provided by the user, multi-objective algorithms such as genetic algorithms are used for optimization to obtain multiple sets of optimal solutions.
[0059] After establishing a vehicle finite element simulation model and abstracting various design variables and possible analysis items from it, the system can be used directly to quickly establish a meta-model for vehicle performance prediction, which does not require human intervention. Combined with the meta-model of vehicle performance prediction, after the user inputs possible optimization goals and constraints, the system uses a multi-objective optimization algorithm to integrate multiple optimization goals and constraints, and automatically generates a feasible optimization plan with universality and flexibility.
[0060] Figure 1 is a flow chart of a method for generating a vehicle design solution according to an exemplary embodiment. Figure 1 As shown, the method is used in a vehicle design terminal and includes the following steps.
[0061] In step S11, a design optimization target and a corresponding first design constraint condition are obtained.
[0062] For example, this embodiment is applied to a vehicle design terminal. Usually, the manufacturing design of a vehicle is to first design each component of the vehicle, generate a component design scheme corresponding to each component, and then splice the component design schemes to form the overall design scheme of the vehicle. Therefore, the vehicle design scheme generated in this embodiment is a vehicle design scheme corresponding to one or more vehicle components. Different vehicle performance prediction metamodels are loaded in the vehicle design terminal for different vehicle components. Relevant staff inputs the design optimization target to be designed in the vehicle design terminal, so that the vehicle design terminal selects a matching vehicle performance prediction metamodel from the database based on the design optimization target. For example, when the design optimization target is the vehicle shell, the relevant staff inputs the vehicle shell option into the vehicle design terminal, and the vehicle design terminal selects a vehicle performance prediction metamodel that matches the vehicle shell from the database.
[0063] In the vehicle design process, there are different design constraints for different design optimization targets. For example, when the design optimization target is vehicle body stiffness, based on the design requirements of vehicle body stiffness, the relevant staff will determine the design constraints corresponding to the vehicle body stiffness, that is, the maximum value limit of the vehicle body material layer thickness, the maximum weight limit of the vehicle body, the vehicle body modal requirement limit, etc. After determining the design optimization target, the relevant staff inputs the first design constraint corresponding to the design optimization target into the vehicle design terminal based on the current design optimization requirements. For example, the relevant staff can set multiple design constraints corresponding to the design optimization target based on the design optimization requirements. The design constraints are used to constrain the vehicle design scheme of the design optimization target so that the vehicle design terminal generates a vehicle design scheme that meets the requirements.
[0064] In step S12, a vehicle performance prediction metamodel matching the design optimization target is determined, wherein the vehicle performance prediction metamodel is used to generate vehicle performance data according to the vehicle design solution.
[0065] For example, in the vehicle design terminal of this embodiment, multiple vehicle performance prediction metamodels are configured based on different types of design optimization targets. The vehicle performance prediction metamodel can be used to predict the performance of the generated vehicle design scheme, and then generate vehicle performance data for evaluating the performance of the vehicle design scheme. In this embodiment, the vehicle performance prediction metamodel can simulate the vehicle design scheme corresponding to the design optimization target, and post-process the simulation results, and evaluate the simulation post-processing results of the vehicle design scheme according to the preset vehicle design scheme evaluation standard of the design optimization target, thereby generating vehicle performance data of the design optimization target.
[0066] The design optimization target input by the relevant staff is identified, and a vehicle performance prediction metamodel matching the design optimization target is determined from the database of the vehicle design terminal, wherein the vehicle performance prediction metamodel is configured with corresponding design variable combinations and simulation modules. Based on the design variable combinations and simulation modules, the vehicle design scheme of the design optimization target is evaluated to generate corresponding vehicle performance data.
[0067] In step S13, a first multi-objective optimization workflow is generated based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the first design constraint.
[0068] For example, in this embodiment, based on the vehicle performance prediction metamodel, the vehicle design terminal can generate a vehicle design solution that meets the constraint conditions according to the design optimization goal and the first design constraint condition. For the design optimization goal provided by the user, combined with the corresponding vehicle performance prediction metamodel, the vehicle design terminal will automatically generate a multi-objective optimization workflow based on the multi-objective optimization algorithm.
