Simulation-based aircraft aerodynamic configuration design method and device

Abstract

The invention provides an aircraft aerodynamic configuration design method and device based on simulation, and relates to the technical field of aircraft aerodynamic configuration design, and the method comprises the steps: obtaining aerodynamic characteristic parameters of aircrafts of different airfoils to construct a multi-source aerodynamic database; a neural network model is established based on the database, airfoil geometric parameters serve as input, the lift coefficient, the resistance coefficient and the lift-drag ratio serve as labels for training, and an aerodynamic performance prediction model is obtained; candidate combinations are generated in an airfoil geometric parameter design space, aerodynamic characteristics of the candidate combinations are predicted through a model, a fitness function is optimized by utilizing dynamic weights, and the combination with the highest fitness is selected as an optimized airfoil parameter by adopting a genetic algorithm; and performing simulation verification on the optimization parameters, extracting a simulation result, comparing the simulation result with the output of the prediction model, judging that the optimization combination is effective when an error meets a preset requirement, and completing aerodynamic configuration design accordingly. The method improves the prediction accuracy and design efficiency of aerodynamic performance.

Description

Technical Field

[0001] This invention relates to the field of aircraft aerodynamic shape design technology, specifically to a simulation-based aircraft aerodynamic shape design method and apparatus. Background Technology

[0002] In modern aircraft design, the aerodynamic performance of airfoils is one of the core factors affecting the overall performance of an aircraft. Traditional airfoil design methods mainly rely on experience and wind tunnel testing, a process that is not only time-consuming but also costly. Designers need to conduct multiple wind tunnel tests to obtain aerodynamic characteristic parameters such as lift coefficient, drag coefficient, and lift-to-drag ratio for different airfoils. However, the limitations of traditional methods lie in the fact that their repeatability and reliability are often affected by experimental conditions, and the scarcity or unrepresentativeness of experimental data can also lead to unsatisfactory final design results. Furthermore, the increasing demand for high-performance and multi-functional aircraft makes it difficult to achieve real-time iteration and optimization using traditional experimental methods alone, thus failing to meet increasingly complex design requirements.

[0003] With the development of computational fluid dynamics (CFD) and machine learning technologies, the prediction and optimization of aerodynamic performance are gradually shifting towards digitalization and intelligence. While CFD can provide relatively accurate predictions of aerodynamic characteristics, its computational complexity and time remain high, limiting its application by designers in rapid iterations. Simultaneously, although machine learning-based models can utilize existing data for training and inference, saving computational resources and time, current approaches often lack mechanisms for dynamically adjusting weights. In practical applications, the optimization requirements for different aerodynamic characteristics are not consistent, leading to the potential neglect of some important aerodynamic features. Furthermore, existing technologies typically rely on static weight allocation during optimization, failing to adaptively adjust optimization strategies based on the actual performance of different airfoils. This lack of dynamic adjustment capability reduces the accuracy and reliability of aerodynamic performance predictions. Therefore, establishing a model that integrates multiple aerodynamic characteristic parameters and achieves dynamic optimization has become a key challenge for improving the accuracy of aerodynamic performance predictions and the efficiency of optimization design.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a simulation-based method and apparatus for designing the aerodynamic shape of aircraft, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A simulation-based aerodynamic shape design method for aircraft, comprising the following steps:

[0008] Step 1: Obtain several sets of aerodynamic characteristic parameters of existing aircraft with different airfoils. The aerodynamic characteristic parameters include airfoil geometric parameters and corresponding lift coefficient, drag coefficient and lift-to-drag ratio. The airfoil geometric parameters are specifically the airfoil thickness distribution curve function, three-dimensional twist angle distribution function and sweep angle change rate function. Map each set of airfoil geometric parameters to the lift coefficient, drag coefficient and lift-to-drag ratio to construct a multi-source aerodynamic database.

[0009] Step 2: Establish a neural network model based on a multi-source aerodynamic database. Use the airfoil geometric parameters in the multi-source aerodynamic database as the input of the model, and use the lift coefficient, drag coefficient and lift-to-drag ratio in the multi-source aerodynamic database as labels to train the neural network model and obtain an aerodynamic performance prediction model.

[0010] Step 3: Sample and generate candidate airfoil geometric parameter combinations within the airfoil geometry parameter design space. Obtain the lift coefficient, drag coefficient, and lift-to-drag ratio for each combination using an aerodynamic performance prediction model. Calculate the aerodynamic performance evaluation index for each combination as its fitness value. The fitness value is used to evaluate the airfoil's aerodynamic performance under that combination. Iteratively optimize the candidate airfoil geometric parameter combinations using a genetic algorithm to select the optimal airfoil parameter combinations. In each iteration, dynamically adjust the weights of the lift coefficient, drag coefficient, and lift-to-drag ratio in the fitness value calculation based on the distribution of lift coefficient, drag coefficient, and lift-to-drag ratio among individuals in the population.

[0011] Furthermore, the airfoil thickness distribution curve function refers to the function of the thickness variation of the airfoil in its chord direction, expressed as a percentage thickness. The thickness distribution map is obtained through experimental data and numerical calculations, and fitted into a mathematical function as the airfoil thickness distribution curve function.

[0012] The three-dimensional torsion angle distribution function refers to the mathematical modeling of the torsion angle from the wing root to the wingtip along the chord or wingspan. Through wind tunnel experiments, the torsion angle at each position is determined and a corresponding function is formed, which serves as the three-dimensional torsion angle distribution function.

[0013] The sweep angle change rate function refers to the rate of change of the sweep angle of the airfoil in the wingspan direction. It is represented by the sweep angle change curve from the wing root to the wing tip and is derived by setting the geometric parameters during airfoil design.

[0014] Each set of airfoil geometric parameters is mapped to its corresponding lift coefficient, drag coefficient, and lift-to-drag ratio to construct a multi-source aerodynamic database.

[0015] Furthermore, based on data from a multi-source aerodynamic database, a neural network model was established. The Long Short-Term Memory (LSTM) network model was selected as the base model, and activation functions and optimization algorithms were chosen. The Tanh function was selected as the activation function, and Adam was selected as the optimization algorithm for the LSTM model.

[0016] The Tanh function expression is as follows:

[0017]

[0018] In the formula, f(x) represents the Tanh function, and the independent variable x represents the weighted sum of the neuron's inputs, that is, the result of the weighted sum of the inputs received by the neuron from the previous layer. Simultaneously, the hyperparameters of the LSTM model are set, including: number of network layers, number of iterations, learning rate, batch size, number of training iterations, number of batches, and number of hidden layer neurons. Specifically, the number of network layers is set to a 3-layer network structure, the number of iterations is set to 180, the learning rate is set to 0.001, the batch size is set to 32, the number of training iterations is set to 100, the number of batches is set to 256, and the number of hidden layer neurons is 32.

[0019] A multi-source aerodynamic database is randomly divided into training and testing sets. An aerodynamic performance prediction model is constructed based on a deep learning algorithm. The airfoil geometry parameters in the training set are used as input, and the lift coefficient, drag coefficient, and lift-to-drag ratio are used as labels to train the model, resulting in a trained aerodynamic performance prediction model. The airfoil geometry parameters in the testing set are then substituted into the trained model to obtain the corresponding prediction results. The error between the prediction results and the actual values ​​in the testing set is calculated. It is then determined whether the error meets a preset error threshold. If it does, the trained model, i.e., the aerodynamic performance prediction model, is output; otherwise, training continues. The error is defined as the mean absolute error, root mean square error, and coefficient of determination between the prediction results and the actual values ​​in the testing set.

[0020] The process of randomly dividing the data into training and testing sets is as follows: the multi-source aerodynamic database is randomly sorted, 70% of the sorted multi-source aerodynamic database is used as the training set, and the remaining 30% is used as the testing set.

