A rudder wing component intelligent lightweight design method based on multi-objective collaborative optimization

Abstract

This invention relates to the technical field of aerospace structural design, and provides an intelligent lightweight design method for rudder components based on multi-objective collaborative optimization. The method includes parametric modeling, establishing an automated aerodynamic simulation process, constructing an aerodynamic performance dataset, building a CNN aerodynamic surrogate model, and multi-objective optimization using the NSGA-II algorithm. By constructing a parameter-driven automated modeling framework, key geometric features of the components are defined as design parameters, achieving parametric representation and automatic generation of the geometric model. Through the constructed CNN aerodynamic surrogate model, the aerodynamic performance can be predicted quickly and accurately based on the input component geometric parameters. By constructing a multi-objective collaborative optimization method that integrates the CNN aerodynamic surrogate model and the NSGA-II algorithm, intelligent lightweight design of aerospace rudder components is achieved, resolving the rigid contradiction between "extreme lightweighting and high aerodynamic efficiency" in traditional design, and meeting the design requirements of rapid iteration of new aerospace equipment.

Description

Technical Field

[0001] This invention relates to the technical field of aerospace structural design, and in particular to an intelligent lightweight design method for rudder components based on multi-objective collaborative optimization. Background Technology

[0002] Extreme lightweighting is a key driving force for cost control and efficient delivery of aerospace components. As a critical functional component in the aerospace field, the lightweight design of control wings can improve equipment payload, reduce launch and operating costs, and enhance equipment maneuverability and mission adaptability, thereby propelling aerospace equipment towards low-cost, large-scale application. The core challenge of control wing lightweight design lies in balancing the inherent contradiction between structural weight reduction and aerodynamic performance. Aerodynamic performance is a core indicator determining the functional realization of a component; however, traditional lightweight design methods, while pursuing structural weight reduction, often affect key aerodynamic parameters due to a reduction in lift area. This contradiction severely restricts the optimization space for control wing performance. Furthermore, the rapid pace and short cycles of developing new aerospace equipment mean that the original trial-and-error development model is no longer sufficient to meet contemporary development needs.

[0003] Current technologies limit the design philosophy of rudder components to single-dimensional aerodynamic performance optimization, generally neglecting the crucial design dimension of structural lightweighting. Existing optimization strategies primarily employ single-objective optimization methods, focusing solely on aerodynamic characteristic optimization, failing to establish a multi-objective collaborative optimization framework that balances the inherent contradiction between rudder lightweighting (minimizing structural weight) and high aerodynamic performance (maximizing lift-to-drag ratio, optimizing handling efficiency, etc.), resulting in difficulties in achieving optimal overall performance. Furthermore, to address the low computational efficiency of traditional optimization processes, current technologies commonly employ surrogate models to replace computational fluid dynamics (CFD) simulations, predicting the optimization target response by constructing approximate mathematical models. While this approach accelerates the design iteration process, it still suffers from efficiency bottlenecks. The accuracy of surrogate models heavily relies on a large number of training samples, and obtaining each sample requires the complete execution of standard CFD procedures such as geometric modeling, mesh generation, and numerical computation. Current simulation processes lack automated processing capabilities, making it difficult to achieve efficient computation of large batches of samples.

[0004] Therefore, in order to solve the above problems, those skilled in the art need to design an intelligent lightweight design method for rudder components based on multi-objective collaborative optimization. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an intelligent lightweight design method for rudder components based on multi-objective collaborative optimization, the design method comprising:

[0006] Step 1: Parametric Modeling

[0007] Based on CAD secondary development technologies (such as SolidWorks API and NX Open), a parameter-driven automated modeling framework is built. The key geometric features of the components are defined as design parameters. The parametric representation of the geometric model is realized through programmatic modeling. According to the changes in design variables, the corresponding component geometric structure and CFD calculation model are automatically generated.

[0008] Step 2: Establish an automated process for pneumatic simulation

[0009] An automatic mesh generation algorithm built using the ANSYS Fluent Meshing script tool generates a structured-unstructured hybrid mesh. The Fluent Journal script is used to automatically configure the solution parameters (including convergence criteria, relaxation factors, etc.). Finally, a Windows batch processing task scheduling system is used to automatically submit multi-condition simulation tasks, thereby establishing a complete automated aerodynamic simulation process.