[0069] Optionally, in one implementation, the above step S13 includes:
[0070] Generate a performance prediction function based on the vehicle performance prediction meta-model;
[0071] Generate a multi-objective optimization condition according to the performance prediction function, the design optimization goal and the first design constraint condition;
[0072] According to the multi-objective optimization conditions, a first multi-objective optimization workflow is generated by a multi-objective optimization algorithm.
[0073] For example, in this embodiment, based on the vehicle performance prediction metamodel, a corresponding performance prediction function is generated, and multi-objective optimization algorithms such as NSGA-II (Non-dominated Sorting Genetic Algorithm II) are used. In combination with the performance prediction function corresponding to the vehicle performance prediction metamodel, the multi-objective optimization results are quickly output, and multiple groups of first multi-objective optimization workflows that meet the design objectives and constraints are generated.
[0074] In step S14, a first vehicle design solution is generated based on the first multi-objective optimization workflow.
[0075] For example, based on the first multi-objective optimization workflow generated in the above steps, a first vehicle design solution corresponding to the design optimization objective is generated.
[0076] In one embodiment, after the above step S14, the method further includes:
[0077] Verifying the first vehicle design solution according to the vehicle performance prediction meta-model to generate first vehicle performance data;
[0078] generating a second design constraint condition based on the first vehicle performance data when the first vehicle performance data does not meet the set performance requirement;
[0079] Generate a second multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal, and the second design constraint;
[0080] Based on the second multi-objective optimization workflow, a second vehicle design solution is generated.
[0081] For example, in this embodiment, after the vehicle design scheme corresponding to the design optimization target is generated through the above steps, the first vehicle design scheme is verified according to the vehicle performance prediction metamodel, and the first vehicle performance data is generated. The first vehicle performance data is verified. When the first vehicle performance data does not meet the set performance requirements, the first constraint condition is adjusted based on the first vehicle performance data to generate the second design constraint condition. Then, based on the adjusted second design constraint condition, the vehicle performance prediction metamodel and the design optimization target, the second vehicle design scheme is generated.
[0082] In this embodiment, the generated vehicle design scheme is verified based on the vehicle performance prediction metamodel, and whether the vehicle design scheme meets the set performance requirements is determined based on the verification results. When the vehicle design scheme does not meet the set performance requirements, the constraints are adjusted and the vehicle design scheme is continuously iterated to finally generate a vehicle design scheme that meets the set performance requirements.
[0083] Through the above method, the design optimization target and the corresponding first design constraint are obtained, and the vehicle performance prediction metamodel matching the design optimization target is determined, wherein the vehicle performance prediction metamodel is used to generate vehicle performance data according to the vehicle design scheme, and the first multi-objective optimization workflow is generated based on the multi-objective optimization algorithm according to the vehicle performance prediction metamodel, the design optimization target and the first design constraint, and the first vehicle design scheme is generated based on the first multi-objective optimization workflow. Thus, based on the vehicle performance prediction metamodel, the design optimization target is optimized according to the first design constraint, and a vehicle design scheme that meets user needs is automatically generated. In the case of avoiding manual intervention, multi-objective optimization of vehicle design is intelligently performed, thereby improving the efficiency of generating vehicle design schemes.
[0084] Figure 2 is a flow chart of a method for generating a vehicle performance prediction meta-model according to an exemplary embodiment. Figure 2 As shown, the method is applied in a vehicle design terminal and includes the following steps.
[0085] In step S21, a plurality of design variable data corresponding to the vehicle performance prediction meta-model is obtained.
[0086] For example, in this embodiment, for the same design optimization target, there are multiple design variable data corresponding to it, and the corresponding multiple design variable data can be determined based on the design common sense of the design optimization target. For example, when the design optimization target is the vehicle body, there are some fixed and unified standards corresponding to the vehicle body design. In addition to the fixed and unified standards, the multiple design variable data of the vehicle body can be the vehicle body weight, vehicle body curvature, vehicle body material, vehicle body width, etc. Based on the current design requirements, the relevant staff inputs the multiple design variable data corresponding to the design optimization target into the corresponding vehicle performance prediction meta-model.
