High-Low Speed Aerodynamic Inverse Design Method for Blended Wing-Body Configuration Based on Conditional Diffusion Model

Abstract

The present invention proposes a high-low speed aerodynamic inverse design method for blended wing-body configurations based on conditional diffusion models. By adopting a three-dimensional airfoil section design method for blended wing-body configurations, five sections of the blended wing-body configuration are selected for optimized inverse design, and physical condition parameters such as aerodynamic force and moment coefficients are embedded in the conditional denoising diffusion probability model, so that the prediction results can more accurately reflect the physical mechanism. The three-dimensional configuration designed by the present invention can effectively improve the aerodynamic performance, and the inverse-designed airfoil meets the predetermined aerodynamic performance requirements. Compared with the prior art, the present invention greatly reduces the training time from the original several weeks to 7.5 hours under the premise of achieving the same optimization performance, improving the design efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of aircraft design, and specifically to a high-low speed aerodynamic inverse design method for a blended wing-body layout based on a conditional diffusion model. Background Art

[0002] The blended wing-body layout is one of the most potential aircraft aerodynamic layouts in the future due to its advantages in aerodynamics, structure, etc. For the aerodynamic design research of the blended wing-body layout, it is currently mainly focused on the high-speed cruise state. However, due to the lack of a tail wing and insufficient rudder effectiveness in the low-speed state, the low-speed aerodynamic characteristics of the blended wing-body layout deteriorate sharply. The deficiencies in low-speed aerodynamic performance and low-speed stall characteristics seriously affect the low-speed field characteristics and flight safety of the blended wing-body layout. Therefore, it is necessary to carry out aerodynamic design research on its low-speed state as well. Traditional three-dimensional aircraft aerodynamic inverse design methods usually use optimization design methods based on surrogate models. However, when facing high-dimensional design variables, surrogate models are extremely prone to the "curse of dimensionality" problem. Another method is the inverse design method. By solving the classical aerodynamic inverse problem, the inverse design is usually time-saving and efficient compared with the direct optimization design method. However, the aerodynamic inverse design under high-dimensional design variables is still a key problem that is difficult to solve at present.

[0003] In recent years, the breakthrough development of deep generative models has provided a new paradigm for overcoming this traditional technical bottleneck of three-dimensional aircraft aerodynamic inverse design. In the Chinese patent application "A Method and System for Inverse Design of Aircraft Layout Based on Conditional Diffusion Model" with the publication number CN119720384A, a method based on a conditional diffusion model is disclosed, which can quickly generate a three-dimensional point cloud model of an aircraft that meets specific aerodynamic performance and stealth performance according to the input design requirements. However, the applicant analyzed this method and found that the single data sample of this method is a point cloud shape with 4,000 points. Although this method can effectively constrain the shape of the three-dimensional aircraft, it greatly increases the training cost of the model, consumes a large amount of time and resources, and is difficult to be applied in research institutions such as universities with limited computing resources. Summary of the Invention

[0004] To solve the above problems existing in the prior art, the present invention proposes a high-low speed aerodynamic inverse design method for blended wing-body configurations based on conditional diffusion models. By adopting a three-dimensional airfoil section design method for blended wing-body configurations, five sections of the blended wing-body configuration are selected for optimized inverse design, and physical condition parameters such as aerodynamic force and moment coefficients are embedded in the conditional denoising diffusion probability model, so that the prediction results can more accurately reflect the physical mechanism. The three-dimensional configuration designed by the present invention can effectively improve the aerodynamic performance, and the inverse-designed airfoil meets the predetermined aerodynamic performance requirements. Compared with the prior art, the present invention greatly reduces the training time from several weeks to 7.5 hours under the premise of achieving the same optimization performance, improving the design efficiency.

