RMNN-based front subframe lightweight design method, device, medium and program product

Abstract

The application discloses a front subframe lightweight design method and device based on RMNN, a medium and a program product, and relates to the technical field of lightweight design of a front subframe.The method comprises the following steps: (1) constructing three simulation models of weight, stress and modal of the front subframe, and deducing a lightweight design model containing stress and modal constraints; (2) generating a population based on uniform experimental design and correlation criteria, simulating and evaluating the population, and constructing a database; (3) constructing a network architecture and training the RMNN; (4) generating a candidate frame set according to a random selection strategy and the RMNN; (5) establishing a radial basis function model, screening an optimal frame, simulating and evaluating the optimal frame, updating the database and the population, returning to step (3) until an optimization target is up to standard, and outputting an optimal solution of optimization parameters. The application can quickly search a feasible region through the RMNN, further combines the radial basis function model to overcome the problems of low search efficiency of a high-dimensional design space and difficulty in identifying the feasible region under multiple constraints, and can quickly optimize the lightweight design of the front subframe.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and swarm intelligence, and more specifically, to a method, apparatus, medium, and program product for lightweight design of a front subframe based on an RMNN (Reverse Mapping Neural Network). Background Technology

[0002] The front subframe, as a key load-bearing component in the front structure of a vehicle, is primarily used to mount the engine, steering system, and suspension mechanism. It bears and transmits the complex loads generated during vehicle operation, directly affecting the vehicle's safety, handling stability, and noise and vibration comfort. In the context of the automotive industry's continuous pursuit of high performance and efficiency, lightweighting has become a crucial goal in front subframe structural design. Reducing weight can effectively lower the overall vehicle mass, thereby improving fuel economy and dynamic response characteristics. However, lightweight design must be carried out while meeting strength and vibration performance constraints: on the one hand, stress analysis must be used to ensure that the stress level of components under extreme conditions remains within the material's allowable range to avoid plastic deformation or fatigue failure; on the other hand, modal analysis must be used to control its natural frequencies and mode shapes, avoiding the main excitation frequencies to prevent resonance and ensuring vehicle ride comfort and structural reliability.

[0003] However, traditional single-objective optimization methods widely used in current engineering practice typically focus on improving a specific performance index and often assume unconstrained conditions. For example, early methods focused solely on structural lightweighting while ignoring practical limitations such as strength or vibration frequency. These methods heavily rely on expert prior knowledge and extensive physical experiments, consuming significant computational resources and having long iteration cycles. When faced with complex design structures and material properties resulting in high-dimensional design spaces and constraints, their search efficiency and convergence accuracy often fail to meet practical needs. Against this backdrop, novel optimization methods based on neural networks are gaining attention. These methods can extract complex implicit relationships between performance indices and design variables from existing design data. Leveraging their powerful nonlinear fitting and generalization capabilities, they significantly accelerate the exploration of the design space, enabling the rapid generation of high-performance designs with lower computational costs. Nevertheless, many existing engineering optimization methods still do not fully integrate the modeling and search capabilities of neural networks, continuing to rely on traditional simulation iteration modes. This results in long optimization cycles, low resource utilization, and difficulty in meeting the demands of modern industry for efficient optimization and rapid response. Summary of the Invention

[0004] To address the challenges of locating the feasible region and achieving low optimization efficiency under high-dimensional constraints in existing lightweight front subframe designs, this invention proposes a lightweight front subframe design method, device, medium, and program product based on RMNN.

[0005] In a first aspect, the present invention provides a lightweight design method for a front subframe based on RMNN, the method comprising the following steps:

[0006] (1) The length, width and thickness of each crossbeam of the former subframe are used as design variables. The model is created in the three-dimensional modeling software. The static load is combined with the static analysis to obtain the weight and stress simulation model. The modal simulation model is constructed in HyperMesh and the lightweight design model with stress and modal constraints is derived.

[0007] (2) Determine the range of design variables and construct the design space by combining the assembly connection relationship between the crossbeam and the longitudinal beam. Within the design space, use uniform experiments to generate After selecting the target population from the candidate populations using the correlation criterion, simulation evaluation was conducted, and a database was constructed.