[0087] For example, in this embodiment, design variable data of a design optimization target in a vehicle design process is collected to generate a design variable sample pool of the design optimization target, and multiple design variable data are sampled from the design variable sample pool based on random sampling rules to form multiple design variable groups, each of which includes multiple different design variable data.
[0088] In step S22 , a plurality of design variable data are simulated and post-processed based on the vehicle finite element model to generate a plurality of vehicle performance data.
[0089] According to the vehicle wired element model, multiple design variable data are simulated, and the simulation results are post-processed to generate vehicle performance data corresponding to each set of design variables. The vehicle wired element model is the basis for vehicle design analysis. According to the specific design problems and design conditions of the design optimization target, the appropriate wired element unit is selected, the design optimization target is discretized, and the vehicle wired element model is assigned corresponding attributes. After simulating the boundary conditions of the design optimization target, the element model is debugged to generate the corresponding vehicle finite element model. The vehicle finite element model is a group of unit combinations connected at the nodes, relying on the node force, and constrained at the nodes. It is used to despise the movement and behavior of the vehicle, and analyze and control the design results based on some design condition assumptions and simplified design conditions.
[0090] For example, the vehicle finite element model in this embodiment may include: (1) vehicle dynamics model: the core part used to describe the vehicle motion based on the combination of design variables, including parameters such as acceleration, speed and position. Common vehicle dynamics models include single-axis model, dual-axis model and multi-body model; (2) vehicle control model: the part used to describe the vehicle control behavior based on the combination of design variables, including steering, acceleration and braking. Common control models include PID (Proportional Integral Derivative) controller, state space controller and fuzzy controller; (3) vehicle environment model: the part used to describe the vehicle surrounding environment based on design variables, including roads, obstacles and other vehicles. Common environmental models include grid maps, laser radar and camera images; (4) vehicle perception model: the part used to describe the vehicle perception ability based on design variables, including obstacle recognition, traffic signal detection and lane line judgment. Common perception models include convolutional neural networks, probabilistic graphical models and image processing algorithms.
[0091] In this embodiment, multiple design variable data are understood based on the vehicle finite element model, and the behavior of the vehicle is predicted, so as to perform tasks such as vehicle control, path planning and decision-making. Multiple design variable data are simulated and post-processed based on the vehicle finite element model to generate multiple vehicle performance data corresponding to each design variable data group.
[0092] Optionally, in one implementation, the above step S22 includes:
[0093] Simulating multiple design variable data according to the vehicle finite element model to generate multiple simulation results;
[0094] Based on a preset evaluation algorithm, multiple simulation results are analyzed to generate multiple vehicle performance data.
[0095] For example, in this embodiment, the multiple design variable data sampled and generated in the above steps are simulated and verified based on the vehicle finite element model to generate multiple corresponding simulation results, and then the multiple simulation results are analyzed based on a preset evaluation algorithm to generate multiple vehicle performance data corresponding to the design optimization target.
[0096] In step S23, a plurality of design variable data and a plurality of vehicle performance data are used as training samples to train an initial prediction meta-model.
[0097] For example, in this embodiment, the initial prediction metamodel is trained based on multiple design variable data and multiple vehicle performance data. Under this active learning workflow, the initial prediction metamodel can be trained through samples generated by the above-mentioned multiple design variable data without human intervention to generate a vehicle prediction metamodel for vehicle performance prediction.
[0098] Optionally, in one implementation, the above step S23 includes:
[0099] Generate training samples according to multiple design variable data and multiple vehicle performance data;
[0100] Determine the sample distribution of training samples;
[0101] Filter target training samples that meet sample distribution conditions from training samples;
[0102] Based on the target training samples, an initial prediction meta-model is trained.