[0005] The present invention is realized through the following technical solutions:

[0006] A high-low speed aerodynamic inverse design method for blended wing-body configurations based on conditional diffusion models, comprising the following steps:

[0007] Step S1: Select the external shape of the reference blended wing-body configuration, and select multiple section reference airfoils for the external shape of the reference blended wing-body configuration; parameterize each section reference airfoil, determine the design variables, and sample in the design space of the design variables to obtain a sample section airfoil data set;

[0008] Step S2: Determine the design state and design objectives according to the flight characteristics of the blended wing-body configuration during high-low speed cruise; according to the design state and design objectives, perform CFD calculations on the sample section airfoil data set obtained in Step S1, and form a complete training sample set from the sample section airfoil data set combined with the aerodynamic force coefficients and aerodynamic moment coefficients obtained by CFD calculations;

[0009] Step S3: Construct a conditional denoising diffusion probability model for aerodynamic inverse design of each section airfoil of the blended wing-body configuration; use the training sample set obtained in Step S2 to train the conditional denoising diffusion probability model to obtain an aerodynamic inverse design model for the blended wing-body configuration;

[0010] The conditional denoising diffusion probability model is a network architecture model including a self-attention mechanism, a residual module, and a Unet network. Among them, conditional parameters are embedded in each layer of the Unet network. The conditional parameters are aerodynamic force coefficients and aerodynamic moment coefficients. The embedding method is: converting the aerodynamic force coefficients and aerodynamic moment coefficients into corresponding tensors through convolution, and splicing them into the input data of the Unet network and the output data of each layer of the Unet network, and then reconstructing the output data of each layer according to the input dimension of the next layer by modifying the number of channels and the amount of data in each channel;

[0011] Step S4: Based on each sectional reference airfoil, according to the design conditions and design objectives determined in Step S2, use the aerodynamic inverse design model of the wing-body fusion layout obtained in Step S3 to perform inverse design on the airfoils of each section of the wing-body fusion layout in the design space of the design variables, so as to obtain the shape of the wing-body fusion layout that meets the requirements of the design objectives.

[0012] Further, in Step S1, according to the spanwise layout stations of the wing-body fusion layout, five sectional reference airfoils are selected in the half model of the wing-body fusion layout, including two sectional reference airfoils in the outer wing region, one sectional reference airfoil in the central body region, and one sectional reference airfoil each at the symmetry plane and the transition station.

[0013] Further, in Step S1, the perturbation CST parameterization method is used to parameterize each sectional reference airfoil, and the coefficients of the shape functions are used as design variables.

[0014] Further, in Step S1, for a single sectional reference airfoil, a total of 10 design variables are used, with 5 design variables on each of the upper and lower surfaces of the sectional reference airfoil. Among them, the coefficient is used to describe the upper surface shape, and the coefficient is used to describe the lower surface shape; the coefficient affects the upper surface shape along the leading edge to trailing edge direction of the airfoil, and the coefficient affects the lower surface shape in the leading edge to trailing edge direction of the airfoil; The design space of is [-0.01, 0.01], and the design space of the remaining coefficients is [-0.02, 0.02].

[0015] Further, in Step S2, the high-speed design conditions are: Ma = 0.8, AOA = 2°, Re = 2.9×10 7 , and the low-speed design conditions are Ma = 0.2, AOA = 8°, Re = 1.0×10 7 ; where Ma is the Mach number, AOA is the angle of attack, and Re is the Reynolds number; the design objective is the aerodynamic performance of the wing-body fusion layout, including three aerodynamic parameters: lift coefficient, drag coefficient, and pitching moment coefficient.

[0016] Further, in Step S4, for each sectional airfoil obtained by inverse design, it is required that compared with the reference airfoil, the lift coefficient increases, the drag coefficient remains unchanged, and the absolute value of the pitching moment coefficient decreases in the high-speed design condition; the lift coefficient remains unchanged, the drag coefficient decreases, and the absolute value of the pitching moment coefficient decreases in the low-speed design condition.

[0017] Further, in Step S4, compared with the reference airfoil, the change range of the design objective is determined according to the maximum value of the deviation of the aerodynamic performance data of each sample sectional airfoil in the sample sectional airfoil dataset from the aerodynamic performance data of the reference airfoil.

[0018] Further, in step S4, after the airfoil inverse design of five cross-sections is completed in the design space using the wing-body fusion layout aerodynamic inverse design model obtained in step S3, the airfoils obtained by inverse design are used to replace the corresponding reference airfoils of the five cross-sections, and a three-dimensional layout reconstruction is performed to obtain a wing-body fusion layout shape that meets the design target requirements.

[0019] Further, the present invention also provides a computer device, including: a processor, a storage medium, and a bus. The storage medium stores program instructions executable by the processor. When the computer device runs, the processor communicates with the storage medium through the bus. When the processor executes the program instructions, it is used to implement the above method.