[0008] (3) Using population weight, stress and modal values ​​as inputs and optimization parameters as outputs, construct a network architecture with 1 input layer, M hidden layers and 1 output layer, and train RMNN according to the database;

[0009] (4) The index matrix is ​​obtained by screening the population using a random selection strategy, and then input into the RMNN to generate a candidate frame set;

[0010] (5) Establish a radial basis function model based on the database, select the optimal chassis and perform simulation evaluation, update the database, update the population using the error-driven feasibility criterion, return to step (3) until the optimization target is achieved, and output the optimal solution of the optimization parameters.

[0011] Furthermore, step (1) specifically includes the following steps:

[0012] The first step is to use the length, width, and thickness of each crossbeam of the front subframe as design variables; and then, in 3D modeling software, to construct a 3D model of the front subframe based on these design variables.

[0013] The second step is to import the three-dimensional model into the finite element analysis software, apply corresponding constraints and force loads according to its static load distribution under actual working conditions, and perform static analysis to obtain the weight simulation model and stress simulation model of the front subframe.

[0014] The third step is to process the 3D model in HyperMesh and build a finite element model for dynamic analysis. Based on this, modal analysis is performed to calculate the natural frequencies and mode shapes of the structure in the free state, thereby constructing a modal simulation model of the front subframe of the automobile.

[0015] The fourth step involves deriving a lightweight design model for the front subframe of a vehicle, based on the weight simulation model, stress simulation model, and modal simulation model, with the goal of minimizing the weight of the front subframe under both stress and modal constraints. The specific expression is as follows:

[0016] ,

[0017] ,

[0018] ,

[0019] ,

[0020] ,

[0021] In the above formula, This indicates finding the optimal solution for the optimized parameters. This indicates the optimized parameters of the car's front subframe. This indicates the length of the first beam. Indicates the first The length of each beam This indicates the width of the first beam. Indicates the first The width of each beam, This indicates the thickness of the first crossbeam. Indicates the first The thickness of each beam, Indicates the number of beams. This indicates that the weight of the front subframe should be minimized. This indicates the weight of the car's front subframe. Indicating the optimization parameters The front subframe was divided into The weight function of the front subframe of the car obtained by summing the components. This indicates the density of the front subframe. Indicating the optimization parameters The corresponding volume of the front subframe, This represents the constraints that the optimization parameters need to satisfy. This indicates that the front subframe is undergoing parameter optimization. The corresponding modal value at that time, This indicates that the front subframe is undergoing parameter optimization. The corresponding modal function at time, This represents the first-order natural frequency of the front subframe. This indicates that the front subframe is undergoing parameter optimization. The corresponding stress calculation function, This indicates the allowable stress of the front subframe. This indicates the design space comprised of optimized parameters for the front subframe.

[0022] Furthermore, step (2) specifically includes the following steps:

[0023] The first step is to combine the assembly relationship between the front subframe crossbeam and the longitudinal beam, analyze the geometric constraints and connection requirements, determine the value range of each design variable, and construct the design space based on the value range of the design variables.

[0024] The second step involves using a uniform experimental design method to generate [samples] within the design space. One candidate population;

[0025] The third step is to screen candidate populations based on the correlation criterion and calculate the score for each candidate population. The specific process for calculating the score based on the correlation criterion is as follows:

[0026] In order of variables, remove the linear correlation between each variable and all previous variables in turn, and update the data matrix with the rank fraction of its residual;

[0027] In reverse order of variables, remove the linear correlation between each variable and all subsequent variables, and update the data matrix with the rank fraction of its residuals;

[0028] Repeat the above two processes until the score of the current data matrix cannot surpass the historical best score or the maximum number of iterations is reached, and finally output the data with the optimal correlation structure.

[0029] The fourth step is to select the candidate population with the highest score as the target population and use the weight simulation model, stress simulation model and modal simulation model of the front subframe of the car to perform simulation evaluation. The obtained weight, stress and modal values ​​of the front subframe of the car are used to build a database.