[0103] In this embodiment, the multiple design variable data and multiple vehicle performance data generated in the above steps are used as training samples to train the initial prediction meta-model. Based on the randomness of the generation of training samples, there are abnormal samples that affect the training results. Therefore, in this embodiment, the training samples need to be screened to improve the accuracy of the vehicle prediction meta-model finally generated. For example, under normal circumstances, the sample distribution corresponding to normal training samples has a certain distribution law. Therefore, according to the data distribution of the training samples, the sample distribution of the training samples is determined, and the training samples are screened according to the sample distribution, and the target training samples that meet the sample distribution are filtered out, and then the initial prediction meta-model is trained based on the target training samples.
[0104] For example, in this embodiment, the design optimization target is determined to be the body stiffness value. Usually, the body stiffness value conforms to the normal distribution. The mean variance of the stiffness in the sample is found according to the body stiffness value, and the sample data that does not obey the normal distribution is filtered out according to the mean variance. Based on the filtered training samples, the initial prediction meta-model is trained.
[0105] Optionally, in another implementation, the above step S23 includes:
[0106] Generate multiple sample feature vectors according to multiple design variable data and multiple vehicle performance data;
[0107] Filtering out multiple target sample feature vectors that match the preset feature vector from multiple sample feature vectors;
[0108] The design variable data and vehicle performance data corresponding to the feature vectors of multiple target samples are used as training samples to train the initial prediction meta-model.
[0109] In this embodiment, sample feature vectors can be used to screen training samples. For example, multiple corresponding sample feature vectors are generated according to multiple design variable data and multiple vehicle performance data generated in the above steps. The database corresponding to the vehicle design terminal includes preset feature vectors generated based on standard variable data and standard vehicle performance data, and each preset feature vector corresponds to a design optimization target. After the design optimization target is determined, a preset feature vector matching the design optimization target is determined from the database, and multiple sample feature vectors are screened based on the preset feature vector to determine multiple target sample feature vectors matching the preset feature vector. The design variables and vehicle performance data corresponding to the multiple target sample feature vectors are used as training samples to train the initial prediction metamodel.
[0110] For example, the vehicle design terminal performs Latin hypercube sampling (a method of approximating random sampling from a multivariate parameter distribution) on the design variables to be optimized to generate 300 groups of samples, and generates a corresponding parameter list. Based on the data in the parameter list, a new finite element file is automatically generated, and simulation software such as Nastran (Simulation Structure) is called for simulation to obtain the eigenvector corresponding to each mode. For the simulation results of the new finite element file, the similarity between the eigenvector of each order and the eigenvector corresponding to the baseline mode is calculated, and the most similar group is found as the mode corresponding to the design variable, and the initial prediction metamodel is trained based on the mode.
[0111] In step S24 , when the prediction accuracy of the initial prediction meta-model is greater than the set accuracy threshold, a vehicle prediction meta-model is generated.
[0112] For example, in the above training process, the prediction accuracy of the prediction metamodel generated by each training is monitored. When the prediction accuracy of the initial prediction metamodel is greater than the set accuracy expectation, the corresponding initial prediction metamodel converges, and the converged initial prediction metamodel is used as the vehicle prediction metamodel.
[0113] In the above method, multiple design variable combinations are input into the vehicle finite element model to generate corresponding simulation results, and then the design variable combinations and corresponding simulation results are used as training samples to train the initial prediction meta-model to generate a vehicle prediction meta-model for vehicle performance prediction. Thus, based on the active learning model training mechanism, a high-precision vehicle performance prediction meta-model is established on the basis of avoiding manual intervention.
[0114] Figure 3 is a schematic diagram of another method for generating a vehicle design solution according to an exemplary embodiment. Figure 3 As shown, the method includes the following process.
[0115] 1. Establishment and evaluation of performance prediction metamodel based on active learning.
[0116] Based on active learning technology, after obtaining the vehicle finite element model, the system automatically samples from the database to generate design variable combinations; automatically calls simulation software to simulate the design variable combinations, and performs post-processing to generate simulation results; further samples are taken from the sample pool to find reliable samples that contribute more to the results based on preset screening conditions; on this basis, algorithms such as automatic machine learning are applied to establish the meta-model, and the model results are evaluated on the validation set. Under this active learning workflow, a vehicle performance prediction meta-model can be established without human intervention and only a small number of samples are needed to achieve high prediction accuracy.