[0020] Further, the present invention also provides a computer-readable storage medium. A computer program is stored on the storage medium, and when the computer program is run by a computer, it is used to implement the above method.

[0021] Beneficial effects:

[0022] The present invention provides a high-low speed aerodynamic inverse design method for wing-body fusion layout based on a conditional diffusion model, constructs a conditional probability denoising diffusion model, which is based on a Markov chain and generates data through two diffusion processes: forward noise addition and reverse denoising; it can directly input aerodynamic parameters to obtain the target airfoil of the wing-body fusion layout. At the same time, compared with the original three-dimensional layout, the three-dimensional layout after replacing with the inverse design airfoil has higher aerodynamic performance in both high-speed and low-speed states. And compared with the prior art, the dimension of design variables is greatly reduced, saving training time and cost. Therefore, this method can greatly improve the design efficiency while ensuring the design accuracy, and can efficiently and accurately complete the three-dimensional cross-section design of the wing-body fusion layout. Description of the drawings

[0023] Figure 1 Schematic diagram of the reference wing-body fusion layout and cross-section station selection for the embodiments of the present invention;

[0024] Figure 2 Schematic diagram of the CST parameter stations and design variables for the embodiments of the present invention;

[0025] Figure 3 Schematic diagram of the upper and lower limits of the airfoil perturbation range for the embodiments of the present invention;

[0026] Figure 4 Schematic diagram of the airfoil grid for the embodiments of the present invention;

[0027] Figure 5 Schematic diagram of the training of the conditional diffusion model for the embodiments of the present invention;

[0028] Figure 6Sampling schematic diagram of the conditional diffusion model in the embodiments of the present invention;

[0029] Figure 7 Schematic structural diagram of the one-dimensional U-net model in the embodiments of the present invention;

[0030] Figure 8 Schematic diagram of the error of the inverse design result provided in the embodiments of the present invention;

[0031] Figure 9 Schematic comparison diagram of the inverse design airfoil sections provided in the embodiments of the present invention;

[0032] Figure 10 Schematic comparison diagram of the predicted aerodynamic performance and the actual aerodynamic performance of the sectional airfoil provided in the embodiments of the present invention;

[0033] Figure 11 Schematic diagram of the airfoil design effect randomly selected from the training sample sets of five sections during the diffusion process in the embodiments of the present invention;

[0034] Figure 12 Schematic diagram of the data distribution of the lift coefficient and drag coefficient in the high-speed state and low-speed state in the airfoil dataset of the first sectional airfoil provided in the embodiments of the present invention;

[0035] Figure 13 Schematic diagram of the pressure coefficient distribution and airfoil comparison of 5 two-dimensional sectional airfoils in the high-speed state provided in the embodiments of the present invention;

[0036] Figure 14 Schematic comparison diagram of the pressure coefficient distribution of 5 two-dimensional sectional airfoils in the low-speed state provided in the embodiments of the present invention;

[0037] Figure 15 Schematic comparison diagram of the pressure coefficient contour maps before and after the design of the three-dimensional wing-body fusion layout in the high-speed state provided in the embodiments of the present invention;

[0038] Figure 16 Schematic comparison diagram of the pressure coefficient contour maps before and after the design of the three-dimensional wing-body fusion layout in the low-speed state provided in the embodiments of the present invention. Detailed implementation manners

[0039] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention clearer and more understandable, and to enable those skilled in the art to better understand the solution of the present invention, the following further describes and completely describes the present invention in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0040] In this embodiment, the three-dimensional profile design of a certain wing-body fusion layout is carried out by using the high and low speed aerodynamic inverse design method of wing-body fusion layout based on conditional diffusion model proposed by the present invention: the aerodynamic optimization inverse design is carried out for the profile airfoil of the wing-body fusion layout, and the optimized airfoil is assembled to the corresponding profile by using the secondary development of CATIA, and the three-dimensional layout model is reconstructed, and finally the CFD numerical calculation verification is carried out on the three-dimensional layout.