[0030] Furthermore, step (3) specifically includes the following steps:

[0031] The first step is to construct a network architecture containing one input layer, M hidden layers, and one output layer. The number of neurons in the input layer is determined by the dimensions of the population's weight, stress, and modal values. M hidden layers are configured, each containing P neurons. A non-linear activation function is introduced after each hidden layer to enhance the model's expressive power. The number of neurons in the output layer is determined by the dimensions of the optimization parameters of the front subframe.

[0032] The second step is to split each sample in the database into input and output. The input data are the population weight, stress and modal values, and the output data are the corresponding optimization parameters. The dataset is divided into training set, validation set and test set in proportions of 70%, 15% and 15%. The training set is used for model parameter learning, the validation set is used for hyperparameter tuning and early stopping to prevent overfitting, and the test set is used for final evaluation of model performance.

[0033] The third step is to train the RMNN using the processed data, calculate the network output through forward propagation, and use the loss function to measure the difference between the output and the true value of the front subframe optimization parameters. Then, the backpropagation algorithm is used in combination with the optimizer to calculate the gradient and update the network weights and bias parameters.

[0034] The fourth step involves iterative training for T cycles while monitoring the performance on the validation set. Training is terminated early when performance no longer improves, ultimately resulting in an RNN that can back-map the optimization parameters of the front subframe from weight, stress, and modal values.

[0035] Furthermore, step (4) specifically includes the following steps:

[0036] The first step involves using a random selection strategy to screen from the target population. The populations in the database are sorted in ascending order based on their weight, stress, and modal values, and the top α% of the populations are selected.

[0037] The second step is to randomly combine the selected weight, stress, and modal values ​​to obtain an index matrix;

[0038] The third step is to use the index matrix as input to the trained RMNN, and then use the RMNN to back-map the index matrix to the design space to obtain a set of candidate frames.

[0039] The fourth step is to perform boundary checks on the candidate chassis set to ensure that the data within it is within the design space.

[0040] Furthermore, step (5) specifically includes the following steps:

[0041] The first step is to use all populations in the database to establish a radial basis function model for the weight, stress, and modal values ​​of the front subframe of the automobile. The expression of the radial basis function model is as follows:

[0042] ，

[0043] ，

[0044] ,

[0045] In the above formula, This represents the radial basis function model constructed with respect to weight. This indicates the optimized parameters of the front subframe. This represents the number of center points of the basis functions. Indicates the first weight Reference vectors, Represents a basis function vector. This indicates that each basis function is used in the front subframe optimization parameters. The calculation results Indicates the first The center point of each basis function Indicates the optimized parameters of the front subframe and basis function center The Euclidean distance between them This represents the radial basis function model constructed for the modal functions. This indicates the first modal value. Reference vectors, This represents a radial basis function model constructed for stress. This indicates the first stress value. One reference vector;

[0046] The second step involves using a radial basis function model to predict the set of optimized parameters for the front subframe. An error-driven feasibility criterion is then used to select the optimal frame for simulation evaluation to update the database. This error-driven feasibility criterion follows these rules:

[0047] When both solutions are feasible, the one with the better objective value wins.

[0048] When one solution is feasible and the other is infeasible, the feasible solution automatically wins.

[0049] When both solutions are infeasible, the solution with the less severe constraint violation should be selected first.

[0050] If two infeasible solutions have the same number of constraint violations, the solution with the smaller number of constraint violations is selected as the better solution.

[0051] The third step is to verify whether the weight, stress and modal values ​​corresponding to the current optimal frame meet the design goals and constraints respectively; if they do, output the optimal solution parameters; otherwise, return to step (3) to continue iterative optimization until all design goals and constraints are met.

[0052] In a second aspect, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the aforementioned RMNN-based lightweight design method for the front subframe.

[0053] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the aforementioned RMNN-based lightweight design method for the front subframe.

[0054] Fourthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the aforementioned RMNN-based lightweight design method for the front subframe.

[0055] In summary, the RMNN-based lightweight design method, device, medium, and program product for front subframes provided by this invention have the following improvements over the limitations of existing technologies:

[0056] 1. To address the challenges of locating and efficiently exploring feasible regions in existing lightweight front subframe designs due to the vast design space inherent in high-dimensional problems, a Real-Time Neural Network (RMNN) is applied to achieve rapid search of feasible regions. Specifically, by constructing an RNN from the performance space (including weight, stress, and modal response) to the design space, and leveraging the multi-parameter and multi-constraint characteristics of front subframe structural simulation, rapid identification and optimization of feasible regions under stress and modal constraints are achieved.