[0117] 2. Intelligent generation of multi-objective optimization solutions based on meta-model.
[0118] Based on the vehicle performance prediction metamodel, the system can intelligently generate a vehicle design solution that meets the constraints entered by the user. Based on the optimization goals provided by the user and the corresponding vehicle performance prediction metamodel, the system automatically establishes a multi-objective optimization workflow, and uses multi-objective optimization algorithms such as NSGAII, combined with the vehicle performance prediction metamodel to quickly output multi-objective optimization results, and generate multiple sets of design variable combinations that meet the design goals and constraints and return them to the user.
[0119] For example, a case of multi-objective material thickness design optimization for body-in-white stiffness and modal analysis (which can be extended to parameterizable design variables of each component):
[0120] 1. The user provides the finite element model of the body in white, the design variables to be optimized (sheet metal ID and its material thickness), the maximum and minimum limits of the material thickness, the target points for post-processing calculations, the stiffness calculation formula, and the baseline modal order.
[0121] 2. Active learning sample pool generation and sample labeling;
[0122] (1) The system performs Latin hypercube sampling (a method of approximating random sampling from a multivariate parameter distribution) on 300 sets of design variables to be optimized, generates a parameter list, automatically generates a new finite element file, and calls Nastran (Structure Simulation) and other simulation software for simulation;
[0123] (2) For stiffness, the system reads the result file after simulation and automatically extracts the stiffness value according to the stiffness calculation formula.
[0124] (3) For the mode (resonance frequency), the system first simulates based on the baseline model to obtain the eigenvector corresponding to each mode. For the simulation results of the new finite element file, the similarity between the eigenvector of each order and the eigenvector corresponding to the baseline mode is calculated, and the most similar group is found as the mode corresponding to the design variable.
[0125] (4) Aggregate data to generate a training sample pool.
[0126] 3. Further sampling:
[0127] (1) Based on the sample distribution in the sample pool, filter out samples with abnormal distribution. For example, assuming that the obtained stiffness value conforms to the normal distribution, find the mean variance of the stiffness in the sample, and filter out sample points that do not conform to the normal distribution based on the mean variance;
[0128] (2) Based on the filtered samples, train the model using a machine learning algorithm (neural networks, automatic machine learning, etc. can be used);
[0129] (3) Make predictions on the validation set and stop if the accuracy is met. If the accuracy is not met, further screening and sampling can be performed to incrementally train the model;
[0130] (4) Obtain a meta-model for stiffness and modal performance prediction, and construct a performance prediction function based on the meta-model (weight prediction can be directly obtained from the sheet metal part thickness ratio and the weight of each part in the baseline model);
[0131] (5) The user selects the optimization goal, such as weight reduction of xxkg, stiffness greater than xx, modality greater than xx, etc., automatically builds a multi-objective optimization workflow, calls multi-objective optimization algorithms such as NSGAII, and automatically generates multiple sets of material thickness combinations;
[0132] (6) The user further simulates and verifies the generated solution, modifies the constraints, and continuously iterates to generate a vehicle design solution.
[0133] Through the above method, a system is provided that can intelligently perform multi-objective optimization of automobile design without too much human intervention. Users only need to upload the baseline model of the whole vehicle or parts and the parameters to be optimized, and the system can establish accurate automobile performance prediction meta-models for different analysis items. Users only need to customize the optimization goals and constraints to quickly obtain multiple groups of optimization solutions. In the process of establishing the automobile performance prediction meta-model, an active learning method is adopted to automatically select a part of the optimal samples from the sample pool, and automatic simulation and post-processing are performed based on the automobile finite element model. After the performance indicators are obtained, they are added to the training samples for centralized training. It is possible to automatically train and evaluate the model with a very small number of samples without human intervention to generate a meta-model for performance prediction.