[0041] This embodiment specifically includes the following steps:

[0042] Step S1: Select the external shape of the reference wing-body fusion layout, and select multiple profile reference airfoils for the external shape of the reference wing-body fusion layout; parameterize each profile reference airfoil, determine the design variables, and sample in the design space of the design variables to obtain a sample profile airfoil data set;

[0043] In this embodiment, specifically, the wing-body fusion layout is selected as the reference external shape, and five profile reference airfoils (half models) are selected according to the spanwise layout stations of the wing-body fusion layout, as Figure 1 shown. The five profiles are two profile reference airfoils (Profile 1 and Profile 2) in the outer wing area, one profile reference airfoil (Profile 3) in the central body area, and one profile reference airfoil at each of the symmetry plane (Root) (Profile 5) and the transition (Kink) (Profile 4) stations. The parameterization method for each sample profile reference airfoil adopts the perturbed CST parameterization method, and the CST basis function form is: , where the class function is used to define the geometric shape, x is the abscissa value of the profile reference airfoil, N1 and N2 are parameters that control the characteristics of the class function, and the specific values in this embodiment are: N1 = 0.5, N2 = 1; the shape function is used to describe the geometric characteristics of the shape in detail, and the basis function of the shape function . On this basis, the general expression of the perturbed CST is: ; N is the order, and the specific value is: N = 4; the coefficient is the design variable. For a single sample profile airfoil in this embodiment, a total of 10 design variables are adopted, 5 for each of the upper and lower surfaces, where is used to describe the upper surface shape, is used to describe the lower surface shape. Since the basis function of the shape function S(x) has a main influence area, the influence areas of the 10 coefficients it contains are as Figure 2 shown. In order to avoid obvious error samples while ensuring the unity of the scale of each profile design variable and the diversity of inverse design, the design space of the design variables selected by the present invention is: the design space of the coefficients {A1, A4, A6, A9} is [-0.01, 0.01], and the design space of the remaining coefficients is [-0.02, 0.02]. Figure 3The upper and lower limits of the perturbation range are given. At the same time, in order to verify the effectiveness of the conditional diffusion model, a sufficient number of aerodynamic samples need to be obtained. 5000 airfoil samples are respectively extracted from five profile stations through Latin hypercube sampling to maximize the generalization ability of the model.

[0044] Step S2: Determine the design state and design objectives according to the flight characteristics of the wing-body fusion layout during high and low speed cruise; according to the design state and design objectives, perform CFD calculations on the sample profile airfoil dataset obtained in step S1, and combine the aerodynamic coefficient and aerodynamic moment coefficient obtained from the CFD calculation of the sample profile airfoil dataset to form a complete training sample set;

[0045] In this embodiment, for the conventional high and low speed cruise states of the wing-body fusion layout, the determined high speed design state is: Ma = 0.8, AOA = 2°, Re = 2.9×10 7 , and the low speed design state is Ma = 0.2, AOA = 8°, Re = 1.0×10 7 ; where Ma is the Mach number, AOA is the angle of attack, and Re is the Reynolds number; the design objective is the aerodynamic performance of the wing-body fusion layout, specifically including the following three aerodynamic parameters: lift coefficient, drag coefficient, and pitching moment coefficient.

[0046] When performing CFD calculations, the CFL3D solver is used for aerodynamic performance evaluation, the k-ω SST turbulence model is selected for the turbulence model, the airfoil grid quantity is 50,000, and the height of the first layer of the grid in the wall normal direction is determined according to the oncoming flow parameters. The calculation grid is as Figure 4 shown. Thus, a complete dataset including all profile airfoil shape parameters and corresponding aerodynamic coefficients is obtained, and this dataset is used for the subsequent network training steps.

[0047] Step S3: Construct a conditional denoising diffusion probability model for aerodynamic inverse design of each profile airfoil of the wing-body fusion layout; use the training sample set obtained in step S2 to train the conditional denoising diffusion probability model to obtain an aerodynamic inverse design model for the wing-body fusion layout;

[0048] In this embodiment, the conditional denoising diffusion probability model is a network architecture model including a self-attention mechanism, a residual module, and a Unet network. This model architecture is a well-known technology in the art; the specific construction process is as follows:

[0049] In this embodiment, the denoising diffusion probabilistic model (DDPM) is changed to a conditional denoising probabilistic diffusion model (CDDPM) based on classifier-free guidance. The specific implementation method is to use the conditional parameter set and the time step t as the embedding conditions for each step of the model. This method can ensure that the model can learn conditional information in the same way as learning time information during training. At the same time, what also needs to be changed are the probability distribution and loss function of each time step. All the remaining mathematical derivation processes are exactly the same as those of DDPM. The expression of the changed loss function is:

[0050]

[0051] In the formula: is the loss function, is the noise added by forward noise addition; represents the prediction of the true noise by the model at time step t; is the original data noise; c represents the conditional parameter set; is the predefined Gaussian noise variance.