[0057] 2. Further, the radial basis function model is combined with the candidate population for prediction and screening, and the optimal solution is evaluated through simulation, effectively overcoming the difficulties of low search efficiency in high-dimensional design space and difficulty in identifying feasible regions under multiple constraints. Specifically, the radial basis function model predicts the performance of the candidate frame set, and the optimal frame is selected by combining error-driven feasibility criteria and evaluated through simulation, thereby improving the efficiency and accuracy of the optimization process.

[0058] 3. A hybrid intelligent optimization framework driven by both a Recurrent Neural Network (RMNN) and a Radial Basis Function (RBF) model was constructed. The RNN inversely infers a set of candidate frames based on given performance metrics, and then the RDF model predicts and filters these candidate frames, thereby improving the convergence speed of the optimization process. Furthermore, with the accumulation of simulation samples during iteration, the prediction accuracy of the neural network continuously improves, enabling accurate reverse deduction of design schemes that meet the constraints.

[0059] In summary, this invention can optimize high-dimensional constrained problems involving complex simulations, improving the optimization efficiency of complex simulations. Furthermore, it enables rapid searching of high-dimensional design spaces under constraints, which is beneficial for the optimization of various complex structures and demonstrates practicality. This invention is not only applicable to the lightweight design of automotive front subframes but also provides a usable algorithmic structure for other high-dimensional constrained problems. Attached Figure Description

[0060] Figure 1A simplified flowchart of a lightweight design method for a front subframe based on RMNN provided by this invention. Detailed Implementation

[0061] To more clearly illustrate the objectives, technical solutions, and advantages of this invention, a detailed description will be provided below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, the technical features in the following embodiments can be combined with each other as long as they do not conflict with each other.

[0062] Please see Figure 1 This invention provides a lightweight design method for a front subframe based on RNN, which is applicable to lightweight design optimization problems involving stress and modal constraints. Specifically, the method includes steps (1) to (5).

[0063] (1) The length, width and thickness of each crossbeam of the former subframe are used as design variables. The model is created in the three-dimensional modeling software. The weight and stress simulation model is obtained by combining static load with static analysis. The modal simulation model is constructed in HyperMesh and the lightweight design model with stress and modal constraints is derived.

[0064] Step (1) specifically includes the following steps:

[0065] The first step is to use the length, width, and thickness of each crossbeam of the front subframe as design variables; then, in 3D modeling software, a 3D model of the front subframe is constructed based on these design variables.

[0066] The second step is to import the 3D model into the finite element analysis software, apply corresponding constraints and force loads according to its static load distribution under actual working conditions, and perform static analysis to obtain the weight simulation model and stress simulation model of the front subframe.

[0067] The third step involves processing the 3D model in HyperMesh and constructing a finite element model for dynamic analysis. Based on this, modal analysis is performed to calculate the natural frequencies and mode shapes of the structure in a free state, thereby constructing a modal simulation model of the front subframe of the automobile.

[0068] The fourth step involves deriving a lightweight design model for the front subframe of a vehicle, based on the weight simulation model, stress simulation model, and modal simulation model, with the goal of minimizing the weight of the front subframe under both stress and modal constraints. The specific expression is as follows:

[0069] ,

[0070] ,

[0071] ,

[0072] ,

[0073] ,

[0074] In the above formula, This indicates finding the optimal solution for the optimized parameters. This indicates the optimized parameters of the car's front subframe. This indicates the length of the first beam. Indicates the first The length of each beam This indicates the width of the first beam. Indicates the first The width of each beam, This indicates the thickness of the first crossbeam. Indicates the first The thickness of each beam, Indicates the number of beams. This indicates that the weight of the front subframe should be minimized. This indicates the weight of the car's front subframe. Indicating the optimization parameters The front subframe was divided into The weight function of the front subframe of the car obtained by summing the components. This indicates the density of the front subframe. Indicating the optimization parameters The corresponding volume of the front subframe, This represents the constraints that the optimization parameters need to satisfy. This indicates that the front subframe is undergoing parameter optimization. The corresponding modal value at that time, This indicates that the front subframe is undergoing parameter optimization. The corresponding modal function at time, This represents the first-order natural frequency of the front subframe. This indicates that the front subframe is undergoing parameter optimization. The corresponding stress calculation function, This indicates the allowable stress of the front subframe. This indicates the design space comprised of optimized parameters for the front subframe.