[0010] Step 3: Construction of Aerodynamic Performance Dataset

[0011] Latin Hypercube Sampling (LHS) is employed to uniformly stratify and sample key geometric parameters of the components within the design space, ensuring optimal spatial filling for each dimension and generating parameter combinations with good spatial distribution characteristics. Then, an established automated aerodynamic simulation workflow is invoked to solve for the aerodynamic performance indicators corresponding to each sample point, forming a complete "design parameter-aerodynamic performance" mapping dataset, which is then normalized. Stratified Random Sampling is used to divide the mapping dataset into training and test sets, ensuring consistency in the probability distribution of the two subsets in the parameter space, laying a data foundation for subsequent surrogate model construction. The use of Latin Hypercube Sampling to sample key geometric parameters of the components ensures good spatial distribution characteristics of the parameter combinations within the design space, providing a comprehensive and representative data foundation for subsequent model training. Furthermore, dividing the dataset into training and test sets using stratified random sampling, while ensuring consistency in their probability distribution in the parameter space, further improves the reliability of the dataset, contributing to enhanced generalization ability and prediction accuracy of the subsequent surrogate model.

[0012] Step 4: Building the CNN aerodynamic surrogate model

[0013] A convolutional neural network model with feature extraction and regression prediction capabilities was built based on the Matlab environment. The aerodynamic performance (lift coefficient CL, drag coefficient CD, and lift-to-drag ratio CL / CD) of the input component was predicted based on the key geometric parameters of the component. The accuracy of the model prediction was evaluated using root mean square error, coefficient of determination, etc., and a CNN aerodynamic surrogate model was constructed.

[0014] Step 5: Multi-objective optimization using the NSGA-II algorithm

[0015] The computational parameters of the Non-dominated Sorting Genetic Algorithm II (NSGA-Ⅱ) are set, an initial population is randomly generated, and each individual represents a set of structural design parameters. The constructed CNN surrogate model is called to calculate the objective function value. Non-dominated sorting and crowding distance calculations are performed, and individuals with low non-dominated sorting levels and large crowding distances are selected to enter the parent population. After crossover and mutation, offspring populations are generated. The parent and offspring populations are merged, and environmental selection is performed based on non-dominated sorting and crowding distance. Elite individuals are retained to form a new generation population. The iteration continues until the termination condition is met, and the first non-dominated frontier is output as the Pareto optimal solution set, which simultaneously satisfies the co-optimization of objective functions f1 and f2.

[0016] Furthermore, the key geometric features of the component include independent variables such as chord length, grid spacing, inner grid thickness, outer grid thickness, and local sweep angle.

[0017] Furthermore, design vectors of corresponding dimensions are established based on the number of design parameters, and the design space of the corresponding dimensions is uniformly stratified and sampled.

[0018] Furthermore, the convolutional neural network model architecture includes an input layer, a convolutional layer, a max pooling layer, a fully connected layer, and a regression layer. The design parameters are used as the input layer, and the output layer consists of the same number of neurons whose aerodynamic performance needs to be predicted. The convolutional layer extracts features from the input and introduces Dropout to reduce the risk of overfitting. Finally, the fully connected layer integrates global information, and the regression layer outputs continuous predicted values.

[0019] Furthermore, the aerodynamic performance includes lift coefficient CL, drag coefficient CD, and lift-to-drag ratio CL / CD.

[0020] Furthermore, the computational parameters of the non-dominated sorting genetic algorithm include the objective function F = (f1, f2), the initial population size N, and the maximum number of iterations T. max Crossover probability P c Probability of mutation P m .

[0021] The present invention has the following beneficial effects:

[0022] (1) This invention constructs a parameter-driven automated modeling framework, defines the key geometric features of components as design parameters, realizes the parametric representation and automatic generation of geometric models, gets rid of the tediousness of traditional manual modeling, and greatly shortens the modeling time.

[0023] (2) The present invention automates aerodynamic simulation through script tools. The entire aerodynamic simulation process does not require manual intervention, which reduces the error caused by human operation and improves the processing efficiency of simulation tasks. It can complete a large number of simulation calculations under different working conditions in a short time.

[0024] (3) The present invention can quickly and accurately predict the aerodynamic performance of a component by constructing a CNN aerodynamic proxy model based on the input component geometric parameters. Compared with the traditional aerodynamic simulation method, the proxy model greatly shortens the performance prediction time. At the same time, the model is evaluated and optimized by using evaluation indicators such as root mean square error and coefficient of determination, which ensures the accuracy of the prediction results and provides a reliable basis for subsequent optimization design.