[0021] The trained aerodynamic performance prediction model achieves end-to-end nonlinear mapping from airfoil geometry parameters to lift coefficient, drag coefficient, and lift-to-drag ratio.

[0022] Furthermore, candidate parameter combinations are generated by sampling within the airfoil geometry design space. The lift coefficient, drag coefficient, and lift-to-drag ratio of each candidate parameter combination are obtained through an aerodynamic performance prediction model. The specific logic behind this is as follows:

[0023] The ranges of the airfoil thickness distribution curve function, the three-dimensional twist angle distribution function, and the sweep angle change rate function are determined to form the design space for airfoil geometric parameters. Within this design space, a Latin hypercube experimental design method is used to generate candidate parameter combinations. Multiple sets of candidate airfoil geometric parameter combinations are uniformly extracted from the design space using space-filling sampling technology. These generated combinations are then input into a pre-trained aerodynamic performance prediction model. The model automatically calculates and outputs the lift coefficient, drag coefficient, and lift-to-drag ratio for each parameter set. Finally, an evaluation matrix containing all candidate airfoil geometric parameter combinations and their corresponding aerodynamic performance indicators, including lift coefficient, drag coefficient, and lift-to-drag ratio, is output.

[0024] Furthermore, the expression for the fitness function is as follows:

[0025]

[0026] In the formula, PPF represents the fitness value of the j-th individual in the population after the i-th iteration optimization, i.e., the aerodynamic performance evaluation index, i is the index of the iteration number, j is the index of the individual in the population, CL(i,j), CD(i,j), and GTR(i,j) are the lift coefficient, drag coefficient, and lift-to-drag ratio of the j-th individual in the population after the i-th iteration optimization, respectively. max (i) and CL min (i) represent the maximum and minimum lift coefficients from the 1st iteration to the i-th iteration, respectively; CD max (i) and CD min (i) represent the maximum and minimum resistance coefficients from the 1st iteration to the i-th iteration, respectively; GTR max (i) and GTR min (i) represent the maximum and minimum boost-to-drag ratios from the 1st iteration to the i-th iteration, respectively; CL ref CD ref and GTR ref The reference values ​​are, in order: lift coefficient reference value, drag coefficient reference value, and lift-to-drag ratio reference value; α i β i and γ i α is a pre-set scaling factor for the i-th iteration process. i >β i >γ i >0, and satisfy α i +β i +γ i =1.

[0027] Furthermore, a genetic algorithm is used to select the candidate airfoil geometry parameters with the highest fitness function value as the optimized airfoil parameter combination. The logic is as follows: Initialize a population of candidate airfoil geometry parameter combinations with a size of M. Each individual represents a set of airfoil geometry parameter codes. A roulette wheel selection strategy is used, and individuals are selected based on their fitness function values. Individuals with higher fitness function values ​​have a greater probability of being selected. Perform a simulated binary crossover operation on the selected individuals, with a crossover probability of Pc, to generate a new generation of parameter combinations. Perform a Gaussian mutation operation on the crossover individuals, with a mutation probability of Pm. Use an elite retention strategy to directly pass the best individual of the current generation to the next generation. Repeat the selection, crossover, and mutation operations until the preset number of iterations N is reached. Finally, the candidate airfoil geometry parameter combination with the highest fitness function value is output as the optimization result, which is the optimized airfoil parameter combination.

[0028] Furthermore, the lift coefficient, drag coefficient, and lift-to-drag ratio are dynamically weighted and optimized based on the fitness function. The specific logic behind this optimization is as follows:

[0029] After each iteration, the distribution of lift coefficient, drag coefficient, and lift-to-drag ratio of all individuals in the current population is statistically analyzed, and the following indicators are calculated;

[0030]

[0031] In the formula, CL opt,i CD opt,i and GTR opt,i Let CL represent the optimization degree indices of lift coefficient, drag coefficient, and lift-to-drag ratio in the i-th generation population, respectively. avg,i CD opt,i and GTR opt,i Let represent the average lift coefficient, average drag coefficient, and average lift-to-drag ratio of the i-th generation population, respectively, where i is the index of the iteration number, and CL ref CD ref and GTR ref The reference values ​​are, in order: lift coefficient reference value, drag coefficient reference value, and lift-to-drag ratio reference value.

[0032] The optimization degree indicators of lift coefficient, drag coefficient, and lift-to-drag ratio are compared with their corresponding preset optimization thresholds. If an optimization degree indicator is greater than or equal to its preset optimization threshold, it means that the parameter has been sufficiently optimized, and its weight is reduced to redirect computing resources to other parameters to be optimized. Conversely, if the optimization degree indicator is less than its preset optimization threshold, the corresponding weight is increased to strengthen the optimization.

[0033] Among them, the proportional relationship of the weights after constraint adjustment always satisfies α. i >β i >γi >0, and satisfy α i +β i +γ i =1; α i β i and γ i This represents the weight after the i-th iteration;

[0034] The specific logic behind dynamically adjusting the weights is as follows:

[0035] The current weights are defined as: α i =α i-1 ,β i =β i-1 γ i =γ i-1 , where i≥1;

[0036] If CL opt,i ≥X CL , then α i =α i-1 *0.9;

[0037] If CD opt,i ≥X CD , then β i =β i-1 *0.9;

[0038] If GTR opt,i ≥X GTR Then γ i =γ i-1 *0.9;

[0039] In the formula, X CL X CD and X GTR These are the preset optimization thresholds corresponding to the optimization degree indicators of the lift coefficient, drag coefficient, and lift-to-drag ratio.

[0040] Conversely, if certain optimization indicators fall below their thresholds, the corresponding weights will increase by 10%.

[0041] Continue normalizing the weights; if the adjusted α i ≤β i Then exchange α i and β i Similarly, ensure α i >β i >γ i .

[0042] Furthermore, CFD simulations were performed to verify the optimized airfoil parameter combination. The specific logic behind this was as follows:

[0043] A three-dimensional mesh model of the aircraft is generated based on optimized airfoil parameter combinations. Steady flow simulation is performed using the SST k-ω turbulence model, with inflow conditions including Mach number Ma and Reynolds number Re. The Reynolds-averaged Navier-Stokes equations are solved to obtain the flow field solution. The pressure coefficient distribution and shear stress distribution on the airfoil surface are calculated by integration from the converged flow field solution. The lift coefficient, drag coefficient, and lift-to-drag ratio are obtained through numerical integration. The calculated lift coefficient, drag coefficient, and lift-to-drag ratio are output as the simulation aerodynamic characteristic parameters. Mesh independence verification requires ensuring that the change of each airfoil geometric parameter is less than 1% after mesh refinement.

[0044] Furthermore, the relative errors of the lift coefficient, drag coefficient, and lift-to-drag ratio in the simulated aerodynamic characteristic parameters are calculated with the corresponding predicted values ​​output by the aerodynamic performance prediction model. When the relative error of the lift coefficient does not exceed 5%, the relative error of the drag coefficient does not exceed 8%, and the relative error of the lift-to-drag ratio does not exceed 10%, the optimized airfoil parameter combination is deemed to have passed the verification. If any index error exceeds the threshold, the process returns to step 2 to adjust the parameters and supplement the training of the neural network model.

[0045] The optimized airfoil parameter combination that has passed verification will be used as the final design scheme to generate an airfoil CAD model that meets engineering requirements, and output a complete design report containing airfoil geometric parameters and aerodynamic performance parameters to guide the manufacturing and assembly of the aircraft's aerodynamic shape; the aerodynamic performance indicators include lift coefficient, drag coefficient and lift-to-drag ratio.