[0075] (2) Determine the range of design variables and construct the design space by combining the assembly connection relationship between the crossbeam and the longitudinal beam. Within the design space, use uniform experiments to generate After selecting the target population from the candidate populations using the correlation criterion, simulation evaluation was performed, and a database was constructed.

[0076] Step (2) specifically includes the following steps:

[0077] The first step is to combine the assembly relationship between the front subframe crossbeam and the longitudinal beam, analyze the geometric constraints and connection requirements, determine the value range of each design variable, and construct the design space based on the value range of the design variables.

[0078] The second step involves using a uniform experimental design method to generate [samples] within the design space. Candidate populations.

[0079] The third step is to screen candidate populations based on the correlation criterion and calculate the score for each candidate population. The specific process for calculating the score based on the correlation criterion is as follows:

[0080] In order of variables, remove the linear correlation between each variable and all previous variables in turn, and update the data matrix with the rank fraction of its residual;

[0081] In reverse order of variables, remove the linear correlation between each variable and all subsequent variables, and update the data matrix with the rank fraction of its residuals;

[0082] Repeat the above two processes until the score of the current data matrix cannot surpass the historical best score or the maximum number of iterations is reached, and finally output the data with the optimal correlation structure.

[0083] The fourth step is to select the candidate population with the highest score as the target population and use the weight simulation model, stress simulation model and modal simulation model of the front subframe of the car to perform simulation evaluation. The obtained weight, stress and modal values ​​of the front subframe of the car are used to build a database.

[0084] (3) Using population weight, stress and modal values ​​as inputs and optimization parameters as outputs, construct a network architecture with 1 input layer, M hidden layers and 1 output layer, and train RMNN based on the database.

[0085] Step (3) specifically includes the following steps:

[0086] The first step is to construct a network architecture consisting of one input layer, M hidden layers, and one output layer. The number of neurons in the input layer is determined by the dimensions of the population's features, such as weight, stress, and modal values. M hidden layers are configured, each containing P neurons. A non-linear activation function is introduced after each hidden layer to enhance the model's expressive power. The number of neurons in the output layer is determined by the dimensions of the optimization parameters of the front subframe.

[0087] The second step is to split each sample in the database into input and output. The input data are the population's weight, stress, and modal values, and the output data are the corresponding optimization parameters. The dataset is then divided into training set, validation set, and test set in proportions of 70%, 15%, and 15%. The training set is used for model parameter learning, the validation set is used for hyperparameter tuning and early stopping to prevent overfitting, and the test set is used for final evaluation of model performance.

[0088] The third step involves training an RMNN using the processed data, calculating the network output through forward propagation, and using a loss function to measure the difference between the output and the true values ​​of the front subframe optimization parameters. Then, the backpropagation algorithm is used in conjunction with the optimizer to calculate the gradient and update the network weights and bias parameters.

[0089] The fourth step involves iterative training for T cycles while monitoring the performance on the validation set. Training is terminated early when performance no longer improves, ultimately resulting in an RNN that can back-map the optimization parameters of the front subframe from weight, stress, and modal values.

[0090] (4) The index matrix is ​​obtained by screening the population using a random selection strategy and then input into the RMNN to generate a candidate frame set.

[0091] Step (4) specifically includes the following steps:

[0092] The first step is to use a random selection strategy to screen from the target population. The populations in the database are sorted in ascending order based on their weight, stress and modal values, and the top α values ​​are selected.

[0093] The second step is to randomly combine the selected weight, stress, and modal values ​​to obtain an index matrix.