[0052] As Figure 5 and Figure 6 shown, both the training process and sampling process of the diffusion model are based on the Markov chain. Traditional models generally directly use the coordinate point positions of the airfoil for training. However, CDDPM is not a dimensionality reduction model. If the airfoil coordinate point samples are directly input for training, the training time of the model will be significantly increased. Therefore, in this embodiment, the design variables are obtained by parameterizing the airfoil with CST, and the design variables are used as the input of the model for training, which can significantly reduce the input dimension and reduce the training time. When training this model, the airfoil data first undergoes CST parameterization to obtain the set of potential design variables. Then, they are input into the model together for forward noise addition to obtain Gaussian noise, and then the new shape is obtained by reverse denoising through the network of CDDPM. The training module of the entire model is the reverse denoising process. For each step of the denoising process, the noise of this step is trained. In this process, CDDPM is trained by minimizing the loss function, so that the predicted noise and the true noise The mean square error is the smallest. To ensure that the prediction results contain information about the conditional parameters, the conditional parameters are embedded in each layer of the Unet network. The conditional parameters are the lift coefficient, the drag coefficient, and the pitching moment coefficient. The embedding method is as follows: The values of the lift coefficient, the drag coefficient, and the pitching moment coefficient are converted into corresponding tensors through convolution and spliced into the input data of the Unet network and the output data of each layer in the Unet network. Then, according to the input dimension of the next layer, the output data of each layer is reconstructed by modifying the number of channels and the amount of data in each channel. This is implemented in each layer of the network to ensure that the network can learn the information of the conditional parameters. This is the key technical feature that distinguishes the conditional denoising diffusion probability model adopted by the present invention from the traditional conditional denoising diffusion probability model. Specifically, as Figure 7 shown, the input for training is and the aerodynamic parameter conditions [ Cl , Cd , Cm ] . The input size is 10×1, and the output size matches the shape design variable size, which is also 10×1. The entire network performs downsampling through a series of one-dimensional residual convolutional blocks to extract the potential high-dimensional features of the data. Compared with the 4000-dimensional input data of the prior art, this method has achieved a dimension reduction across orders of magnitude.

[0053] In the downsampling stage, the model introduces a residual network layer to transfer the input source information and enhance the stability of training. The residual connection is a connection method that directly spans the neural network layers, which allows information to be directly transferred between different convolutional layers, thereby retaining more feature information and reducing the problem of gradient vanishing. In addition, the model also introduces a self-attention mechanism to capture the correlation between different data, thereby obtaining the global characteristic information of each parameter. This mechanism can adaptively assign weights to different parameters, enabling the model to focus on more important conditional data information.

[0054] The stride of all one-dimensional convolutional layers is 1, and the stride is 2 during downsampling to reduce the resolution. In the upsampling stage, various features are fused through skip connections. The additional time step information is processed through positional encoding of sine and cosine functions and added to each layer together with the conditional information encoding to ensure that the model can read the forward progress information and the conditional information.

[0055] Except for the output layer, LeakyReLU is used as the non-linear activation function of the network layer in the remaining layers. The number of forward noise addition steps and the number of reverse denoising steps in the entire diffusion process are both 1000 steps. The Adam optimizer is used for backpropagation to update the gradient. The batch size during training is set to 256, the initial learning rate is set to 0.0001, and the number of training epochs is set to 3000.

[0056] During sampling generation, the new Gaussian noise samples sampled from the prior distribution and the aerodynamic parameter conditions are directly used as inputs and passed through reverse denoising to obtain new samples. Finally, the geometric shape of the new airfoil is obtained through CST inverse parameterization.