[0046] The present invention also provides a simulation-based aerodynamic shape design device for aircraft, which is used to implement the above-mentioned simulation-based aerodynamic shape design method for aircraft, including:

[0047] The data acquisition and database construction module is used to acquire several sets of aerodynamic characteristic parameters of existing different airfoil aircraft. The aerodynamic characteristic parameters include airfoil geometric parameters and corresponding lift coefficient, drag coefficient and lift-drag ratio. The airfoil geometric parameters are specifically the airfoil thickness distribution curve function, three-dimensional twist angle distribution function and sweep angle change rate function. Each set of airfoil geometric parameters is mapped to lift coefficient, drag coefficient and lift-drag ratio to construct a multi-source aerodynamic database.

[0048] The model building module establishes a neural network model based on a multi-source aerodynamic database. The airfoil geometric parameters in the multi-source aerodynamic database are used as the input of the model, and the lift coefficient, drag coefficient and lift-to-drag ratio in the multi-source aerodynamic database are used as labels to train the neural network model and obtain an aerodynamic performance prediction model.

[0049] The airfoil parameter optimization module is used to sample and generate candidate airfoil geometric parameter combinations within the airfoil geometry parameter design space. It obtains the lift coefficient, drag coefficient, and lift-to-drag ratio for each combination through an aerodynamic performance prediction model. The module then calculates the aerodynamic performance evaluation index for each combination as its fitness value. This fitness value is used to evaluate the airfoil's aerodynamic performance under that combination. Based on a genetic algorithm, the module iteratively optimizes the candidate airfoil geometric parameter combinations to select the optimal airfoil parameter combinations. Furthermore, in each iteration, the module dynamically adjusts the weights of the lift coefficient, drag coefficient, and lift-to-drag ratio in the fitness value calculation based on the distribution of these parameters among individuals in the population.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] This invention establishes a multi-source aerodynamic database and combines it with deep learning technology to construct a highly efficient aerodynamic performance prediction model, achieving a precise mapping between airfoil geometric parameters and aerodynamic characteristic parameters. This method significantly improves the prediction accuracy of airfoil aerodynamic performance, reduces the need for traditional wind tunnel experiments, and thus lowers the time and cost in the design process. Furthermore, by employing a genetic algorithm and dynamic weight optimization mechanism, the model can adaptively adjust its optimization strategy based on the actual performance of different airfoils. This process not only improves design flexibility and efficiency but also ensures effective attention to aerodynamic characteristics at different design stages. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0053] Figure 2 This is a schematic diagram of the overall device module of the present invention;

[0054] Figure 3-5 The three-dimensional scatter plot of the corresponding aerodynamic evaluation index after the free combination of absolute deviations. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0056] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0057] Example:

[0058] Please see Figure 1 The present invention provides a technical solution:

[0059] A simulation-based aerodynamic shape design method for aircraft, comprising the following steps:

[0060] Step 1: Obtain several sets of aerodynamic characteristic parameters of existing aircraft with different airfoils. The aerodynamic characteristic parameters include airfoil geometric parameters and corresponding lift coefficient, drag coefficient and lift-to-drag ratio. The airfoil geometric parameters are specifically the airfoil thickness distribution curve function, three-dimensional twist angle distribution function and sweep angle change rate function. Map each set of airfoil geometric parameters to the lift coefficient, drag coefficient and lift-to-drag ratio to construct a multi-source aerodynamic database.

[0061] In this embodiment, the airfoil thickness distribution curve function refers to the function of the thickness change of the airfoil in its chord direction, expressed as a percentage thickness. The thickness distribution map is obtained through experimental data and numerical calculations, and fitted into a mathematical function as the airfoil thickness distribution curve function.

[0062] The three-dimensional torsion angle distribution function refers to the mathematical modeling of the torsion angle from the wing root to the wingtip along the chord or wingspan. Through wind tunnel experiments, the torsion angle at each position is determined and a corresponding function is formed, which serves as the three-dimensional torsion angle distribution function.

[0063] The sweep angle change rate function refers to the rate of change of the sweep angle of the airfoil in the wingspan direction. It is represented by the sweep angle change curve from the wing root to the wing tip and is derived by setting the geometric parameters during airfoil design.

[0064] Each set of airfoil geometric parameters is mapped to its corresponding lift coefficient, drag coefficient, and lift-to-drag ratio to construct a multi-source aerodynamic database.

[0065] The selection of airfoil thickness distribution curve function, three-dimensional twist angle distribution function, and sweep angle rate of change function as airfoil geometric parameters in this invention is significant because it allows for a comprehensive capture of the key factors influencing aerodynamic performance by finely describing the airfoil's geometric characteristics. The airfoil thickness distribution curve function reflects the thickness variation of the airfoil along the chord direction, directly affecting lift and drag generation; the three-dimensional twist angle distribution function describes the torsional changes of the airfoil from the root to the tip, helping to optimize airflow characteristics; and the sweep angle rate of change function affects the airfoil's aerodynamic efficiency and stability. The comprehensive application of these parameters provides a foundation for constructing an accurate aerodynamic model, facilitating more efficient airfoil design and optimization.

[0066] The advantage of step 1 lies in its ability to systematically summarize and analyze abundant aerodynamic data by acquiring the aerodynamic characteristic parameters of existing airfoil aircraft and constructing a multi-source aerodynamic database. Specifically, the mapping relationship between airfoil geometry parameters and lift coefficient, drag coefficient, and lift-to-drag ratio provides a solid foundation for subsequent model training. This data-driven approach avoids the shortcomings of purely relying on experience and improves the scientific rigor and accuracy of the design.

[0067] Compared to existing technologies, the advantages of this step lie in its integration of multiple airfoil aerodynamic characteristic parameters and the establishment of a database, enabling the model to be optimized by learning from a large amount of aerodynamic characteristic data. This method is more efficient than traditional wind tunnel testing, allowing for rapid iteration of design solutions, significantly shortening the development cycle and reducing costs.

[0068] In this invention, this step provides high-quality data support for the overall solution, ensuring the reliability of subsequent model training and optimization processes. By establishing a detailed and accurate aerodynamic feature database, subsequent neural network models can better predict aerodynamic performance, thereby effectively improving the performance of aircraft aerodynamic shape design and promoting innovation and optimization in the overall design process.

[0069] Step 2: Establish a neural network model based on a multi-source aerodynamic database. Use the airfoil geometric parameters in the multi-source aerodynamic database as the input of the model, and use the lift coefficient, drag coefficient and lift-to-drag ratio in the multi-source aerodynamic database as labels to train the neural network model and obtain an aerodynamic performance prediction model.

[0070] In this embodiment, a neural network model is established based on data from a multi-source aerodynamic database. The Long Short-Term Memory (LSTM) network model is selected as the base model. Activation functions and optimization algorithms are selected, with the Tanh function chosen as the activation function and Adam chosen as the optimization algorithm for the LSTM model.

[0071] The Tanh function expression is as follows:

[0072]

[0073] In the formula, f(x) represents the Tanh function, and the independent variable x represents the weighted sum of the neuron's inputs, that is, the result of the weighted sum of the inputs received by the neuron from the previous layer. Simultaneously, the hyperparameters of the LSTM model are set, including: number of network layers, number of iterations, learning rate, batch size, number of training iterations, number of batches, and number of hidden layer neurons. Specifically, the number of network layers is set to a 3-layer network structure, the number of iterations is set to 180, the learning rate is set to 0.001, the batch size is set to 32, the number of training iterations is set to 100, the number of batches is set to 256, and the number of hidden layer neurons is 32.