[0025] (4) This invention uses the NSGA-II algorithm for multi-objective optimization. By reasonably setting the calculation parameters, and through the processes of population generation, selection, crossover, mutation and environmental selection, the Pareto optimal solution set is finally output. This solution set can simultaneously satisfy the synergistic optimization of multiple objective functions, providing designers with a series of optimal design schemes. On the basis of batch generation of high-precision samples and rapid prediction of aerodynamic performance, it realizes the synergistic optimization of structural weight reduction and aerodynamic efficiency, and provides key technical support for the comprehensive performance improvement of the next generation of aerospace equipment.

[0026] (5) This invention realizes the intelligent lightweight design of aerospace rudder components by constructing a multi-objective collaborative optimization method that integrates a CNN aerodynamic proxy model with the NSGA-II algorithm. It solves the rigid contradiction between "extreme lightweight and high aerodynamic efficiency" in traditional design, meets the design requirements of rapid iteration of new aerospace equipment, and promotes the upgrading of aerospace structural design towards intelligence and efficiency. Attached Figure Description

[0027] Figure 1 This is a flowchart of the lightweight design in this invention.

[0028] Figure 2 This is a schematic diagram of the key geometric features of the grid rudder in this invention.

[0029] Figure 3 This is a schematic diagram of the lift coefficient predicted by the CNN aerodynamic surrogate model of this invention.

[0030] Figure 4 This is a schematic diagram of the drag coefficient predicted by the CNN aerodynamic surrogate model in this invention.

[0031] Figure 5 This is a schematic diagram of the grid rudder optimization results of the NSGA-II algorithm based on the CNN proxy model in this invention. Detailed Implementation

[0032] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. However, these embodiments are not intended to limit the present invention. Any similar structures and similar variations of the present invention should be included in the protection scope of the present invention. The commas in the present invention all indicate the relationship between and. The English letters in the present invention are case-sensitive.

[0033] like Figures 1-2 As shown, a lightweight design is implemented for the grid fins of a certain space launch vehicle. The specific design method is as follows:

[0034] Step 1, Parametric Modeling

[0035] Based on SolidWorks secondary development technology, a grid rudder parametric modeling system was constructed using VBA macro programming technology. Key geometric features of the grid rudder, such as grid spacing d, chord length c, inner grid thickness t1, outer grid thickness t2, and local sweep angle α, were used as design parameters. A Visual Basic user interface was developed to allow for the numerical input of these design parameters. After the user inputs the parameter values ​​on the interface, the system automatically calls the program to create and update the 3D model of the grid rudder and the external flow field model based on the input design parameters. Figure 2 As shown, this establishes a parametric modeling framework, providing a foundation for subsequent performance prediction and optimization design.

[0036] Step 2, Automated Process of Pneumatic Simulation

[0037] An automatic mesh generation algorithm built using the ANSYS Fluent Meshing script tool generates a hybrid structured-unstructured polyhedral mesh. For the complex geometry of the grid fin, a boundary layer structured mesh is applied to the surfaces of the rocket body and grid fin, while an unstructured polyhedral mesh is used in the external flow field. The script tool programmatically drives key parameters such as mesh size, number of layers, and growth rate. For the solution, a k-ωSST turbulence model is used for closure based on the Reynolds-averaged Navier-Stokes (RANS) equations, and a second-order upwind scheme is set for spatial discretization. The lift and drag coefficients are automatically solved using the Fluent Journal script. Finally, a Windows batch processing task scheduling system is used to implement parameterized configuration and batch submission of multi-condition simulation tasks through workflow scripts, forming an automated aerodynamic simulation process from geometric modeling and mesh generation to numerical solution.

[0038] Step 3, Construction of aerodynamic performance dataset

[0039] The geometry of the grid rudder can be represented as a 5-dimensional design vector, i.e., X = [dc t1 t2α]. Considering the practical engineering constraints of each design variable, its value range is determined. The Latin hypercube sampling (LHS) method is used to perform uniform stratified sampling of this 5-dimensional design space. To verify the feasibility and computational efficiency of this patented method, 100 samples were extracted in the preliminary research stage. Among them, X i Let d represent the i-th sample. i c represents the grid spacing of the grid rudder in the i-th sample. i t1 represents the chord length of the grid rudder in the i-th sample. i t2 represents the inner grid thickness of the grid rudder in the i-th sample. i α represents the outer grid thickness of the grid rudder in the i-th sample. i Let represent the local sweep angle of the grid fin in the i-th sample; based on the automated aerodynamic simulation process established in step 2, calculate the aerodynamic coefficients of each sample point at 0.8 Ma and an angle of attack of 10°, and generate a training dataset containing 100 sample points. Among them, Y i CL represents the output aerodynamic performance of the i-th sample. i CD represents the lift coefficient obtained from the i-th sample. i This represents the resistance coefficient obtained from the i-th sample; and the dataset is normalized and divided into training and test sets according to a certain ratio.