[0134] Figure 4 is a block diagram of a vehicle design scheme generation device according to an exemplary embodiment, referring to Figure 4 The device 100 includes: an acquisition module 110, a determination module 120, a generation module 130 and an execution module 140.
[0135] The acquisition module 110 is configured to acquire a design optimization target and a corresponding first design constraint condition;
[0136] The determination module 120 is configured to determine a vehicle performance prediction metamodel that matches the design optimization target, wherein the vehicle performance prediction metamodel is used to generate vehicle performance data according to the vehicle design solution;
[0137] The generating module 130 is configured to generate a first multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the first design constraint;
[0138] The execution module 140 is configured to generate a first vehicle design solution based on a first multi-objective optimization workflow.
[0139] Optionally, the apparatus 100 further includes a training module, the training module including:
[0140] An acquisition submodule configured to acquire a plurality of design variable data corresponding to a vehicle performance prediction meta-model;
[0141] A first generating submodule is configured to simulate and post-process a plurality of design variable data based on a vehicle finite element model to generate a plurality of vehicle performance data;
[0142] A training submodule is configured to train an initial prediction meta-model using a plurality of design variable data and a plurality of vehicle performance data as training samples;
[0143] The second generation submodule is configured to generate a vehicle prediction metamodel when the prediction accuracy of the initial prediction metamodel is greater than a set accuracy threshold.
[0144] Optionally, the training submodule is configured as follows:
[0145] Generate training samples according to multiple design variable data and multiple vehicle performance data;
[0146] Determine the sample distribution of training samples;
[0147] Filter target training samples that meet sample distribution conditions from training samples;
[0148] Based on the target training samples, an initial prediction meta-model is trained.
[0149] Optionally, the training submodule is configured as follows:
[0150] Generate multiple sample feature vectors according to multiple design variable data and multiple vehicle performance data;
[0151] Filtering out multiple target sample feature vectors that match the preset feature vector from multiple sample feature vectors;
[0152] The design variable data and vehicle performance data corresponding to the feature vectors of multiple target samples are used as training samples to train the initial prediction meta-model.
[0153] Optionally, the first generating submodule is configured as follows:
[0154] Simulating multiple design variable data according to the vehicle finite element model to generate multiple simulation results;
[0155] Based on a preset evaluation algorithm, multiple simulation results are analyzed to generate multiple vehicle performance data.
[0156] Optionally, the generation module is configured to:
[0157] Generate a performance prediction function based on the vehicle performance prediction meta-model;
[0158] Generate a multi-objective optimization condition according to the performance prediction function, the design optimization goal and the first design constraint condition;
[0159] According to the multi-objective optimization conditions, a first multi-objective optimization workflow is generated by a multi-objective optimization algorithm.
[0160] Optionally, the apparatus 100 includes an iteration module, and the iteration module is configured to:
[0161] Verifying the first vehicle design solution according to the vehicle performance prediction meta-model to generate first vehicle performance data;
[0162] generating a second design constraint condition based on the first vehicle performance data when the first vehicle performance data does not meet the set performance requirement;
[0163] Generate a second multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal, and the second design constraint;
[0164] Based on the second multi-objective optimization workflow, a second vehicle design solution is generated.
[0165] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0166] The present disclosure also provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the steps of the method for generating a vehicle design solution provided by the present disclosure.
[0167] Figure 55 is a block diagram of an electronic device according to an exemplary embodiment. For example, device 500 may be provided as a server. Figure 5 The device 500 includes a processing component 522, which further includes one or more processors, and a memory resource represented by a memory 532 for storing instructions executable by the processing component 522, such as an application. The application stored in the memory 532 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 522 is configured to execute instructions to perform the above-mentioned method for generating a vehicle design solution.
[0168] Device 500 may also include a power component 526 configured to perform power management of device 500, a wired or wireless network interface 550 configured to connect device 500 to a network, and an input / output interface 558. Device 500 may operate based on an operating system stored in memory 532.