[0074] A multi-source aerodynamic database is randomly divided into training and testing sets. An aerodynamic performance prediction model is constructed based on a deep learning algorithm. The airfoil geometry parameters in the training set are used as input, and the lift coefficient, drag coefficient, and lift-to-drag ratio are used as labels to train the model, resulting in a trained aerodynamic performance prediction model. The airfoil geometry parameters in the testing set are then substituted into the trained model to obtain the corresponding prediction results. The error between the prediction results and the actual values ​​in the testing set is calculated. It is then determined whether the error meets a preset error threshold. If it does, the trained model, i.e., the aerodynamic performance prediction model, is output; otherwise, training continues. The error is defined as the mean absolute error, root mean square error, and coefficient of determination between the prediction results and the actual values ​​in the testing set.

[0075] The process of randomly dividing the data into training and testing sets is as follows: the multi-source aerodynamic database is randomly sorted, 70% of the sorted multi-source aerodynamic database is used as the training set, and the remaining 30% is used as the testing set.

[0076] The trained aerodynamic performance prediction model achieves end-to-end nonlinear mapping from airfoil geometry parameters to lift coefficient, drag coefficient, and lift-to-drag ratio.

[0077] The advantage of step 2 lies in the fact that by establishing a neural network model based on a multi-source aerodynamic database, it is possible to achieve complex nonlinear mapping between airfoil geometric parameters and aerodynamic characteristics. This method, through deep learning technology, can effectively process large amounts of aerodynamic data, extract features, and perform pattern recognition, thereby improving the accuracy and reliability of aerodynamic performance prediction.

[0078] Compared to existing technologies, this method offers the advantage that traditional aerodynamic performance predictions typically rely on empirical formulas or simplified models, while step 2, by utilizing advanced deep learning algorithms, can capture complex aerodynamic behavior more comprehensively and accurately. This data-driven approach not only improves prediction accuracy but also reduces the need for expensive wind tunnel testing, thereby lowering development costs and time.

[0079] In this invention, this step provides powerful data processing and predictive capabilities for the overall design, offering a more accurate foundation for subsequent airfoil optimization and design. By establishing an efficient aerodynamic performance prediction model, the overall design process is accelerated, the optimization process becomes more efficient, and ultimately, it contributes to the rapid development and design of high-performance aircraft.

[0080] Step 3: Sample and generate candidate airfoil geometric parameter combinations within the airfoil geometry parameter design space. Obtain the lift coefficient, drag coefficient, and lift-to-drag ratio for each combination using an aerodynamic performance prediction model. Calculate the aerodynamic performance evaluation index for each combination as its fitness value. The fitness value is used to evaluate the airfoil's aerodynamic performance under that combination. Iteratively optimize the candidate airfoil geometric parameter combinations using a genetic algorithm to select the optimal airfoil parameter combinations. In each iteration, dynamically adjust the weights of the lift coefficient, drag coefficient, and lift-to-drag ratio in the fitness value calculation based on the distribution of lift coefficient, drag coefficient, and lift-to-drag ratio among individuals in the population.

[0081] In this embodiment, candidate parameter combinations are generated by sampling within the airfoil geometry design space. The lift coefficient, drag coefficient, and lift-to-drag ratio of each candidate parameter combination are obtained through an aerodynamic performance prediction model. The specific logic is as follows:

[0082] The ranges of the airfoil thickness distribution curve function, the three-dimensional twist angle distribution function, and the sweep angle change rate function are determined to form the design space for airfoil geometric parameters. Within this design space, a Latin hypercube experimental design method is used to generate candidate parameter combinations. Multiple sets of candidate airfoil geometric parameter combinations are uniformly extracted from the design space using space-filling sampling technology. These generated combinations are then input into a pre-trained aerodynamic performance prediction model. The model automatically calculates and outputs the lift coefficient, drag coefficient, and lift-to-drag ratio for each parameter set. Finally, an evaluation matrix containing all candidate airfoil geometric parameter combinations and their corresponding aerodynamic performance indicators, including lift coefficient, drag coefficient, and lift-to-drag ratio, is output.

[0083] The fitness function is expressed as follows:

[0084]

[0085] In the formula, PPF represents the fitness value of the j-th individual in the population after the i-th iteration optimization, i.e., the aerodynamic performance evaluation index, i is the index of the iteration number, j is the index of the individual in the population, CL(i,j), CD(i,j), and GTR(i,j) are the lift coefficient, drag coefficient, and lift-to-drag ratio of the j-th individual in the population after the i-th iteration optimization, respectively. max (i) and CL min(i) represent the maximum and minimum lift coefficients from the 1st iteration to the i-th iteration, respectively; CD max (i) and CD min (i) represent the maximum and minimum resistance coefficients from the 1st iteration to the i-th iteration, respectively; GTR max (i) and GTR min (i) represent the maximum and minimum boost-to-drag ratios from the 1st iteration to the i-th iteration, respectively; CL ref CD ref and GTR ref The reference values ​​are, in order: lift coefficient reference value, drag coefficient reference value, and lift-to-drag ratio reference value; α i β i and γ i α is a pre-set scaling factor for the i-th iteration process. i >β i >γ i >0, and satisfy α i +β i +γ i =1. In this embodiment, the initial values ​​of α0, β0, and γ0 are 0.5, 0.3, and 0.2, respectively; CL ref CD ref and GTR ref The values ​​were set to 0.7, 0.06, and 10 respectively.

[0086] Set to α i >β i >γ i The reasons are as follows: The lift coefficient is a direct indicator of whether an aircraft can generate sufficient lift, and therefore crucial to its flight capability. Improving the lift coefficient is the primary design objective, hence its highest weight is assigned to ensure its impact is prioritized during optimization. The drag coefficient directly relates to the aircraft's energy consumption and speed performance; while important, its impact is slightly less significant than the lift coefficient. Therefore, its weight is set to medium to ensure that drag reduction is considered during optimization without overshadowing the importance of lift. The lift-to-drag ratio is a performance indicator that comprehensively considers lift and drag. Although it reflects the aircraft's efficiency, its impact is usually determined based on both lift and drag. Therefore, it is assigned the lowest weight to avoid overemphasizing the lift-to-drag ratio during optimization, which could negatively impact the optimization results for lift and drag.

[0087] In the given formula, the dependent variable PPF of the fitness function reflects the aerodynamic performance of the candidate airfoil; a higher value indicates that the airfoil's aerodynamic performance is closer to the ideal state. The purpose of PPF design is to comprehensively consider three key aerodynamic performance indicators—lift coefficient, drag coefficient, and lift-to-drag ratio—to evaluate the overall performance of the airfoil. By optimizing PPF, better flight performance can be achieved in the design, such as increasing lift, reducing drag, and improving the lift-to-drag ratio, thereby enabling the aircraft to have higher efficiency and better flight characteristics in practical applications.

[0088] The independent variables CL, CD, and GTR are directly correlated with PPF because they are the core factors constituting PPF. Changes in CL directly affect lift generation, while changes in CL affect aircraft drag, and GTR reflects the combined efficiency of lift and drag. The difference between the values ​​of the independent variables and their reference values ​​directly affects the calculated PPF. Specifically, the closer CL, CD, and GTR are to their reference values, the higher the PPF value, indicating better aerodynamic performance of the airfoil.

[0089] In the fitness function, the dependent variable PPF exhibits a positive and negative correlation with the independent variables CL, CD, and GTR. Specifically, when CL is close to CL... ref When the PPF increases, the PPF decreases; conversely, when the CL deviates from the CL... ref At that time, the PPF will decrease. Similarly, for CDs, when a CD approaches a CD... ref When PPF rises, if CD deviates from CD ref Then PPF decreases. For GTR, if GTR is close to GTR... ref If the PPF value is close to the reference value, the PPF value will increase; if it deviates from the reference value, the PPF value will decrease. This design intention is to encourage designers to make each variable as close as possible to its reference value during the optimization process, so as to improve the overall aerodynamic performance.