[0094] The third step is to use the index matrix as input to the trained RMNN, and then use the RMNN to back-map the index matrix to the design space to obtain a set of candidate frames.

[0095] The fourth step is to perform boundary checks on the candidate chassis set to ensure that the data within it is within the design space.

[0096] (5) Establish a radial basis function model based on the database, select the optimal chassis and perform simulation evaluation, update the database, update the population using the error-driven feasibility criterion, return to step (3) until the optimization target is achieved, and output the optimal solution of the optimization parameters.

[0097] Step (5) specifically includes the following steps:

[0098] The first step is to use all populations in the database to establish a radial basis function model for the weight, stress, and modal values ​​of the front subframe of the automobile. The expression of the radial basis function model is as follows:

[0099] ，

[0100] ，

[0101] ,

[0102] In the above formula, This represents the radial basis function model constructed with respect to weight. This indicates the optimized parameters of the front subframe. This represents the number of center points of the basis functions. Indicates the first weight Reference vectors, Represents a basis function vector. This indicates that each basis function is used in the front subframe optimization parameters. The calculation results Indicates the first The center point of each basis function Indicates the optimized parameters of the front subframe and basis function center The Euclidean distance between them This represents the radial basis function model constructed for the modal functions. This indicates the first modal value. Reference vectors, This represents a radial basis function model constructed for stress. This indicates the first stress value. Reference vectors.

[0103] The second step involves using a radial basis function model to predict the set of optimized parameters for the front subframe. An error-driven feasibility criterion is then used to select the optimal frame for simulation evaluation to update the database. This error-driven feasibility criterion follows these rules:

[0104] When both solutions are feasible, the one with the better objective value wins.

[0105] When one solution is feasible and the other is infeasible, the feasible solution automatically wins.

[0106] When both solutions are infeasible, the solution with the less severe constraint violation should be selected first.

[0107] If two infeasible solutions have the same number of constraint violations, the solution with the smaller number of constraint violations is selected as the better solution.

[0108] The third step is to verify whether the weight, stress and modal values ​​corresponding to the current optimal frame meet the design goals and constraints respectively; if they do, output the optimal solution parameters; otherwise, return to step (3) to continue iterative optimization until all design goals and constraints are met.

[0109] Example 1: This example uses the CEC2017 benchmark set to illustrate the optimization performance of the proposed lightweight front subframe design method based on RNN. The expression of the benchmark function is as follows:

[0110] ,

[0111] ,

[0112] In the above formula, This indicates minimizing the objective function. This represents the objective function that needs to be minimized. It is a displacement vector. Indicates the optimization parameters. Indicates the number of optimization parameters. Indicates the first One optimization parameter, Indicates the first to Among the optimization parameters, the first one is... One optimization parameter, Represents the constraint function.

[0113] To further illustrate the superiority of the method in this embodiment, a lightweight design method for the front subframe based on RNN in this embodiment is compared with another algorithm, GF-SAEAs. The maximum number of simulation evaluations in this embodiment is set to 1000, and the number of optimization parameters is set to 30. The mean and standard deviation of the results of 25 independent runs are compared. The results are shown in Table 1. The results show that, with the same number of simulation evaluations, the method in this embodiment is significantly better than the GF-SAEAs algorithm. It can be considered that the method in this embodiment can effectively solve the lightweight design optimization problem with stress and modal constraints.

[0114] Table 1. Comparison of optimization results of different methods

[0115]

[0116] This invention provides a lightweight design method for front subframes based on RNNs. The RNN is trained using a constructed database, and then a candidate frame set is generated based on a random selection strategy and the RNN to achieve rapid search of the feasible region. The set is then predicted using a radial basis function model and screened using an error-driven feasibility criterion to obtain the optimal frame, which is then evaluated through simulation. This improves the convergence speed and accuracy of the optimization process and provides a systematic solution for lightweight design optimization problems with stress and modal constraints.

[0117] In a second aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of a front subframe lightweight design method based on RMNN.

[0118] The memory can be volatile or non-volatile, or a combination of both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory of this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0119] The processor can be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the processor's hardware or by software instructions. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0120] The method steps of this invention can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in this application, or combinations thereof.