[0057] In this embodiment, the dataset samples are divided into a training set and a test set according to a ratio of 9:1. The training set is used for the training of the CDDPM network to optimize the weights of the model; the test set is used for the evaluation of the model's inverse design ability. The learning rate is determined to be 0.0001, the batch size is determined to be 256, and the number of forward propagation steps is determined to be 1000; in the test phase, the size of the random noise is kept at 10×1, and its value is randomly generated normally to meet the requirements of Gaussian noise. Then, the aerodynamic parameters of the test set are input into the CDDPM network, and the corresponding airfoil design variables are predicted from the reverse denoising process. Figure 8 Show the mean absolute error and root mean square error between the numerical values of the shape parameters of each predicted airfoil and the numerical values of the shape parameters of the real airfoil in all samples under five airfoil sections. It can be seen that the mean absolute error and root mean square error of the five airfoil section test sets are both below 0.03, which preliminarily indicates that the trained CDDPM model can inverse design the airfoil under the given aerodynamic parameter conditions. To more intuitively show the gap between the predicted airfoil inverse designed by CDDPM and the real airfoil, Figure 9 The prediction results of an airfoil randomly selected from the five-section test set samples are given, and the predicted values of the airfoil are in good agreement with the actual expected values. Figure 10 Show the comparison between the predicted values and the real values of the lift coefficients of 5 sections in the test set. In the figure, the horizontal axis is the real value, the vertical axis is the predicted value, and the legend marks the correlation coefficients of the two in turn. The lift coefficients of the airfoils obtained through inverse design at each section station are closely distributed near the standard line. To more carefully observe the changes of the airfoil during the model training process, Figure 11 Show the airfoil design effects randomly selected from the training sample sets of five sections during the diffusion process. It can be seen from the figure that in the initial stage of reverse diffusion, the sample generation has extremely high randomness, so the situation of the upper and lower airfoil surfaces crossing appears. As the denoising process progresses, the shape of the predicted samples gradually approaches the real samples. Therefore, when the denoising reaches a certain level, the shape of the predicted samples will show an obvious change trend. When the denoising is completed, the difference between the predicted airfoil and the real airfoil reaches the minimum.

[0058] Step S4: Based on the reference airfoils of each section, according to the design state and design goal determined in step S2, use the airfoil-body integrated layout aerodynamic inverse design model obtained in step S3 to inverse design the airfoils of each section of the airfoil-body integrated layout in the design space of the design variables, and obtain the airfoil-body integrated layout shape that meets the design goal requirements.

[0059] Since the three-dimensional airfoil section design method is adopted in the present invention, it is necessary to replace the reference airfoils of the five sections with airfoils having better performance, so as to improve the aerodynamic performance of the three-dimensional wing-body blending layout. The design objectives of the airfoils of the five sections obtained by inverse design are shown in Table 1: for the high-speed state, it is required that the lift coefficient increases, the drag coefficient remains unchanged, and the absolute value of the pitching moment coefficient decreases; when performing inverse design for the low-speed state, it is required that the lift coefficient remains unchanged, the drag coefficient decreases, and the absolute value of the pitching moment coefficient decreases. When performing inverse design, reasonably determining the design objectives, that is, reasonably determining the change range of the design objectives compared with the reference airfoil, is also one of the key factors for the final success of the inverse design method. In the present invention, by analyzing the distribution of the aerodynamic performance data of each sample airfoil section in the sample airfoil section dataset obtained in step S2, it is determined that the maximum deviation of the aerodynamic performance data of each sample airfoil section in the sample airfoil section dataset from the aerodynamic performance data of the reference airfoil is 15%. In order to ensure the inverse design effect of the model, the present invention further reduces it by 5% on the basis of the maximum deviation, and finally takes the change interval of the design objectives as 10%, thereby obtaining Table 1. Figure 12 The distributions of the lift coefficient and the drag coefficient of the first airfoil section in the high-speed state and the low-speed state are given. The vertical coordinate represents the relative error of the aerodynamic performance of the first airfoil section with respect to the aerodynamic performance of the reference airfoil.

[0060] Table 1 Design objectives of inverse design

[0061]

[0062] According to the design states determined in step S2, the high- and low-speed target aerodynamic parameters of each airfoil section are shown in Table 2. After the design is completed, Figure 13 and Figure 14 are the comparison diagrams of the high- and low-speed pressure coefficient distributions and the designed airfoil results of the five sections. Figure 13 It shows that the airfoil sections after inverse design according to the design states change significantly compared with the reference airfoil. Each airfoil section generally shows a decrease in relative thickness and an increase in camber, improving the lift coefficient in the high-speed state and keeping the lift coefficient in the low-speed state basically unchanged.