[0090] The rationality of the fitness function PPF is reflected in several aspects: First, using an exponential function effectively transforms performance deviations into a non-negative evaluation index, which gradually decreases as CL, CD, and GTR deviate from their reference values. This aligns with the design optimization objective: the closer the aerodynamic performance indicators are to the ideal state, the higher the evaluation index. Second, the formula employs absolute values ​​and normalization, making performance indicators of different natures comparable through the calculation of relative deviations, facilitating comprehensive evaluation within the same framework. Finally, the weighting coefficients reflect the importance of different indicators to overall performance while also satisfying the condition that their sum equals 1, allowing for a reasonable weighting of the influence of each dependent variable, thereby achieving a comprehensive evaluation of the aircraft's aerodynamic performance. This structural design ensures the logic and effectiveness of the formula.

[0091] Table 1: Statistical Table of 20 Groups of Aerodynamic Performance Evaluation Indices

[0092] Lift coefficient Maximum lift coefficient Minimum lift coefficient drag coefficient Maximum drag coefficient Minimum drag coefficient Rise-to-drag ratio Maximum lift-to-drag ratio Minimum lift-to-drag ratio Aerodynamic performance evaluation index 0.5 1.2 0.2 0.05 0.15 0.02 10 12 8 0.85 0.6 1.3 0.3 0.06 0.14 0.01 9.5 11 7 0.8 0.7 1.4 0.4 0.04 0.13 0.03 11 13 9 0.88 0.65 1.5 0.35 0.07 0.16 0.04 10.5 14 8.5 0.82 0.55 1.1 0.25 0.045 0.12 0.02 12 15 10 0.87 0.75 1.6 0.5 0.08 0.17 0.05 9 10 6 0.76 0.8 1.7 0.4 0.09 0.18 0.06 8 10.5 5 0.73 0.45 1.2 0.1 0.05 0.11 0.03 10.2 13 7.5 0.84 0.5 1.4 0.3 0.065 0.14 0.02 11.5 12 9 0.79 0.85 1.8 0.5 0.1 0.2 0.07 7.5 9 5 0.7 0.4 1.1 0.2 0.05 0.12 0.02 12.5 14 11 0.86 0.35 1.3 0.15 0.06 0.13 0.01 13 15 10 0.81 0.9 1.9 0.4 0.11 0.21 0.08 6.5 8 4 0.68 0.25 1.2 0.05 0.03 0.1 0.01 14 16 12 0.9 0.95 2 0.5 0.12 0.22 0.09 5 7 3 0.65 0.3 1.1 0.1 0.04 0.11 0.02 13.5 15 11 0.83 0.7 1.4 0.3 0.07 0.15 0.03 10.8 12.5 9 0.78 0.75 1.5 0.4 0.09 0.19 0.05 9.2 11 8 0.75 0.6 1.3 0.2 0.08 0.16 0.04 10.1 12 9.5 0.81 0.55 1.25 0.15 0.05 0.14 0.02 11.1 13 10 0.83

[0093] Please see Figure 3-5 It should be noted that the absolute deviations of the lift coefficient, drag coefficient, and lift-to-drag ratio in the figure correspond to |CL-CL in the formula. ref |item,|CD-CD ref |Xiang and|GTR-GTR ref | item.

[0094] In this data analysis, the aerodynamic performance evaluation index ranged from 0.65 to 0.9. Airfoils with a PPF > 0.85 exhibited superior overall aerodynamic performance, indicating that their lift, drag, and lift-to-drag ratio were close to the reference values. Airfoils with a PPF < 0.7 deviated significantly from the reference values, resulting in a decrease in overall aerodynamic efficiency. Furthermore, the lift-to-drag ratio significantly affected the PPF. For example, at CL = 0.4, despite lower lift, the high lift-to-drag ratio (GTR = 12.5) still resulted in a PPF of 0.86, superior to most high-lift airfoils.

[0095] Although some airfoils in the data (such as CL=0.7, CD=0.04, GTR=11) have excellent lift-to-drag ratios, their drag coefficients (CD=0.04) are significantly lower than the reference value (CD). ref =0.06), which may lead to stability issues in actual flight. This highlights the importance of dynamic weight adjustment: if the initial weights overemphasize lift, the optimization of drag or lift-to-drag ratio may be neglected. By iteratively adjusting the weights through a genetic algorithm, the relationship among the three can be balanced. For example, in later optimization, α can be reduced and γ increased to improve the lift-to-drag ratio. In addition, the CFD verification stage (e.g., error thresholds: CL≤5%, CD≤8%, GTR≤10%) effectively screens out designs with large deviations in the prediction model, ensuring that the final airfoil (e.g., candidates with PPF≥0.85) has consistency between simulation and prediction, meeting the actual engineering requirements.

[0096] In summary, this method achieves efficient aerodynamic shape design through a closed-loop process of neural network prediction, genetic algorithm optimization, and CFD verification. The dynamic weighting strategy and rigorous verification standards are key to improving design reliability.

[0097] The genetic algorithm selects the candidate airfoil geometry parameters with the highest fitness function value as the optimized airfoil parameter combination. The logic is as follows: Initialize a population of candidate airfoil geometry parameter combinations with a size of M. Each individual represents a set of airfoil geometry parameter codes. Use a roulette wheel selection strategy to select individuals based on their fitness function value, where individuals with higher fitness function values ​​have a greater probability of being selected. Perform a simulated binary crossover operation on the selected individuals, with a crossover probability of Pc, to generate a new generation of parameter combinations. Perform a Gaussian mutation operation on the crossover individuals, with a mutation probability of Pm. Use an elite retention strategy to directly pass the best individual of the current generation to the next generation. Repeat the selection, crossover, and mutation operations until the preset number of iterations N is reached. Finally, output the candidate airfoil geometry parameter combination with the highest fitness function value as the optimization result, which is the optimized airfoil parameter combination.

[0098] The lift coefficient, drag coefficient, and lift-to-drag ratio are dynamically weighted and optimized based on the fitness function. The specific logic behind this optimization is as follows:

[0099] After each iteration, the distribution of lift coefficient, drag coefficient, and lift-to-drag ratio of all individuals in the current population is statistically analyzed, and the following indicators are calculated;

[0100]

[0101] In the formula, CL opt,i CD opt,i and GTR opt,i Let CL represent the optimization degree indices of lift coefficient, drag coefficient, and lift-to-drag ratio in the i-th generation population, respectively. avg,i CD opt,i and GTR opt,i Let represent the average lift coefficient, average drag coefficient, and average lift-to-drag ratio of the i-th generation population, respectively, where i is the index of the iteration number, and CL ref CD ref and GTR ref The reference values ​​are, in order: lift coefficient reference value, drag coefficient reference value, and lift-to-drag ratio reference value.

[0102] In the given formula above, the dependent variable CL opt,i CD opt,i and GTR opt,i These metrics reflect the degree of improvement in the aerodynamic performance of the current population compared to reference values. Specifically, higher values ​​indicate that the performance metric is closer to the ideal state, thereby improving the overall aerodynamic efficiency of the aircraft. Technically, this optimization allows for continuous adjustment of weights during the design process, enabling subsequent iterations to focus more on poorly performing metrics, thus achieving a comprehensive performance improvement.

[0103] Independent variable CL avg,iCD avg,i and GTR avg,i This represents the average aerodynamic performance indicators of the current population, reflecting the overall performance of the population. Its relationship with the dependent variable lies in the fact that the degree of optimization of the dependent variable is derived by comparing it with its respective reference value. Specifically, if the average lift coefficient is close to its reference value, it indicates that the indicator is well optimized; if the average drag coefficient deviates from the reference value, then CD... opt,i A decrease indicates that the indicator needs further optimization. This relationship shows that the performance of the independent variable directly affects the degree of optimization of the dependent variable, and thus affects the overall fitness.