[0121] Software implementation can be achieved by executing functional modules (such as procedures, functions, etc.). Software code can be stored in memory and executed by the processor. Memory can be implemented in the processor or outside the processor.

[0122] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the aforementioned lightweight design method for the front subframe based on RMNN.

[0123] Computer storage media can include various media that can store program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0124] Fourthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of a lightweight design method for a front subframe based on RMNN.

[0125] Specifically, computer program products include: data signals, data signals embodied in a carrier wave, or computer-readable storage media.

[0126] It should be noted that the technical solutions described in this invention can be combined arbitrarily without conflict.

[0127] Those skilled in the art will readily understand that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A lightweight design method for a front subframe based on RMNN, characterized in that, The method includes: (1) The length, width and thickness of each crossbeam of the front subframe of the automobile are used as design variables; a three-dimensional model of the front subframe is constructed based on the design variables in a three-dimensional modeling software; the three-dimensional model is imported into a finite element analysis software, and corresponding constraints and force loads are applied according to the static load distribution under actual working conditions to perform static analysis and obtain the weight simulation model and stress simulation model of the front subframe; the three-dimensional model is processed in HyperMesh, and a finite element model for dynamic analysis is constructed to perform modal analysis, calculate the natural frequencies and mode shapes of the structure in the free state, and thus construct the modal simulation model of the front subframe of the automobile; based on the weight simulation model, stress simulation model and modal simulation model, with the minimization of the weight of the front subframe of the automobile as the optimization objective, the lightweight design model of the front subframe of the automobile is derived under the two constraints of stress and mode, and the specific expression is as follows: ， ， ， ， ， In the above formula, This indicates finding the optimal solution for the optimized parameters. The vector variable representing the optimization parameters of the front subframe of the automobile. This indicates the length of the first beam. Indicates the first The length of each beam This indicates the width of the first beam. Indicates the first The width of each beam, This indicates the thickness of the first crossbeam. Indicates the first The thickness of each beam, Indicates the number of beams. This indicates that the weight of the front subframe should be minimized. This indicates the weight of the car's front subframe. Represents the vector variables of optimization parameters The front subframe was divided into The weight function of the front subframe of the car obtained by summing the components. This indicates the density of the front subframe. Represents the vector variables of optimization parameters The corresponding volume of the front subframe, This represents the constraints that the optimization parameters need to satisfy. The vector variable representing the front subframe in the optimization parameters The corresponding modal value at that time, The vector variable representing the front subframe in the optimization parameters The corresponding modal function at time, This represents the first-order natural frequency of the front subframe. The vector variable representing the front subframe in the optimization parameters The corresponding stress calculation function, This indicates the allowable stress of the front subframe. The design space is represented by vector variables that constitute the optimization parameters of the front subframe; (2) Determine the range of design variables and construct the design space by combining the assembly connection relationship between the crossbeam and the longitudinal beam. Use the uniform experimental design method to generate within the design space. After selecting the target population from the candidate populations using the correlation criterion, simulation evaluation was conducted, and a database was constructed. (3) Using population weight, stress and modal values ​​as inputs and optimization parameters as outputs, construct a network architecture with 1 input layer, M hidden layers and 1 output layer, and train RMNN according to the database; (4) The target population is selected by random selection strategy to obtain the index matrix, which is then input into the RMNN to generate a candidate frame set; (5) Establish a radial basis function model based on the database, select the optimal chassis and perform simulation evaluation, update the database, update the population using the error-driven feasibility criterion, return to step (3) until the optimization target is achieved, and output the optimal solution of the optimization parameters.