[0063] Table 2 Target aerodynamic parameters / (aerodynamic parameters of the reference airfoil)

[0064]

[0065] Table 3 shows the specific aerodynamic parameters of the airfoils in each section after inverse design. According to the design state requirements in the previous text, compared with the baseline airfoil, the lift coefficient of the inverse design airfoil has increased by approximately 7.78% on average in the high-speed state; the drag coefficient has decreased by approximately 9.23% on average in the low-speed state. The change in aerodynamic parameters meets the 10% requirement of the design space, indicating the potential of this method to improve performance in inverse design and achieving the preset target requirements.

[0066] Table 3 Aerodynamic Parameters of Inverse Design Airfoils

[0067]

[0068] After the two-dimensional design is completed, the baseline airfoils corresponding to the five sections are replaced with the inverse design airfoils for three-dimensional layout reconstruction. Using the secondary development method of CATIA, the obtained inverse design airfoils are imported into CATIA to generate the corresponding wings. When reconstructing the three-dimensional layout, to ensure a smooth transition, splines are used to generate the leading and trailing edges of the layout. The two ends of the spline are tangent to the Y-axis, and the tension is set to 0.05; then, an xy plane perpendicular to the z-axis is constructed at the root, and the multi-section surface command is used to generate the upper and lower surfaces of the central main body area. The leading and trailing edges of the central main body area use splines as guide lines, and the xy plane is constructed as the support surface. Finally, using the leading and trailing edge splines of the outer wing area as guide lines and the upper and lower surface contours of the central body as the support surface, the same multi-surface command is used to generate the upper and lower surface contours of the outer wing area, thus achieving a smooth transition and ensuring the smoothness of the three-dimensional layout.

[0069] Figure 15 and Figure 16 This is a comparison of the three-dimensional aerodynamic layout calculation results. It can be seen from the figure that in the low-speed state, the area of the negative pressure zone at the leading edge of the wing of the three-dimensional layout replaced with the inverse design airfoil is smaller than that of the original three-dimensional layout, which results in a decrease in the lift coefficient by 0.0026. However, the drag coefficient of the overall wing-body fusion layout also decreases by 18 counts, and finally the lift-to-drag ratio increases by 0.364, achieving the purpose of improving the performance of the wing-body fusion layout after inverse design. Similarly, for the high-speed state, the area of the negative pressure zone at the leading edge of the upper surface and the front part of the central body area of the three-dimensional layout replaced with the inverse design airfoil is significantly larger than that of the original three-dimensional layout, increasing the lift coefficient of the entire three-dimensional layout by 0.0035 and the lift-to-drag ratio by 0.58.

[0070] In this embodiment, the total training time for 5 profiles is 7.5 h, and the training is carried out on an NVIDIA GeForce RTX2080Ti graphics card; the sampling time for 5 airfoil samples is 4.61 s, and the time for obtaining a single three-dimensional layout through secondary development of CATIA is 22 s. Therefore, the total time consumed from the entire training to the sampling process is less than 8 h, which is greatly reduced compared with the prior art. Moreover, the inverse design effect is good, and the lift-drag ratios of the generated three-dimensional layouts are improved under both high- and low-speed conditions.

[0071] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Without departing from the principle and spirit of the present invention, those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A high-low speed aerodynamic inverse design method for a blended wing-body configuration based on a conditional diffusion model, characterized in that: It includes the following steps: Step S1: Select the baseline wing-body blended layout shape, and select multiple sectional baseline airfoils for the baseline wing-body blended layout shape; parameterize each sectional baseline airfoil, determine the design variables, and sample in the design space of the design variables to obtain a sample sectional airfoil dataset; Step S2: Determine the design conditions and design objectives according to the flight characteristics of the wing-body blended layout during high and low speed cruise; according to the design conditions and design objectives, perform CFD calculations on the sample sectional airfoil dataset obtained in Step S1, and form a complete training sample set from the aerodynamic coefficient and aerodynamic moment coefficient obtained by combining the sample sectional airfoil dataset with CFD calculations; Step S3: Construct a conditional denoising diffusion probability model for aerodynamic inverse design of each sectional airfoil of the wing-body blended layout; use the training sample set obtained in Step S2 to train the conditional denoising diffusion probability model to obtain an aerodynamic inverse design model for the wing-body blended layout; The conditional denoising diffusion probability model is a network architecture model including a self-attention mechanism, a residual module, and a Unet network. Among them, conditional parameters are embedded in each layer of the Unet network. The conditional parameters are the aerodynamic coefficient and aerodynamic moment coefficient. The embedding method is: convert the aerodynamic coefficient and aerodynamic moment coefficient into corresponding tensors through convolution, and splice them into the input data of the Unet network and the output data of each layer of the Unet network, and then reconstruct the output data of each layer according to the input dimension of the next layer by modifying the number of channels and the amount of data in each channel; Step S4: Based on each sectional baseline airfoil, according to the design conditions and design objectives determined in Step S2, use the aerodynamic inverse design model of the wing-body blended layout obtained in Step S3 to perform inverse design on the airfoils of each section of the wing-body blended layout in the design space of the design variables to obtain a wing-body blended layout shape that meets the design objective requirements.