[0104] In this formula, when the independent variable CL avg,i When it approaches its reference value, the dependent variable CL opt,i The value of the dependent variable will approach 1, indicating a high degree of optimization; conversely, if the difference between the independent variable and the reference value increases, the value of the dependent variable will decrease. The same logic applies to CL. opt,i and GTR opt,i In other words, the optimization index increases when the average value is close to the reference value, and decreases when it deviates from the reference value. This positive and negative correlation effectively guides the design process, enabling more targeted optimization of performance indicators in subsequent iterations.

[0105] The optimization degree indicators of lift coefficient, drag coefficient, and lift-to-drag ratio are compared with their corresponding preset optimization thresholds. If an optimization degree indicator is greater than or equal to its preset optimization threshold, it means that the parameter has been sufficiently optimized, and its weight is reduced to redirect computing resources to other parameters to be optimized. Conversely, if the optimization degree indicator is less than its preset optimization threshold, the corresponding weight is increased to strengthen the optimization.

[0106] Among them, the proportional relationship of the weights after constraint adjustment always satisfies α. i >β i >γ i >0, and satisfy α i +β i +γ i =1; α i β i and γ i This represents the weight after the i-th iteration;

[0107] The specific logic behind dynamically adjusting the weights is as follows:

[0108] The current weights are defined as: α i =α i-1 ,β i =β i-1 γ i =γ i-1 , where i≥1;

[0109] If CL opt,i ≥X CL , then α i =α i-1 *0.9;

[0110] If CD opt,i ≥X CD , then β i =β i-1 *0.9;

[0111] If GTR opt,i ≥X GTR Then γ i =γ i-1 *0.9;

[0112] In the formula, X CL X CD and X GTR These are the preset optimization thresholds corresponding to the optimization degree indicators of the lift coefficient, drag coefficient, and lift-to-drag ratio.

[0113] Conversely, if certain optimization indicators fall below their thresholds, the corresponding weights will increase by 10%.

[0114] Continue normalizing the weights; if the adjusted α i ≤β i Then exchange α i and β i Similarly, ensure α i >β i >γ i .

[0115] During dynamic weight adjustment, normalization ensures that the sum of the weight coefficients is always 1, maintaining a reasonable scale for the fitness function. The core purpose of forced weight swapping is to maintain the optimization priority of aerodynamic parameters. Since the lift coefficient typically has the greatest impact on aircraft performance, followed by the drag coefficient, and then the lift-to-drag ratio, this constraint prevents the algorithm from deviating excessively from the design criteria due to short-term optimization trends, such as significant CD optimization in a particular generation. This ensures that the long-term search always prioritizes lift, with drag as a secondary factor and lift-to-drag ratio as a supplement, thus meeting practical engineering needs. This mechanism enhances the stability and rationality of the optimization process by forcing the weight order.

[0116] The advantage of step 3 lies in its ability to efficiently explore the design space by sampling within the airfoil geometry design space and combining it with an aerodynamic performance prediction model to rapidly evaluate the performance of candidate parameter combinations. This significantly reduces reliance on traditional trial-and-error methods and allows the optimization process to focus more on improving key aerodynamic characteristics by dynamically adjusting the fitness function weights.

[0117] Compared to existing technologies, its advantages are as follows: Traditional airfoil optimization typically relies on manual experience or limited parameter scanning, often requiring significant computational resources and time. This step, however, combines aerodynamic performance prediction models and genetic algorithms to rapidly select the optimal solution from a large pool of candidate parameters, significantly improving optimization efficiency and accuracy. Furthermore, the dynamic weight optimization mechanism can adjust target priorities based on real-time optimization progress, giving the optimization process greater adaptability.

[0118] In this invention, this step plays a crucial role in promoting the overall design. On the one hand, it closely integrates the exploration of the design space with aerodynamic performance prediction, providing a clear evaluation basis for the optimization process. On the other hand, through the efficient search of the genetic algorithm, it can quickly approximate the optimal design parameters, laying a solid foundation for subsequent CFD simulation verification and final design finalization. This saves both time and resources while ensuring high-performance output of the aircraft's aerodynamic shape design.

[0119] The specific logic behind the CFD simulation verification of the optimized airfoil parameter combination is as follows:

[0120] A three-dimensional mesh model of the aircraft is generated based on optimized airfoil parameter combinations. Steady flow simulation is performed using the SST k-ω turbulence model, with inflow conditions including Mach number Ma and Reynolds number Re. The Reynolds-averaged Navier-Stokes equations are solved to obtain the flow field solution. The pressure coefficient distribution and shear stress distribution on the airfoil surface are calculated by integration from the converged flow field solution. The lift coefficient, drag coefficient, and lift-to-drag ratio are obtained through numerical integration. The calculated lift coefficient, drag coefficient, and lift-to-drag ratio are output as the simulation aerodynamic characteristic parameters. Mesh independence verification requires ensuring that the change of each airfoil geometric parameter is less than 1% after mesh refinement.

[0121] Calculate the relative errors between the lift coefficient, drag coefficient, and lift-to-drag ratio in the simulated aerodynamic characteristic parameters and the corresponding predicted values ​​output by the aerodynamic performance prediction model. If the relative error of the lift coefficient does not exceed 5%, the relative error of the drag coefficient does not exceed 8%, and the relative error of the lift-to-drag ratio does not exceed 10%, the optimized airfoil parameter combination is deemed to have passed the verification. If any index error exceeds the threshold, return to step 2 to adjust the parameters and supplement the training of the neural network model.

[0122] The optimized airfoil parameter combination that has passed verification will be used as the final design scheme to generate an airfoil CAD model that meets engineering requirements, and output a complete design report containing airfoil geometric parameters and aerodynamic performance parameters to guide the manufacturing and assembly of the aircraft's aerodynamic shape; the aerodynamic performance indicators include lift coefficient, drag coefficient and lift-to-drag ratio.

[0123] The advantage of this process lies in its ability to effectively evaluate and confirm the accuracy of the aerodynamic performance prediction model by performing CFD simulations on the optimized airfoil parameter combinations. This process obtains realistic aerodynamic characteristic parameters through numerical simulation, ensuring that the designed airfoil's performance under actual flight conditions can be verified, thereby increasing the reliability of the design scheme.

[0124] Compared to existing technologies, this method offers the advantage of relying on laboratory wind tunnel testing for traditional aerodynamic performance verification, which is costly and time-consuming. CFD simulation, on the other hand, allows for rapid flow field solutions on a computer, reducing the need for physical experiments, lowering the overall development cycle and cost, and enabling more detailed flow field analysis, thus improving design accuracy.

[0125] In this invention, this step plays a crucial role in promoting the overall design. On the one hand, it provides strong verification support for the optimized airfoil design, ensuring the practical feasibility of the design scheme; on the other hand, through the precise calculation of aerodynamic characteristic parameters, it further enhances the credibility of the design process, providing a reliable basis for the final generation of an airfoil CAD model that meets engineering requirements. This will help promote the efficient manufacturing and assembly of aircraft aerodynamic shapes and improve the overall project success rate.

[0126] Please see Figure 2 A simulation-based aerodynamic shape design device for aircraft, comprising:

[0127] The data acquisition and database construction module is used to acquire several sets of aerodynamic characteristic parameters of existing different airfoil aircraft. The aerodynamic characteristic parameters include airfoil geometric parameters and corresponding lift coefficient, drag coefficient and lift-drag ratio. The airfoil geometric parameters are specifically the airfoil thickness distribution curve function, three-dimensional twist angle distribution function and sweep angle change rate function. Each set of airfoil geometric parameters is mapped to lift coefficient, drag coefficient and lift-drag ratio to construct a multi-source aerodynamic database.