[0040] Step 4, Construction of CNN aerodynamic surrogate model

[0041] A one-dimensional convolutional neural network (1D-CNN) for regression tasks was built using Matlab. Five design parameters of the grid rudder were used as input layers, and the output layer consisted of two neurons: the lift coefficient (CL) and the drag coefficient (CD). Two convolutional layers were designed. The first convolutional layer extracted local features, which were then activated by ReLU and subjected to max pooling for dimensionality reduction. A second convolutional layer was then superimposed to further abstract higher-order features, and 10% Dropout was introduced to reduce the risk of overfitting. Finally, a fully connected layer integrated global information, and the regression layer output continuous predicted values. The prediction results are shown below. Figures 3-4 As shown, the constructed CNN aerodynamic agent model is trained and tested using training and testing sets.

[0042] Step 5, NSGA-II algorithm multi-objective optimization

[0043] Based on the structural weight reduction and aerodynamic performance optimization problem of grid rudders, the above five key geometric features (d, c, t1, t2, α) are used as design variables. The multi-objective optimization objectives are minimizing volume (weight reduction) and maximizing lift-to-drag ratio (aerodynamic efficiency). Constraints are set (CL ≥ 0.25), and the feasible region of the design variables is limited, transforming the optimization problem into a minimization problem. The optimization objective is:

[0044]

[0045] And satisfy:

[0046]

[0047] In the formula, f1 and f2 are objective functions; V is the volume of the grid rudder;

[0048] The NSGA-II algorithm based on a CNN surrogate model was used to solve the above optimization problem. The calculation of the objective function f2 was performed using the CNN surrogate model in all populations. The algorithm parameters were set as follows: initial population size 80; maximum number of iterations 100; crossover probability 0.7; mutation probability 0.02. The optimization results are as follows: Figure 5 As shown, the Pareto optimal solution set contains 40 Pareto points. Selecting the optimal solution can obtain the optimal geometric parameters of the grid rudder. Inputting the geometric parameters into the parametric modeling form interface can output the optimal grid rudder geometry.

[0049] This invention achieves intelligent lightweight design of aerospace rudder components by constructing a multi-objective collaborative optimization method that integrates a CNN aerodynamic proxy model with the NSGA-II algorithm. It solves the rigid contradiction between "extreme lightweighting and high aerodynamic efficiency" in traditional design, meets the design requirements of rapid iteration of new aerospace equipment, and promotes the upgrading of aerospace structural design towards intelligence and efficiency.