2. The high- and low-speed aerodynamic inverse design method for a blended wing-body configuration based on a conditional diffusion model according to claim 1, wherein: In Step S1, according to the spanwise layout stations of the wing-body blended layout, select five sectional baseline airfoils in the half model of the wing-body blended layout, including two sectional baseline airfoils in the outer wing area, one sectional baseline airfoil in the central body area, and one sectional baseline airfoil at the symmetry plane and the transition station respectively.

3. The aerodynamic inverse design method for high and low speed of blended wing body layout based on conditional diffusion model according to claim 1, characterized in that: In Step S1, use the perturbed CST parameterization method to parameterize each sectional baseline airfoil, and use the coefficients of the shape functions as the design variables.

4. The high and low speed aerodynamic inverse design method for wing-body fusion layout based on conditional diffusion model according to claim 3, characterized in that: In step S1, for a single sectional reference airfoil, a total of 10 design variables are adopted, with 5 design variables on each of the upper and lower surfaces of the sectional reference airfoil. Among them, the coefficient is used to describe the upper surface profile, and the coefficient is used to describe the lower surface profile; the coefficient affects the upper surface profile along the leading edge to trailing edge direction of the airfoil, and the coefficient affects the lower surface profile in the leading edge to trailing edge direction of the airfoil; has a design space of [-0.01, 0.01], and the remaining coefficients have a design space of [-0.02, 0.02].

5. The high- and low-speed aerodynamic inverse design method for a blended wing-body configuration based on a conditional diffusion model according to claim 2, wherein: In step S2, the high-speed design state is: Ma = 0.8, AOA = 2°, Re = 2.9×10 7 , and the low-speed design state is Ma = 0.2, AOA = 8°, Re = 1.0×10 7 ; where Ma is the Mach number, AOA is the angle of attack, and Re is the Reynolds number; the design objective is the aerodynamic performance of the wing-body fusion layout, including three aerodynamic parameters: lift coefficient, drag coefficient, and pitching moment coefficient.

6. The high - low speed aerodynamic inverse design method for blended wing - body configuration based on conditional diffusion model according to claim 5, characterized in that: In Step S4, for each sectional airfoil obtained by inverse design, it is required that compared with the baseline airfoil, the lift coefficient increases in the high-speed design condition, the drag coefficient remains unchanged, and the absolute value of the pitching moment coefficient decreases; the lift coefficient remains unchanged, the drag coefficient decreases, and the absolute value of the pitching moment coefficient decreases in the low-speed design condition.

7. The high- and low-speed aerodynamic inverse design method for a blended wing-body configuration based on the conditional diffusion model according to claim 6, wherein: In Step S4, compared with the baseline airfoil, the change range of the design objective is determined according to the maximum value of the deviation of the aerodynamic performance data of each sample sectional airfoil in the sample sectional airfoil dataset from the aerodynamic performance data of the baseline airfoil.

8. The high- and low-speed aerodynamic inverse design method for a blended wing-body configuration based on a conditional diffusion model according to claim 1, wherein: In step S4, after the airfoil inverse design of five profiles is completed in the design space using the wing-body blended layout aerodynamic inverse design model obtained in step S3, the airfoils obtained by the inverse design are used to replace the corresponding baseline airfoils of the five profiles, and a three-dimensional layout reconstruction is performed to obtain the wing-body blended layout shape that meets the design target requirements.

9. A computer device, comprising: A processor, a storage medium, and a bus, where the storage medium stores program instructions executable by the processor. When the computer device runs, the processor communicates with the storage medium through the bus. It is characterized in that: when the processor executes the program instructions, it is used to implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, on which a computer program is stored, characterized in that: When the computer program runs on a computer, it is used to implement the method according to any one of claims 1 to 8.

Images