[0128] The model building module establishes a neural network model based on a multi-source aerodynamic database. The airfoil geometric parameters in the multi-source aerodynamic database are used as the input of the model, and the lift coefficient, drag coefficient and lift-to-drag ratio in the multi-source aerodynamic database are used as labels to train the neural network model and obtain an aerodynamic performance prediction model.

[0129] The airfoil parameter optimization module is used to sample and generate candidate airfoil geometric parameter combinations within the airfoil geometry parameter design space. It obtains the lift coefficient, drag coefficient, and lift-to-drag ratio for each combination through an aerodynamic performance prediction model. The module then calculates the aerodynamic performance evaluation index for each combination as its fitness value. This fitness value is used to evaluate the airfoil's aerodynamic performance under that combination. Based on a genetic algorithm, the module iteratively optimizes the candidate airfoil geometric parameter combinations to select the optimal airfoil parameter combinations. Furthermore, in each iteration, the module dynamically adjusts the weights of the lift coefficient, drag coefficient, and lift-to-drag ratio in the fitness value calculation based on the distribution of these parameters among individuals in the population.

[0130] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0131] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0132] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0133] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A simulation-based aerodynamic shape design method for aircraft, characterized in that, The specific steps include: Step 1: Obtain several sets of aerodynamic characteristic parameters of existing aircraft with different airfoils. The aerodynamic characteristic parameters include airfoil geometric parameters and corresponding lift coefficient, drag coefficient and lift-to-drag ratio. The airfoil geometric parameters are specifically the airfoil thickness distribution curve function, three-dimensional twist angle distribution function and sweep angle change rate function. Map each set of airfoil geometric parameters to the lift coefficient, drag coefficient and lift-to-drag ratio to construct a multi-source aerodynamic database. Step 2: Establish a neural network model based on a multi-source aerodynamic database. Use the airfoil geometric parameters in the multi-source aerodynamic database as the input of the model, and use the lift coefficient, drag coefficient and lift-to-drag ratio in the multi-source aerodynamic database as labels to train the neural network model and obtain an aerodynamic performance prediction model. Step 3: Sample and generate candidate airfoil geometric parameter combinations within the airfoil geometry parameter design space. Obtain the lift coefficient, drag coefficient, and lift-to-drag ratio for each combination using an aerodynamic performance prediction model. Calculate the aerodynamic performance evaluation index for each combination as its fitness value. The fitness value is used to evaluate the airfoil's aerodynamic performance under that combination. Iteratively optimize the candidate airfoil geometric parameter combinations using a genetic algorithm to select the optimal airfoil parameter combinations. In each iteration, dynamically adjust the weights of the lift coefficient, drag coefficient, and lift-to-drag ratio in the fitness value calculation based on the distribution of lift coefficient, drag coefficient, and lift-to-drag ratio among individuals in the population.

2. The simulation-based aerodynamic shape design method for aircraft according to claim 1, characterized in that: The airfoil thickness distribution curve function refers to the function of the thickness variation of the airfoil in its chord direction, expressed as a percentage thickness. The thickness distribution map is obtained through experimental data and numerical calculations, and fitted into a mathematical function as the airfoil thickness distribution curve function. The three-dimensional torsion angle distribution function refers to the mathematical modeling of the torsion angle from the wing root to the wingtip along the chord or wingspan. Through wind tunnel experiments, the torsion angle at each position is determined and a corresponding function is formed, which serves as the three-dimensional torsion angle distribution function. The sweep angle change rate function refers to the rate of change of the sweep angle of the airfoil in the wingspan direction. It is represented by the sweep angle change curve from the wing root to the wing tip and is derived by setting the geometric parameters during airfoil design. Each set of airfoil geometric parameters is mapped to its corresponding lift coefficient, drag coefficient, and lift-to-drag ratio to construct a multi-source aerodynamic database.

3. The simulation-based aerodynamic shape design method for aircraft according to claim 1, characterized in that: Based on data from a multi-source aerodynamic database, a neural network model was established. The Long Short-Term Memory (LSTM) network model was selected as the base model. Activation functions and optimization algorithms were selected, with the Tanh function chosen as the activation function and Adam as the optimization algorithm for the LSTM model. The Tanh function expression is as follows: In the formula, f(x) represents the Tanh function, and the independent variable x represents the weighted sum of the neuron's inputs, that is, the result of the weighted sum of the inputs received by the neuron from the previous layer. Simultaneously, the hyperparameters of the LSTM model are set, including: number of network layers, number of iterations, learning rate, batch size, number of training iterations, number of batches, and number of hidden layer neurons. Specifically, the number of network layers is set to a 3-layer network structure, the number of iterations is set to 180, the learning rate is set to 0.001, the batch size is set to 32, the number of training iterations is set to 100, the number of batches is set to 256, and the number of hidden layer neurons is 32. A multi-source aerodynamic database is randomly divided into training and testing sets. An aerodynamic performance prediction model is constructed based on a deep learning algorithm. The airfoil geometry parameters in the training set are used as input, and the lift coefficient, drag coefficient, and lift-to-drag ratio are used as labels to train the model, resulting in a trained aerodynamic performance prediction model. The airfoil geometry parameters in the testing set are then substituted into the trained model to obtain the corresponding prediction results. The error between the prediction results and the actual values ​​in the testing set is calculated. It is then determined whether the error meets a preset error threshold. If it does, the trained model, i.e., the aerodynamic performance prediction model, is output; otherwise, training continues. The error is defined as the mean absolute error, root mean square error, and coefficient of determination between the prediction results and the actual values ​​in the testing set. The process of randomly dividing the data into training and testing sets is as follows: the multi-source aerodynamic database is randomly sorted, 70% of the sorted multi-source aerodynamic database is used as the training set, and the remaining 30% is used as the testing set. The trained aerodynamic performance prediction model achieves end-to-end nonlinear mapping from airfoil geometry parameters to lift coefficient, drag coefficient, and lift-to-drag ratio.

4. The simulation-based aerodynamic shape design method for aircraft according to claim 1, characterized in that: Candidate parameter combinations are generated by sampling within the airfoil geometry design space. The lift coefficient, drag coefficient, and lift-to-drag ratio of each candidate parameter combination are obtained through an aerodynamic performance prediction model. The specific logic is as follows: The ranges of the airfoil thickness distribution curve function, the three-dimensional twist angle distribution function, and the sweep angle change rate function are determined to form the design space for airfoil geometric parameters. Within this design space, a Latin hypercube experimental design method is used to generate candidate parameter combinations. Multiple sets of candidate airfoil geometric parameter combinations are uniformly extracted from the design space using space-filling sampling technology. These generated combinations are then input into a pre-trained aerodynamic performance prediction model. The model automatically calculates and outputs the lift coefficient, drag coefficient, and lift-to-drag ratio for each parameter set. Finally, an evaluation matrix containing all candidate airfoil geometric parameter combinations and their corresponding aerodynamic performance indicators, including lift coefficient, drag coefficient, and lift-to-drag ratio, is output.