[0050] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

Claims

1. A smart lightweight design method for rudder / wing components based on multi-objective collaborative optimization, characterized in that, The design method includes: Step 1: Parametric Modeling A parameter-driven automated modeling framework is built based on CAD secondary development technology. The key geometric features of the components are defined as design parameters. The parametric representation of the geometric model is realized through programmatic modeling. Based on the changes in design variables, the corresponding component geometric structure and CFD calculation model are automatically generated. Step 2: Establish an automated process for pneumatic simulation An automatic mesh generation algorithm built using script tools generates a structured-unstructured hybrid mesh, and the solution parameters are automatically configured using scripts. Finally, a Windows batch processing task scheduling system is used to automatically submit multi-condition simulation tasks, thereby establishing a complete automated aerodynamic simulation process. Step 3: Construction of Aerodynamic Performance Dataset The Latin hypercube sampling method is used to uniformly stratify and sample the key geometric parameters of the components in the design space to generate parameter combinations. Then, the established automated aerodynamic simulation process is called to solve the aerodynamic performance index corresponding to each sample point, forming a complete "design parameter-aerodynamic performance" mapping dataset, which is then normalized. The mapping dataset is divided into training and test sets using a stratified random sampling method. Step 4: Construction of CNN aerodynamic surrogate model A convolutional neural network model with feature extraction and regression prediction capabilities is built to predict the aerodynamic performance of the input component based on its key geometric parameters. The accuracy of the prediction is evaluated, and a CNN aerodynamic surrogate model is constructed. Step 5: Multi-objective optimization using the NSGA-II algorithm Set the computational parameters of the non-dominated sorting genetic algorithm, randomly generate an initial population, and represent a set of structural design parameters for each individual. Call the constructed CNN surrogate model to calculate the objective function value; perform non-dominated sorting and crowding distance calculation, select individuals with low non-dominated sorting level and large crowding distance to enter the parent population, and generate offspring population through crossover and mutation; merge the parent and offspring populations, and perform environmental selection based on non-dominated sorting and crowding distance, retain elite individuals to form a new generation population, iterate until the termination condition, and output the first non-dominated frontier as the Pareto optimal solution set; The convolutional neural network model architecture includes an input layer, a convolutional layer, a max pooling layer, a fully connected layer, and a regression layer. The design parameters are used as the input layer, and the output layer consists of the same number of neurons as the aerodynamic performance to be predicted. The convolutional layer extracts features from the input and introduces Dropout to reduce the risk of overfitting. Finally, the fully connected layer integrates global information, and the regression layer outputs continuous predicted values. Step 1 specifically includes the following steps: Based on SolidWorks secondary development technology, a grid rudder parametric modeling system was constructed using VBA macro programming technology. The grid spacing d, chord length c, inner grid thickness t1, outer grid thickness t2, and local sweep angle α of the grid rudder were used as design parameters. By developing a Visual Basic user interface, the numerical input function of the design parameters was realized. After the user inputs the values ​​of each parameter in the form interface, the system automatically calls the program to complete the creation and updating of the grid rudder 3D model and the external flow field model based on the input design parameters, thereby establishing a parametric modeling framework and providing a foundation for subsequent performance prediction and optimization design. Step 2 specifically includes the following steps: An automatic mesh generation algorithm built using the ANSYS Fluent Meshing script tool generates a hybrid structured-unstructured polyhedral mesh. For the complex geometry of the grid fin, a boundary layer structured mesh is applied to the surfaces of the rocket body and grid fin, while an unstructured polyhedral mesh is used in the external flow field. The script tool programmatically drives key parameters such as mesh size, number of layers, and growth rate. For the solution, a k-ωSST turbulence model is used for closure based on the Reynolds-averaged Navier-Stokes (RANS) equations, and a second-order upwind scheme is set for spatial discretization. The lift and drag coefficients are automatically solved using the Fluent Journal script. Finally, a Windows batch processing task scheduling system is used to implement parameterized configuration and batch submission of multi-condition simulation tasks through workflow scripts, forming an automated aerodynamic simulation process from geometric modeling and mesh generation to numerical solution. Step 4 specifically includes the following steps: A one-dimensional convolutional neural network (1D-CNN) for regression tasks was built using Matlab. Five design parameters of the grid rudder were used as input layers, and the output layer consisted of two neurons: the lift coefficient CL and the drag coefficient CD. Two convolutional layers were designed. The first convolutional layer extracted local features, which were then activated by ReLU and reduced in dimensionality by max pooling. The second convolutional layer was then superimposed to further abstract higher-order features, and 10% Dropout was introduced to reduce the risk of overfitting. Finally, a fully connected layer was used to integrate global information, and the regression layer output continuous predicted values. The constructed CNN aerodynamic proxy model was trained and tested using training and test sets. Step 5 specifically includes the following steps: Based on the structural weight reduction and aerodynamic performance optimization problem of grid fins, the above 5 key geometric features ( d , c , t 1. t 2. α Using the design variables as the design objectives, and minimizing volume and maximizing lift-to-drag ratio as the multi-objective optimization goals, constraints are set, and the feasible region of the design variables is limited. This transforms the optimization problem into a minimization problem, with the optimization objective being: In the formula, f 1, f 2 is the objective function; V The volume of the grid rudder; The NSGA-II algorithm based on a CNN surrogate model is used to solve the above optimization problem. Among all populations, the optimization objective function... f The calculations for 2 all call the CNN proxy model.

2. The intelligent lightweight design method for rudder components based on multi-objective collaborative optimization according to claim 1, characterized in that, The key geometric features of the component include chord length, grid spacing, inner grid thickness, outer grid thickness, and local sweep angle.

3. The intelligent lightweight design method for rudder components based on multi-objective collaborative optimization according to claim 1, characterized in that, In step 3, design vectors of corresponding dimensions are established according to the number of design parameters, and the design space of the corresponding dimensions is uniformly stratified and sampled.

4. The intelligent lightweight design method for rudder components based on multi-objective collaborative optimization according to claim 1, characterized in that, The aerodynamic performance includes lift coefficient, drag coefficient, and lift-to-drag ratio.

5. The intelligent lightweight design method for rudder components based on multi-objective collaborative optimization according to claim 1, characterized in that, The computational parameters of the non-dominated sorting genetic algorithm include the objective function, initial population size, maximum number of iterations, crossover probability, and mutation probability.

Images