[0169] In another exemplary embodiment, a computer program product is also provided, which contains a computer program executable by a programmable device and has code parts for executing the above-mentioned method for generating a vehicle design solution when the computer program is executed by the programmable device.
[0170] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the present disclosure. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0171] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A method for generating a vehicle design scheme, characterized in that: include: Obtaining a design optimization objective and a corresponding first design constraint condition; Determining a vehicle performance prediction metamodel that matches the design optimization goal, wherein the vehicle performance prediction metamodel is used to generate vehicle performance data according to the vehicle design solution; Generate a first multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the first design constraint; Based on the first multi-objective optimization workflow, a first vehicle design solution is generated.
2. The generation method according to claim 1, characterized in that: The vehicle performance prediction meta-model is generated by the following steps: Acquiring a plurality of design variable data corresponding to the vehicle performance prediction meta-model; Simulating and post-processing the plurality of design variable data based on a vehicle finite element model to generate a plurality of vehicle performance data; Using the plurality of design variable data and the plurality of vehicle performance data as training samples, training an initial prediction meta-model; When the prediction accuracy of the initial prediction meta-model is greater than a set accuracy threshold, the vehicle prediction meta-model is generated.
3. The generation method according to claim 2, characterized in that: The method of using the plurality of design variable data and the plurality of vehicle performance data as training samples to train an initial prediction meta-model comprises: generating the training samples according to the plurality of design variable data and the plurality of vehicle performance data; Determining the sample distribution of the training samples; Filtering target training samples that meet the sample distribution condition from the training samples; The initial prediction meta-model is trained based on the target training samples.
4. The generation method according to claim 2, characterized in that: The method of using the plurality of design variable data and the plurality of vehicle performance data as training samples to train an initial prediction meta-model comprises: generating a plurality of sample feature vectors according to the plurality of design variable data and the plurality of vehicle performance data; Filtering out a plurality of target sample feature vectors that match a preset feature vector from the plurality of sample feature vectors; The design variable data and vehicle performance data corresponding to the plurality of target sample feature vectors are used as training samples to train the initial prediction meta-model.
5. The generation method according to claim 2, characterized in that: The simulating and post-processing the plurality of design variable data based on the vehicle finite element model to generate a plurality of vehicle performance data includes: Simulating the plurality of design variable data according to the vehicle finite element model to generate a plurality of simulation results; The plurality of simulation results are analyzed based on a preset evaluation algorithm to generate the plurality of vehicle performance data.
6. The generation method according to claim 1, characterized in that: The step of generating a first multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the first design constraint condition comprises: Based on the vehicle performance prediction meta-model, generating a performance prediction function; Generate a multi-objective optimization condition according to the performance prediction function, the design optimization goal and the first design constraint condition; According to the multi-objective optimization conditions, the first multi-objective optimization workflow is generated by the multi-objective optimization algorithm.
7. The generation method according to claim 1, characterized in that: The method comprises: Verifying the first vehicle design solution according to the vehicle performance prediction meta-model to generate first vehicle performance data; generating a second design constraint condition based on the first vehicle performance data when the first vehicle performance data does not meet the set performance requirement; generating a second multi-objective optimization workflow based on the multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the second design constraint; Based on the second multi-objective optimization workflow, a second vehicle design solution is generated.
8. A device for generating a vehicle design scheme, characterized in that: include: An acquisition module is configured to acquire a design optimization target and a corresponding first design constraint condition; a determination module configured to determine a vehicle performance prediction metamodel that matches the design optimization target, wherein the vehicle performance prediction metamodel is used to generate vehicle performance data according to a vehicle design solution; A generating module configured to generate a first multi-objective optimization workflow based on a multi-objective optimization algorithm according to the vehicle performance prediction meta-model, the design optimization goal and the first design constraint; An execution module is configured to generate a first vehicle design solution based on the first multi-objective optimization workflow.
9. An electronic device, characterized in that: include: a memory having a computer program stored thereon; A processor, configured to execute the computer program in the memory to implement the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the program instructions are executed by a processor, the steps of the method described in any one of claims 1 to 7 are implemented.