5. The simulation-based aerodynamic shape design method for aircraft according to claim 1, characterized in that: The fitness function is expressed as follows: In the formula, PPF represents the fitness value of the j-th individual in the population after the i-th iteration optimization, i.e., the aerodynamic performance evaluation index, i is the index of the iteration number, j is the index of the individual in the population, CL(i,j), CD(i,j), and GTR(i,j) are the lift coefficient, drag coefficient, and lift-to-drag ratio of the j-th individual in the population after the i-th iteration optimization, respectively. max (i) and CL min (i) represent the maximum and minimum lift coefficients from the 1st iteration to the i-th iteration, respectively; CD max (i) and CD min (i) represent the maximum and minimum resistance coefficients from the 1st iteration to the i-th iteration, respectively; GTR max (i) and GTR min (i) represent the maximum and minimum boost-to-drag ratios from the 1st iteration to the i-th iteration, respectively; CL ref CD ref and GTR ref The reference values ​​are, in order: lift coefficient reference value, drag coefficient reference value, and lift-to-drag ratio reference value; α i β i and γ i α is a pre-set scaling factor for the i-th iteration process. i >β i >γ i >0, and satisfy α i +β i +γ i =1.

6. The simulation-based aerodynamic shape design method for aircraft according to claim 1, characterized in that: The genetic algorithm selects the candidate airfoil geometry parameters with the highest fitness function value as the optimized airfoil parameter combination. The logic is as follows: Initialize a population of candidate airfoil geometry parameter combinations with a size of M. Each individual represents a set of airfoil geometry parameter codes. Use a roulette wheel selection strategy to select individuals based on their fitness function value; individuals with higher fitness function values ​​have a greater probability of being selected. Perform a simulated binary crossover operation on the selected individuals, with a crossover probability of Pc, to generate a new generation of parameter combinations. Perform a Gaussian mutation operation on the crossover individuals, with a mutation probability of Pm. Use an elite retention strategy to directly pass the best individual of the current generation to the next generation. Repeat the selection, crossover, and mutation operations until the preset number of iterations N is reached. Finally, output the candidate airfoil geometry parameter combination with the highest fitness function value as the optimization result, which is the optimized airfoil parameter combination.

7. The simulation-based aerodynamic shape design method for aircraft according to claim 6, characterized in that: The lift coefficient, drag coefficient, and lift-to-drag ratio are dynamically weighted and optimized based on the fitness function. The specific logic behind this optimization is as follows: After each iteration, the distribution of lift coefficient, drag coefficient, and lift-to-drag ratio of all individuals in the current population is statistically analyzed, and the following indicators are calculated; In the formula, CL opt,i CD opt,i and GTR opt,i Let CL represent the optimization degree indices of lift coefficient, drag coefficient, and lift-to-drag ratio in the i-th generation population, respectively. avg,i CD opt,i and GTR opt,i Let represent the average lift coefficient, average drag coefficient, and average lift-to-drag ratio of the i-th generation population, respectively, where i is the index of the iteration number, and CL ref CD ref and GTR ref The reference values ​​are, in order: lift coefficient reference value, drag coefficient reference value, and lift-to-drag ratio reference value. The optimization degree indicators of lift coefficient, drag coefficient, and lift-to-drag ratio are compared with their corresponding preset optimization thresholds. If an optimization degree indicator is greater than or equal to its preset optimization threshold, it means that the parameter has been sufficiently optimized, and its weight is reduced to redirect computing resources to other parameters to be optimized. Conversely, if the optimization degree indicator is less than its preset optimization threshold, the corresponding weight is increased to strengthen the optimization. Among them, the proportional relationship of the weights after constraint adjustment always satisfies α. i >β i >γ i >0, and satisfy α i +β i +γ i =1; α i β i and γ i This represents the weight after the i-th iteration; The specific logic behind dynamically adjusting the weights is as follows: The current weights are defined as: α i =α i-1 ,β i =β i-1 γ i =γ ii1 , where i≥1; If CL opt,i ≥X CL , then α i =α i-1 *0.9; If CD opt,i ≥ X CD , then β i = β i-1 * 0.9; If GTR opt,i ≥X GTR Then γ i =γ i-1 *0.9; In the formula, X CL X CD and X GTR These are the preset optimization thresholds corresponding to the optimization degree indicators of the lift coefficient, drag coefficient, and lift-to-drag ratio. Conversely, if certain optimization indicators fall below their thresholds, the corresponding weights will increase by 10%. Continue normalizing the weights; if the adjusted α i ≤β i Then exchange α i and β i Similarly, ensure α i >β i >γ i .

8. The simulation-based aerodynamic shape design method for aircraft according to claim 7, characterized in that: The specific logic behind the CFD simulation verification of the optimized airfoil parameter combination is as follows: A three-dimensional mesh model of the aircraft is generated based on optimized airfoil parameter combinations. Steady flow simulation is performed using the SST k-ω turbulence model, with inflow conditions including Mach number Ma and Reynolds number Re. The Reynolds-averaged Navier-Stokes equations are solved to obtain the flow field solution. The pressure coefficient distribution and shear stress distribution on the airfoil surface are calculated by integration from the converged flow field solution. The lift coefficient, drag coefficient, and lift-to-drag ratio are obtained through numerical integration. The calculated lift coefficient, drag coefficient, and lift-to-drag ratio are output as the simulation aerodynamic characteristic parameters. Mesh independence verification requires ensuring that the change of each airfoil geometric parameter is less than 1% after mesh refinement.

9. The simulation-based aerodynamic shape design method for aircraft according to claim 8, characterized in that: Calculate the relative errors between the lift coefficient, drag coefficient, and lift-to-drag ratio in the simulated aerodynamic characteristic parameters and the corresponding predicted values ​​output by the aerodynamic performance prediction model. If the relative error of the lift coefficient does not exceed 5%, the relative error of the drag coefficient does not exceed 8%, and the relative error of the lift-to-drag ratio does not exceed 10%, the optimized airfoil parameter combination is deemed to have passed the verification. If any index error exceeds the threshold, return to step 2 to adjust the parameters and supplement the training of the neural network model. The optimized airfoil parameter combination that has passed verification will be used as the final design scheme to generate an airfoil CAD model that meets engineering requirements, and output a complete design report containing airfoil geometric parameters and aerodynamic performance parameters to guide the manufacturing and assembly of the aircraft's aerodynamic shape; the aerodynamic performance indicators include lift coefficient, drag coefficient and lift-to-drag ratio.

10. A simulation-based aerodynamic shape design device for aircraft, characterized in that: The simulation-based aerodynamic shape design device for aircraft is used to implement the simulation-based aerodynamic shape design method for aircraft as described in any one of claims 1-9, comprising: The data acquisition and database construction module is used to acquire several sets of aerodynamic characteristic parameters of existing different airfoil aircraft. The aerodynamic characteristic parameters include airfoil geometric parameters and corresponding lift coefficient, drag coefficient and lift-drag ratio. The airfoil geometric parameters are specifically the airfoil thickness distribution curve function, three-dimensional twist angle distribution function and sweep angle change rate function. Each set of airfoil geometric parameters is mapped to lift coefficient, drag coefficient and lift-drag ratio to construct a multi-source aerodynamic database. The model building module establishes a neural network model based on a multi-source aerodynamic database. The airfoil geometric parameters in the multi-source aerodynamic database are used as the input of the model, and the lift coefficient, drag coefficient and lift-to-drag ratio in the multi-source aerodynamic database are used as labels to train the neural network model and obtain an aerodynamic performance prediction model. The airfoil parameter optimization module is used to sample and generate candidate airfoil geometric parameter combinations within the airfoil geometry parameter design space. It obtains the lift coefficient, drag coefficient, and lift-to-drag ratio for each combination through an aerodynamic performance prediction model. The module then calculates the aerodynamic performance evaluation index for each combination as its fitness value. This fitness value is used to evaluate the airfoil's aerodynamic performance under that combination. Based on a genetic algorithm, the module iteratively optimizes the candidate airfoil geometric parameter combinations to select the optimal airfoil parameter combinations. Furthermore, in each iteration, the module dynamically adjusts the weights of the lift coefficient, drag coefficient, and lift-to-drag ratio in the fitness value calculation based on the distribution of these parameters among individuals in the population.

Images