2. The method as described in claim 1, characterized in that, Step (2) specifically includes the following steps: The first step is to combine the assembly relationship between the front subframe crossbeam and the longitudinal beam, analyze the geometric constraints and connection requirements, determine the value range of each design variable, and construct the design space based on the value range of the design variables. The second step involves using a uniform experimental design method to generate [samples] within the design space. One candidate population; The third step is to screen candidate populations based on the correlation criterion and calculate the score for each candidate population. The specific process for calculating the score based on the correlation criterion is as follows: In order of variables, remove the linear correlation between each variable and all previous variables in turn, and update the data matrix with the rank fraction of its residual; In reverse order of variables, remove the linear correlation between each variable and all subsequent variables, and update the data matrix with the rank fraction of its residuals; Repeat the above two processes until the score of the current data matrix cannot surpass the historical best score or the maximum number of iterations is reached, and finally output the data with the optimal correlation structure. The fourth step is to select the candidate population with the highest score as the target population and use the weight simulation model, stress simulation model and modal simulation model of the front subframe of the car to perform simulation evaluation. The obtained weight, stress and modal values ​​of the front subframe of the car are used to build a database.

3. The method as described in claim 1, characterized in that, Step (3) specifically includes the following steps: The first step is to construct a network architecture consisting of one input layer, M hidden layers, and one output layer. The number of neurons in the input layer is determined by the dimensions of the population's weight, stress, and modal value features. M hidden layers are configured, each containing P neurons, and a nonlinear activation function is introduced after each hidden layer. The number of neurons in the output layer is determined by the dimensions of the optimization parameters of the front subframe. The second step is to split each sample in the database into input and output. The input data are the population weight, stress and modal values, and the output data are the corresponding optimization parameters. The dataset is divided into training set, validation set and test set in proportions of 70%, 15% and 15%. The training set is used for model parameter learning, the validation set is used for hyperparameter tuning and early stopping to prevent overfitting, and the test set is used for final evaluation of model performance. The third step is to train the RMNN using the processed data, calculate the network output through forward propagation, and use the loss function to measure the difference between the output and the true value of the front subframe optimization parameters. Then, the backpropagation algorithm is used in combination with the optimizer to calculate the gradient and update the network weights and bias parameters. The fourth step involves iterative training for T cycles while monitoring the performance on the validation set. Training is terminated early when performance no longer improves, ultimately resulting in an RNN that can back-map the optimization parameters of the front subframe from weight, stress, and modal values.

4. The method as described in claim 1, characterized in that, Step (4) specifically includes the following steps: The first step involves using a random selection strategy to screen from the target population. The populations in the database are sorted in ascending order based on their weight, stress, and modal values, and the top α% of the populations are selected. The second step is to randomly combine the selected weight, stress, and modal values ​​to obtain an index matrix; The third step is to use the index matrix as input to the trained RMNN, and then use the RMNN to back-map the index matrix to the design space to obtain a set of candidate frames. The fourth step is to perform boundary checks on the candidate chassis set to ensure that the data within it is within the design space.

5. The method as described in claim 1, characterized in that, Step (5) specifically includes the following steps: The first step is to use all populations in the database to establish a radial basis function model for the weight, stress, and modal values ​​of the front subframe of the automobile. The expression of the radial basis function model is as follows: ， ， ， In the above formula, This represents the radial basis function model constructed with respect to weight. This indicates the optimized parameters of the front subframe. This represents the number of center points of the basis functions. Indicates the first weight Reference vectors, Represents a basis function vector. This indicates that each basis function is used in the front subframe optimization parameters. The calculation results Indicates the first The center point of each basis function Indicates the optimized parameters of the front subframe and the center point of the basis functions The Euclidean distance between them This represents the radial basis function model constructed for the modal functions. This indicates the first modal value. Reference vectors, This represents a radial basis function model constructed for stress. This indicates the first stress value. One reference vector; The second step involves using a radial basis function model to predict the set of optimized parameters for the front subframe. An error-driven feasibility criterion is then used to select the optimal frame for simulation evaluation to update the database. This error-driven feasibility criterion follows these rules: When both solutions are feasible, the one with the better objective value wins. When one solution is feasible and the other is infeasible, the feasible solution automatically wins. When both solutions are infeasible, the solution with the less severe constraint violation should be selected first. If two infeasible solutions have the same number of constraint violations, the solution with the smaller number of constraint violations is selected as the better solution. The third step is to verify whether the weight, stress, and modal values ​​corresponding to the current optimal frame meet the design goals and constraint requirements respectively; if they do, the optimal solution parameters are output. Otherwise, return to step (3) to continue iterative optimization until all design goals and constraints are met.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1-5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-5.

Images